Doc. no. | D2139R3 |
Date: | 2020-07-27 |
Project: | Programming Language C++ |
Audience: | Evolution Working Group Incubator |
Library Evolution Working Group Incubator | |
Reply to: | Alisdair Meredith <[email protected]> |
*this
by reference [depr.capture.this]char*
streams [depr.str.strstreams]move_iterator
access [depr.move.iter.elem]shared_ptr
atomic access [depr.util.smartptr.shared.atomic]basic_string
capacity [depr.string.capacity]Revision 3
Revision of the paper for the 2020 August mailing, following SG16 feedback.
- Updated integrated wording following SG16 feedback.
- In D.10, flagged wording recommended for removal with a distinct background to simplify eventual review.
- Noted SG16 feedback on clause D.20-D.23.
Revision 2
Revision of the paper for the 2020 July mailing, following telecon feedback.
- Fixed whitespace bugs in the grammar for
relational-expression
- Added comment to introduction on general feedback on the topic of removing deprecated libraries
- Record LEWGI feedback from the telecon on 2020-07-13
- Add initial LEWGI feedback to integrated wording for eventual LWG review
- Note in D.9 that the next C standard will likely deprecate
<stdalign.h>
and<stdbool.h>
- Add a note on the historical use of
rel_ops
in the standard, prior to its moving to D.11- Provided (without encouragement) wording to undeprecate strstreams in D.12
- Added cross-reference to the
spanstream
proposal for D.12- For D.13, note that the related term POF was removed from C++17
- Fix original reason for deprecation in D.14/D.15, which had nothing to do with modules!
- For D.16, add cross-reference to the paper that introduced more
iterator_traits
defaults- Note for D.24 that the next C standard will likely remove
ATOMIC_VAR_INIT
Revision 1
Revision of the paper for the 2020 June mailing, following early feedback.
- Clean up document format using a html validator
- Note which compiler versions start issuing warnings
- Record EWGI review of subclauses D.1 - D.4.
- Start collecting integrated wording for eventual core review, where EWGI have provided a recommendation.
- Note that users can work around deprecation in D.1 with a simple unary
+
- Update wording for EWGI preference to remove just floating point interaction in D.1
- Propose Annex C wording for D.1
- Completely revised recommendations for D.3 (comma expressions in array bounds), guided by feedback on R0 of this paper.
- Propose Annex C wording for D.3 following EWGI recommendation to remove.
- Propose Annex C wording for D.4 following EWGI recommendation to remove.
- A more fine-grained analysis of D.5 (volatile operations), as this deprecation is really four cases in one clause.
- Initial drafting of core wording to remove D.5, if we so choose.
- Note that no currently available compiler warns on the deprecation of D.6 (Redeclare
constexpr
members)- Example of an iterator implementation idiom that requires slightly more work to resolve D.8 deprecation.
- Note that Clang 10 also now warns on the deprecations in D.8.
- Added a note to D.9 that we expect compilers will never give a deprecation warning on the use of these headers
- For D.10 (library Requires paragraphs), revised suggestion of Mandates to Preconditions where both run-time and compile-time requirements applied.
Original
Original version of the paper for the 2020 post-Prague mailing.
This paper evaluates all the existing deprecated facilities in the C++20 standard, and recommends removing a subset from Annex D in C++23, either by removing from the standard entirely, or by undeprecating and restoring to the main text.
With the release of a new C++ standard, we get an opportunity to revisit the features identified for deprecation, and consider if we are prepared to clear any out yet, either by removing completely from the standard, or by reversing the deprecation decision and restoring the feature to full service. This paper makes no attempt to offer proposals for removing features other than those deprecated in Annex D, nor does it attempt to identify new candidates for deprecation.
In an ideal world, the start of every release cycle would cleanse the list of deprecated features entirely, allowing the language and library to evolve cleanly without holding too much deadweight. This paper is an attempt to consolidate all likely requests (and papers) to review isolated features on Annex D, and gain efficiency by performing one consolidated review. In practice, C++ has some long-term deprecated facilities that are difficult to remove, and equally difficult to rehabilitate. It seems reasonable that work could be split into follow-up papers if a direction has appeal, but needs more work for consensus to advance; one possible goal of this paper is to discover which features we may be interested in following up while maintaining the tempo of an efficient high level review.
The benefits of making the choice to remove features early in a standard cycle is that we will get the most experience we can from the bleeding-edge adopters whether a particular removal is more problematic than expected - but even this data point is limited, as bleeding-edge adopters typically have less reliance on deprecated features, eagerly adopting the newer replacement facilities.
However, do note that with the three year release cadence for the C++ standard, we will often be re-evaluating features whose deprecated status has barely reached print.
We have precedent that Core language features are good targets for removal, typically taking two standard cycles to remove a deprecated feature, although often prepared to entirely remove a feature even without a period of deprecation, if the cause is strong enough.
The library experience has been mixed, with no desire to remove anything
without a period of deprecation (other than gets
) and no precedent
prior to C++17 for actually removing deprecated features.
The precedent set for the library in C++17 seems to be keener to clear out the old code more quickly though, even removing features deprecated as recently as the previous standard, with the ink still drying on the text! Accordingly, this paper will be fairly aggressive in its attempt to clear out old libraries. This is perhaps more reasonable for the library clauses than the core language, as the Zombie Names clause, 16.5.4.3.1 [zombie.names] allows vendors to continue shipping features long after they have left the standard, as long as their existing customers rely on them.
Informal feedback on library deprecation prior to LEWGI review leans in two directions: deprecation is for life, and we should never remove anything, as there is no reason for the library to ever break client code; and alternatively the with the addition of zombie names there is clear intent library vendors should continue supporting deprecated features as long as their customers demand them, without violating conformance, and that removal from the standard removes maintenance costs from wg21 (placed instead on library vendors) and that this paper should have been stronger in its recommendations to remove.
We will review each deprecated facility, and make a strong and a weak recommendation. The strong recommendation is the preferred direction of the authors, who lean towards early removal. There will also be a weak recommendation, which is an alternative proposal for the evolution groups to consider, if the strong recommendation does not find favor. Finally, wording is generally provided for both the strong and weak recommendations, which will be collated into a unified Proposed Wording section once the various recommendations have been given direction by the corresponding evolution group.
All proposed wording is relative to N4861, which best corresponds to the C++20 DIS.
The intent is that the incubator groups review these proposals by telecon throughout 2020, indicating which features should be given more attention to gain implementation experience in time to inform the evolution groups when face-to-face meetings resume in (hopefully) 2021. Recommendations will be tracked in the table below.
Checklist for recommendations:
Subclause | Introduced | Deprecated | Paper | Feature | Strong Recommendation | Weak Recommendation | Incubator Recommendation | Final Decision |
---|---|---|---|---|---|---|---|---|
D.1 | C++98 | C++20 | P1120R0 | Arithmetic conversion on enumerations | No action | Remove now | Partial removal | pending... |
D.2 | C++11 | C++20 | P0806R2 | Implicit capture of *this by reference |
Remove now | No action | No action | pending... |
D.3 | C++98 | C++20 | P1161R3 | Comma operator in subscript expressions | P2128R1 | Remove now | Remove now | pending... |
D.4 | C++98 | C++20 | P1120R0 | Array comparisons | Remove now | No action | Remove now | pending... |
D.5 | C++98 | C++20 | P1152R4 | Deprecated use of volatile |
No action | Remove now | pending... | pending... |
D.6 | C++11 | C++17 | P0386R2 | Reclare constexpr members |
Remove now | No action | pending... | pending... |
D.7 | C++98 | C++20 | P1815R2 | Non-local use of TU-local entities | No action | No action yet | pending... | pending... |
D.8 | C++98 | C++11 | N3203 | Implicit special members | Remove on copy Undeprecate on dtor |
No action yet | pending... | pending... |
D.9 | C++98 | C++98 | C <*.h> headers |
Undeprecate | Remove now | New paper | pending... | |
D.10 | C++98 | C++20 | Requires: clauses | Remove now | No action | Defer to LWG | pending... | |
D.11 | C++98 | C++20 | P0768R1 | relops |
Remove now | Fix mandates | Remove now | pending... |
D.12 | C++98 | C++98 | char * streams |
No action | No action yet | No action | pending... | |
D.13 | C++11 | C++20 | P0767R1 | Deprecated type traits | Fix mandates | Remove now | Remove now | pending... |
D.14 | C++11 | C++20 | P1831R1 | volatile tuple API |
No action | No action | New paper | pending... |
D.15 | C++17 | C++20 | P1831R1 | volatile variant API |
Remove now | No action | Remove now | pending... |
D.16 | C++98 | C++17 | P0174R2 | std::iterator |
Remove now | Undeprecate | Remove now | pending... |
D.17 | C++11 | C++20 | P1252R2 | move_iterator::operator-> |
No action | Remove now | Remove now | pending... |
D.18 | C++11 | C++20 | P0718R2 | C API to use shared_ptr atomically |
Remove now | Fix mandates | SG1 | pending... |
D.19 | C++98 | C++20 | P0966R1 | basic_string::reserve() |
No action | Remove now | Remove now | pending... |
D.20 | C++11 | C++17 | P0618R0 | <codecvt> |
Remove now | No action yet | Remove now | pending... |
D.21 | C++11 | C++17 | P0618R0 | wstring_convert et al. |
Remove now | Fix mandates | SG16 | pending... |
D.22 | C++11 | C++20 | P0482R6 | Deprecated locale category facets | No action | No action | SG16 | pending... |
D.23 | C++17 | C++20 | P0482R6 | filesystem::u8path |
Fix mandates | Remove now | SG16 | pending... |
D.24 | C++11 | C++20 | P0883R2 | atomic operations | No action | Partially Remove | SG1 | pending... |
One sign of the success we have had with previous papers is that roughly 70% of the subclauses in Annex D were added in the latest C++20 standard, and only one feature remains that was deprecated by the C++11 standard. Notably, that is the last remaining core feature that was deprecated prior to C++20.
First deprecated: C++20
Warning since: Clang 10
Warn on emum/float compare: MSVC 19.22
Warn comparing enums: Clang 3.3 (earliest I could test) gcc 4.1.2 (earliest I could test)
Not tested: EDG
This feature was deprecated as part of the effort to make the new spaceship operator do the right thing, adopted by paper P1120R0. It potentially impacts on code written against C++98 and later standards.
With the introduction of the 3-way comparison "spaceship" operator,
there was a concern to avoid some implicit comparisons that might be lossy (due
to rounding of floating point values) or giving up intended type safety (by
using enums rather than integer types to indicate more than just a value).
While the 3-way comparison operator is specified to reject such comparisons,
the existing comparison operators were granted ongoing compatibility in these
cases, but deprecated. It is likely that most, but not all, such usage relying
on implicit conversion is a latent bug, and all such code would be clearer to
folks reading the code if the implicit conversions were made explicit. Note
that it is still possible to avoid explicit casts by using the unary
+
operator, forcing integral promotion.
While we would like to recommend the removal of these deprecated comparisons, bringing all the comparison operators into harmony for which types they interoperate on, the case for all uses being latent bugs is not as strong as for array comparisons (D.4), therefore it seems reasonable to allow another standard cycle encouraging compilers to issue deprecation warnings so that users can clean up their code. Hence, the strong recommendation is to do nothing for C++23, and strongly reconsider for C++26. The weak recommendation is to take decisive action and remove now, harmonizing the comparison operators.
Strong recommendation: Take no action.
No change to draft.
Weak recommendation: Remove this feature from C++23
7.4 Usual arithmetic conversions [expr.arith.conv]
- Many binary operators that expect operands of arithmetic or enumeration type cause conversions and yield result types in a similar way. The purpose is to yield a common type, which is also the type of the result. This pattern is called the usual arithmetic conversions, which are defined as follows:
- — If either operand is of scoped enumeration type (9.7.1), no conversions are performed; if the other operand does not have the same type, the expression is ill-formed.
- — If either operand is of unscoped enumeration type (9.7.1), and the other operand is of a different enumeration type or a floating-point type, the expression is ill-formed.
- — If either operand is of type
long double
, the other shall be converted tolong double
.- — Otherwise, if either operand is
double
, the other shall be converted todouble
.- — Otherwise, if either operand is
float
, the other shall be converted tofloat
.- — Otherwise, the integral promotions (7.3.6) shall be performed on both operands.56 Then the following rules shall be applied to the promoted operands:
- — If both operands have the same type, no further conversion is needed.
- — Otherwise, if both operands have signed integer types or both have unsigned integer types, the operand with the type of lesser integer conversion rank shall be converted to the type of the operand with greater rank.
- — Otherwise, if the operand that has unsigned integer type has rank greater than or equal to the rank of the type of the other operand, the operand with signed integer type shall be converted to the type of the operand with unsigned integer type.
- — Otherwise, if the type of the operand with signed integer type can represent all of the values of the type of the operand with unsigned integer type, the operand with unsigned integer type shall be converted to the type of the operand with signed integer type.
- — Otherwise, both operands shall be converted to the unsigned integer type corresponding to the type of the operand with signed integer type.
If one operand is of enumeration type and the other operand is of a different enumeration type or a floating-point type, this behavior is deprecated (D.1).
D.1 Arithmetic conversion on enumerations [depr.arith.conv.enum]
The ability to apply the usual arithmetic conversions (7.4) on operands where one is of enumeration type and the other is of a different enumeration type or a floating-point type is deprecated. [Note: Three-way comparisons (7.6.8) between such operands are ill-formed. —end note] [Example:enum E1 { e };enum E2 { f };bool b = e <= 3.7; // deprecatedint k = f - e; // deprecatedauto cmp = e <=> f; // error—end example]
No impact on library wording is anticipated by this removal.
Draft compatibility note for Annex C.
Concerns were raised about ongoing C compatibility. Any recommendation on removing deprecated support in D.1 will be forwarded to our WG14 liaison, to hopefully have a co-ordinated process for removing such features.
Concerns were raised over several issues of interoperating with different enumeration types, including expressions used to initialize global variables, and for array bounds. Similarly, metaprogramming idioms from C++98 would often use enumerations to denote result values (saving on storage for a static data member), and comparing such named values from different instantiations of the same template is the expected usage.
Another concern was raised for idioms for externally extending another enumeration without intruding on the original:
enum original { e_FIRST, ... , e_LAST }; enum extended { e_NEXT = e_LAST + 1, ... , e_MORE };
with the intent that the extended enumeration can interoperate with the original.
There seems fewer concerns for removing the interoperation with floating point
types, although the was some discussion of the workarounds. There is concern
that the "simple" forced promotion through unary operator
+
is too clever/cute, but that static_cast
, or
C-style casts, are verbose and risk converting to a different type than the
previous implicit integer promotion.
There is an ongoing concern of gaining more specific feedback on implementation experience with the deprecation warning before making any change, and further concerns about gaining experience with any removal that might silently change behavior in SFINAE contexts.
SF | F | N | A | SA |
1 | 2 | 7 | 6 | 4 |
No consensus
SF | F | N | A | SA |
8 | 1 | 4 | 8 | 6 |
No consensus
SF | F | N | A | SA |
1 | 2 | 10 | 7 | 1 |
No consensus
SF | F | N | A | SA |
7 | 10 | 4 | 0 | 1 |
Progress with this part
EWGI Review: Modified weak recommendation (remove just the enum/floating-point behavior)
7.4 Usual arithmetic conversions [expr.arith.conv]
- Many binary operators that expect operands of arithmetic or enumeration type cause conversions and yield result types in a similar way. The purpose is to yield a common type, which is also the type of the result. This pattern is called the usual arithmetic conversions, which are defined as follows:
- — If either operand is of scoped enumeration type (9.7.1), no conversions are performed; if the other operand does not have the same type, the expression is ill-formed.
- — Otherwise, if either operand is of unscoped enumeration type (9.7.1) and the other operand is of a floating-point type, the expression is ill-formed.
- — Otherwise, if either operand is of type
long double
, the other shall be converted tolong double
.- — Otherwise, if either operand is
double
, the other shall be converted todouble
.- — Otherwise, if either operand is
float
, the other shall be converted tofloat
.- — Otherwise, the integral promotions (7.3.6) shall be performed on both operands.56 Then the following rules shall be applied to the promoted operands:
- — If both operands have the same type, no further conversion is needed.
- — Otherwise, if both operands have signed integer types or both have unsigned integer types, the operand with the type of lesser integer conversion rank shall be converted to the type of the operand with greater rank.
- — Otherwise, if the operand that has unsigned integer type has rank greater than or equal to the rank of the type of the other operand, the operand with signed integer type shall be converted to the type of the operand with unsigned integer type.
- — Otherwise, if the type of the operand with signed integer type can represent all of the values of the type of the operand with unsigned integer type, the operand with unsigned integer type shall be converted to the type of the operand with signed integer type.
- — Otherwise, both operands shall be converted to the unsigned integer type corresponding to the type of the operand with signed integer type.
- If
oneeither operand is of unscoped enumeration type and the other operand is of a different enumeration typeor a floating-point type, this behavior is deprecated (D.1).No impact on library wording is anticipated by this removal.C.1.X Clause 7: Expressions [diff.cpp20.expr]
Affected subclause: 7.4 Note for editors: [expr.arith.conv]
Change: Unscoped enumerations do not implicitly promote to integral type in expressions with floating point types.
Rationale:
Provide rationaleEffect on original feature: A valid C++ 2020 involving both an unscoped enumeration and a floating point value will be rejected as ill-formed in this International Standard. Either argument could be explicitly conerted with a cast, or the enumeration could be explicitly promoted to an integer with unary operator +, for no change of meaning since C++ 2020. [Example:
enum multipliers { gigaseconds = 1'000'000 }; double century_ish = 3.14 * gigaseconds; // ill-formed; previously well-formed double century_est = 3.14 * +gigaseconds; // OK—end example]
C.5.3 Clause 7: expressions [diff.expr]
Affected subclause: 7.4 Note for editors: [expr.arith.conv]
Change: Enumerations cannot be used in expressions with floating point types.
Rationale:
Provide rationaleEffect on original feature: Deletion of semantically well-defined feature.
Difficulty of converting: Could be automated. Violations will be diagnosed by the C++ translator. The fix is to add a cast, or explicitly promote the
enum
object to an integer with unary operator+
. For example:How widely used: Common.
D.1 Arithmetic conversion on enumerations [depr.arith.conv.enum]
- The ability to apply the usual arithmetic conversions (7.4) on operands where one is of unscoped enumeration type and the other is of a different enumeration type
or a floating-point typeis deprecated. [Note: Three-way comparisons (7.6.8) between such operands are ill-formed. —end note] [Example:—end example]enum E1 { e }; enum E2 { f };bool b = e <= 3.7; // deprecatedint k = f - e; // deprecated auto cmp = e <=> f; // error
*this
by reference [depr.capture.this]First deprecated: C++20
Warning since: EDG 5.1, gcc 9, MSVC 19.22
No warnings: Clang
This feature was deprecated in C++20 by P0806R2. Its removal would potentially impact on programs written against C++11 or a later standard.
The concern addressed by the paper is that the implicit capture of
this
as a reference to data members on a default-capture
that copies is surprising and often misleading. C++20 introduced the explicit
capture of this
as a pointer, or *this
by value, to
clearly disambiguate the different use cases. However, as the syntax to
migrate to was introduced only in C++20, there was a clear desire for at least
one standard retaining the implicit capture support as deprecated while users
migrate their code at a time of their own choosing.
In the interest of clear code and fewer bugs from the surprising implicit defaults, the strong recommendation of this paper is to complete the work started in P0806 and remove the implicit capture from C++23. The weak recommendation is to retain deprecated support for another three years, and strongly reconsider removal in C++26. There is no suggestion to undeprecate.
Strong recommendation: Remove this feature from C++23
Draft core wording to enforce the removal (simply removing the deprecation notice is not enough).
D.2 Implicit capture of*this
by reference[depr.capture.this]
For compatibility with prior C++ International Standards, a lambda-expression with capture-default=
(7.5.5.2) may implicitly capture*this
by reference. [Example:struct X {int x;void foo(int n) {auto f = [=]() { x = n; }; // deprecated: x means this->x, not a copy thereofauto g = [=, this]() { x = n; }; // recommended replacement}};—end example]
No impact on library wording is anticipated by this removal.
Draft compatibility note for Annex C.
Weak recommendation: Take no action yet, strongly reconsider removal in C++26.
No change to draft.
Concerns were raised about a lack of user experience with these, and a concern that the proposed fix for code that removes the deprecated syntax is actually ill-formed in earlier dialects of the language, prior to C++17. Removing this feature may make supporting long-lived code bases that support multiple C++ dialects awkward.
There was no enthusiasm to undeprecate the feature without a separate paper.
SF | F | N | A | SA |
2 | 2 | 7 | 6 | 2 |
No consensus
EWGI Review: Take no action for C++23.
First deprecated: C++20
Warning since: Clang 9, EDG 6.0, gcc 10, MSVC 19.25
A more complete treatment of removing this feature and directly replacing it with support for multi-index array operators is presented in paper P2128R1 to which we might prefer to defer all handling of this topic.
This feature was deprecated for C++20 by paper P1161R3 to allow for a future extension in the language to support a simpler syntax for multiple-dimension arrays. Its removal would impact on code written against the C++98 and later standards, and potentially interoperating with C language headers.
There is a question over the value of removing this deprecated feature without also adding the extension for multiple-dimension arrays, which goes beyond the scope of this simpler review paper. The conservative view is that this would lead to code breakage for no clear benefit,so the strong recommendation is to do retain this feature as part of the overall review. The alternative view is that immediately applying a new meaning to existing syntax as part of an updated standard is a risk we could avoid by requiring at least one standard to make such code ill-formed, rather than merely deprecated. The weak recommendation is to remove this feature from C++23 to better enable repurposing the syntax in C++26, or beyond.
On as separate note regarding wording, we observe that the convention for deprecating a feature in the core language is to normatively define the deprecation in the core clauses (4-15) and then call back to that deprecation notice from Annex D. The C++20 standard has this backwards for this particular deprecation, with the two references to deprecation in the core clauses being Note:s, and the only normative text being in Annex D. At least one of the two Note:s should become normative, if the feature is retained.
Strong recommendation: Clean up wording.
7.6.1.1 Subscripting [expr.sub]
[Note:A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression is deprecated; see D.3.—end note]7.6.20 Comma operator [expr.comma]
[Note:A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression is deprecated; see D.3.—end note]
Weak recommendation: Remove this feature from C++23
7.6.1.1 Subscripting [expr.sub]
[Note:A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression is ill-formed.deprecated; see D.3. —end note]7.6.20 Comma operator [expr.comma]
- [Note: A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression is ill-formed (7.6.1.1)
deprecated; see D.3. —end note]No impact on library wording is anticipated by this removal.C.1.X Clause 7: Expressions [diff.cpp20.expr]
Affected subclause: 7.6.1.1 Note for editors: [expr.sub]
Change: Cannot parse unparenthesized comma expressions as subscript expressions.
Rationale: The old behavior was surprising for users familiar with other languages. It was removed to allow a future extension to support overloading the subscript operator for multiple arguments.
Effect on original feature: A valid C++ 2020 program using a comma expression as the argument to a subscript expression will be rejected as ill-formed in this International Standard. The intended comma expression can be enclosed in parentheses for no change of meaning since C++ 2020. [Example:
void f(int *a, int b, int c) { a[b,c]; // ill-formed; previously well-formed a[(b,c)]; // OK }Add an example of a DSEL—end example]
C.5.3 Clause 7: expressions [diff.expr]
Affected subclause: 7.6.1.1 Note for editors: [expr.sub]
Change: Cannot parse unparenthesized comma expressions as subscript expressions.
Rationale: The old behavior was surprising for users familiar with other languages. It was removed to allow a future extension to support overloading the subscript operator for multiple arguments.
Effect on original feature: Deletion of semantically well-defined feature.
Difficulty of converting: Could be automated. Violations will be diagnosed by the C++ translator. The fix is to add parentheses. For example:
void f(int *a, int b, int c) { a[b,c]; // ill-formed; previously well-formed a[(b,c)]; // OK }How widely used: Rare.
D.3 Comma operator in subscript expressions [depr.comma.subscript]
A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression (7.6.1.1) is deprecated. [Note: A parenthesized comma expression is not deprecated. —end note] [Example:void f(int *a, int b, int c) {a[b,c]; // deprecateda[(b,c)]; // OK}—end example]
Several concerns were raised that might be better addressed for when EWG take a final decision. First, there is the existing paper, P2128, that seeks to both remove this deprecated feature, and replace it with a new meaning. One option is to delegate entirely to that paper, but loud concerns were also raised about changing meaning, and silently changing behavior, in the same standard. There is some support for removing the deprecated behavior now, and holding P2128 back until C++26. From the perspective of this paper, the notion is to proceed and treat P2128 independently.
Concerns were raised about implementation experience. While it is notable that all current compiler with C++20 support warn on this deprecation, some of those compilers are only a month or two old, and we get experience only when folks compile in the experimental C++20 modes. While we expect experience to grow, it is still minimal. It was also noted that the original paper performed a survey of popular open source libraries to assess impact, and hit only a vanishingly small set of matches in a deprecated Boost library: link
A final concern was raised about ongoing C compatibility, which will be forwarded to our WG14 liaison, to hopefully have a co-ordinated process for removing this feature.
SF | F | N | A | SA |
8 | 9 | 2 | 1 | 1 |
Follow up with this direction
EWGI Review: Accept the weak recommendation.
First deprecated: C++20
Warning since: Clang 10, MSVC 19.22
QoI Warning Clang 3.3 (earliest I could test)
No warnings: gcc
Not tested: EDG
This feature was deprecated as part of the effort to make the new spaceship operator do the right thing, adopted by paper P1120R0. It potentially impacts on code written against C++98 and later standards.
The deprecated comparison operator for arrays was not so much a deliberately designed feature, but accidental oversight that array-to-pointer decay would kick in, and so we compare whether two arrays are literally the same array at the same address, rather than whether two distinct arrays have the same contents. Identity tests are typically performed by explicitly taking the address of the objects we wish to validate; it would be highly unusual to rely on an implicit pointer decay to perform that task. The trick, if performed intentionally, offers no efficiency gain over explicitly taking the address of the array, but would fool a large number of subsequent code readers and reviewers who are not familiar with this trick. We do note that function comparison performs exactly the same decay, and users are not surprised that comparing functions is an identity test.
Given the likelihood that any usage of this comparison operator is a bug waiting to be detected, this could be a real concern for software reliability. Therefore, the strong recommendation is to remove this feature immediately from C++23. However, note for wording, that the comparison of an array with a pointer value is not deprecated in C++20, is reasonably idiomatic for practitioners of C++, and is intended to continue to be supported. This mimics the behavior of the spaceship operator.
Given the feature was so recently deprecated, the weak recommendation is to sit on the feature for another 3 years giving users more time to find (presumably benign) uses in their code and correct at a time of their choosing. We would expect to strongly recommend the removal again in another three years though, rather than consider undeprecation for such a feature.
Strong recommendation: Remove this feature from C++23
7.6.9 Relational operators [expr.rel]
- The relational operators group left-to-right. [Example:
a<b<c
means(a<b)<c
and not(a<b)&&(b<c)
. — end example]The lvalue-to-rvalue (7.3.1)relational-expression : compare-expression relational-expression < compare-expression relational-expression > compare-expression relational-expression <= compare-expression relational-expression >= compare-expression, array-to-pointer (7.3.2),and function-to-pointer(7.3.3) standard conversions are performed on the operands. If at least one of the operands is a pointer, array-to-pointer conversions (7.3.2) are performed.The comparison is deprecated if both operands were of array type prior to these conversions (D.4).- The converted operands shall have arithmetic, enumeration, or pointer type. The operators
<
(less than),>
(greater than),<=
(less than or equal to), and>=
(greater than or equal to) all yieldfalse
ortrue
. The type of the result isbool
.- The usual arithmetic conversions (7.4) are performed on operands of arithmetic or enumeration type. If both operands are pointers, pointer conversions (7.3.11) and qualification conversions (7.3.5) are performed to bring them to their composite pointer type (7.2.2). After conversions, the operands shall have the same type.
- The result of comparing unequal pointers to objects ...
7.6.10 Equality operators [expr.eq]
equality-expression : relational-expression equality-expression == relational-expression equality-expression != relational-expression
- The
==
(equal to) and the!=
(not equal to) operators group left-to-right. The lvalue-to-rvalue (7.3.1), array-to-pointer (7.3.2),and function-to-pointer (7.3.3) standard conversions are performed on the operands.The comparison is deprecated if both operands were of array type prior to these conversions (D.4).- If at least one of the operands is a pointer, array-to-pointer conversions (7.3.2), pointer conversions (7.3.11), function pointer conversions (7.3.13), and qualification conversions (7.3.5) are performed on both operands to bring them to their composite pointer type (7.2.2).
Comparing pointers is defined as follows:- The converted operands shall have arithmetic, enumeration, pointer, or pointer-to-member type, or type
std::nullptr_t
. The operators==
and!=
both yieldtrue
orfalse
, i.e., a result of typebool
. In each case below, the operands shall have the same type after the specified conversions have been applied.- Comparing pointers is defined as follows:
- — If one pointer represents the address of a complete object, and another pointer represents the address one past the last element of a different complete object,79 the result of the comparison is unspecified.
- — Otherwise, if the pointers are both null, both point to the same function, or both represent the same address (6.8.2), they compare equal.
- — Otherwise, the pointers compare unequal.
- If at least one of the operands is a pointer to member, ...
No impact on library wording is anticipated by this removal.C.1.X Clause 7: Expressions [diff.cpp20.expr]
Affected subclause: 7.6.9 and 7.6.10 Note for editors: [expr.rel] and [expr.eq]
Change: Cannot compare two objects of array type.
Rationale: The old behavior was confusing, as it did not compare the contents of the two arrays, but compare their addresses. Depending on context, this would either report whether the two arrays were the same object, or have an unspecified result.
Effect on original feature: A valid C++ 2020 program directly comparing two array objects will be rejected as ill-formed in this International Standard. [Example:
int arr1[5]; int arr2[5]; bool same = arr1 == arr2; // ill-formed; previously well-formed bool idem = arr1 == +arr2; // compare addresses, unspecified result—end example]
C.5.3 Clause 7: expressions [diff.expr]
Affected subclause: 7.6.9 and 7.6.10 Note for editors: [expr.rel] and [expr.eq]
Change: Cannot compare two objects of array type.
Rationale: The old behavior was confusing, as it did not compare the contents of the two arrays, but compare their addresses. Depending on context, this would either report whether the two arrays were the same object, or have an unspecified result.
Effect on original feature: Deletion of feature with unspecified behavior.
Difficulty of converting: Violations will be diagnosed by the C++ translator.
How widely used: Rare. In the cases where the result is well defined, it is reporting whether both arguments are the same object, using the same name.
D.4 Array comparisons [depr.array.comp]
Equality and relational comparisons (7.6.10, 7.6.9) between two operands of array type are deprecated. [Note: Three-way comparisons (7.6.8) between such operands are ill-formed. —end note] [Example:int arr1[5];int arr2[5];bool same = arr1 == arr2; // deprecated, same as &arr1[0] == &arr2[0],// does not compare array contentsauto cmp = arr1 <=> arr2; // error—end example]
Weak recommendation: Take no action.
No change to draft.
It was noted that comparison of pointer values is often unspecified, rather than well-defined, unless both arrays happen to be members of the same class, or are both elements of the same multi-dimensional array.
General agreement that this is a good opportunity to remove a landmine from the language that offers little benefit, even when correctly used as intended.
A concern was raised about ongoing C compatibility, which will be forwarded to our WG14 liaison, to hopefully have a co-ordinated process for removing this feature.
SF | F | N | A | SA |
9 | 9 | 3 | 0 | 0 |
Follow up with this direction
EWGI Review: Accept the strong recommendation.
First deprecated: C++20
Warning since: Clang 10, EDG 6.0, gcc 10
No warnings: MSVC
The volatile
keyword is an original part of the C legacy for C++,
and describes constraints on programs intended to model hardware changing
values beyond the program's control. As this entered the type system of
C++, certain interactions were discovered to be troublesome, and latent
bugs that could be detected at the time of program translation go unreported.
The paper
breaks down each context where the volatile
keyword can be used,
and deprecated those uses that are unconditionally dangerous, or serve no
good purpose. This paper is the first opportunity to go further, and
remove those use cases after 3 years of giving users deprecation warnings.
A quick micro-analysis suggests the main concerns of the first two paragraphs
are read/modify/write operations, where by the nature of volatile objects, the
value being rewritten may have changed since read and modified. This kind of
pattern is most likely in old (pre-C++11) code using volatile
as a
poor proxy for atomic. Since we will have over a decade of real atomic support
in the language when C++23 ships, it could be desirable to further encourage
such code (when compiled in the latest dialect) to adapt to the memory model
and its stronger guarantees.
The third paragraph addresses function arguments and return values. These are
temporary or elided objects created entirely by the compiler, and guaranteed to
not display the uncertainty of value implied by the volatile
keyword. As such, any use is redundant and misleading, so it would be helpful
to remove this facility sooner rather than later, and have one fewer oddity to
teach when learning (and understanding) the language. The biggest concern
would be for compatibility with C code, that may still use this feature in its
headers. To mitigate, me may consider removing volatile
function
parameters and return values for only functions with
extern "C++"
linkage.
The fourth paragraph considers volatile qualifier in structured bindings, and can affect only code written since C++17, that will have been deprecated as long as it was non-deprecated when C++23 is published. It would be good to remove this now, before more deprecated code is written.
As outright removal is introducing potential breakage into programs that have been compiling successfully since C++98, this paper strongly recommends to hold this feature in Annex D as deprecated for another 3 years, and strongly consider taking further action for C++26. However, as there is the potential to diagnose real bugs for users to fix, the weak recommendation is to remove these deprecated use cases immediately from C++23. It would also be possible to review each of the 4 noted usages separately, and remove only the features with lowest risk from removal, notably paragraphs 3 and 4.
Strong recommendation: Take no action.
No change to draft.
Weak recommendation: Remove this feature from C++23
Note that the core terms arithmetic type and pointer type refer only to cv-unqualified types, and cv-qualifiers are incorporated only for the larger classification of scalar types; therefore, much of the deprecated wording in clause 7 is already redundant - this will be core issue 2448. In principle, we need to add words to enable the feature that C++20 intended to deprecate! The wording issue goes back to C++98 though, as current implementations provide the intended deprecated behavior, not a literal interpretation of the standard.
7.6.1.5 Increment and decrement [expr.post.incr]
- The value of a postfix
++
expression is the value of its operand. [Note: The value obtained is a copy of the original value. —end note] The operand shall be a modifiable lvalue. The type of the operand shall be an arithmetic type other thancvbool
, or a pointer to a complete object type.An operand with volatile-qualified type is deprecated; see D.5.The value of ...7.6.2.2 Increment and decrement [expr.pre.incr]
- The operand of prefix
++
is modified (3.1) by adding1
. The operand shall be a modifiable lvalue. The type of the operand shall be an arithmetic type other thancvbool
, or a pointer to a completely-defined object type.An operand with volatile-qualified type is deprecated; see D.5.The result is ...7.6.19 Assignment and compound assignment operators [expr.ass]
A simple assignment whose left operand is of avolatile
-qualified type is deprecated (D.5) unless the (possibly parenthesized) assignment is a discarded-value expression or an unevaluated operand.Draft a replacement paragraph so that assignment tovolatile
-qualified types remains valid, but only through discarded-value expressions. If not an ABI consideration, could we make the built-in assignment operators forvolatile
-qualified types returnvoid
?- The behavior of an expression of the form
E1
op= E2
is equivalent toE1 = E1
opE2
except thatE1
is evaluated only once.Such expressions are deprecated ifForE1
hasvolatile
-qualified type; see D.5.+=
and-=
,E1
shall either have arithmetic type or be a pointer to a possibly cv-qualified completely-defined object type. In all other cases,E1
shall have arithmetic type.9.3.3.5 Functions [dcl.fct]
It is expected that removing support for volatile-qualified function parameters will have ripples more widely through the standard, and that potentially cv-qualified types may become potentially const-qualified types in a number of places, after a deeper audit.
- The parameter-declaration-clause determines the arguments that can be specified, and their processing, when the function is called. [Note: The parameter-declaration-clause is used to convert the arguments specified on the function call; see 7.6.1.2. —end note] If the parameter-declaration-clause is empty, the function takes no arguments. A parameter list consisting of a single unnamed parameter of non-dependent type
void
is equivalent to an empty parameter list. Except for this special case, a parameter shall not have type cvvoid
. A parameterwithshall not have a volatile-qualified typeis deprecated; see D.5. If the parameter-declaration-clause terminates with an ellipsis or a function parameter pack (13.7.3), the number of arguments shall be equal to or greater than the number of parameters that do not have a default argument and are not function parameter packs. Where syntactically correct and where “...” is not part of an abstract-declarator, “, ...” is synonymous with “...”. [Example: The declarationint printf(const char*, ...);declares a function that can be called with varying numbers and types of arguments.printf("hello world"); printf("a=%d b=%d", a, b);However, the first argument must be of a type that can be converted to aconst char*
. —end example] [Note: The standard header<cstdarg>
(17.13.1) contains a mechanism for accessing arguments passed using the ellipsis (see 7.6.1.2 and 17.13). —end note]
A volatile-qualified return type is deprecated; see D.5.A return type shall not be volatile-qualified. [Note: References to volatile-qualified types are permitted, as reference types do not have a top level cv-qualifier. —end note]9.6 Structured binding declarations [dcl.struct.bind]
As a reminder, the grammar for a structured binding is the following simple-declaration:attribute-specifier-seqopt decl-specifier-seq ref-qualifieropt [ identifier-list ] initializer ;
- A structured binding declaration introduces the identifiers
v0
,v1
,v2
, . . . of the identifier-list as names (6.4.1) of structured bindings. Let cv denote the cv-qualifiers in the decl-specifier-seq and S consist of the storage-class-specifiers of the decl-specifier-seq (if any). A cv that includesvolatile
is deprecated; see D.5. First, a variable with a unique namee
is introduced. If the assignment-expression in the initializer has array typeA
and no ref-qualifier is present,e
is defined byattribute-specifier-seqopt Sand each element is copy-initialized or direct-initialized from the corresponding element of the assignment-expression as specified by the form of the initializer. Otherwise,cv A e ;
e
is defined as-if byattribute-specifier-seqopt decl-specifier-seq ref-qualifieroptwhere the declaration is never interpreted as a function declaration and the parts of the declaration other than the declarator-id are taken from the corresponding structured binding declaration. The type of the id-expressione
initializer;
e
is calledE
. [Note:E
is never a reference type (7.2). —end note]Ideally, this whole paragraph should be rewritten to avoid introducing a cv that can never include thevolatile
qualifier.- If the initializer refers to one of the names introduced by the structured binding declaration, the program is ill-formed.
- If
E
is an array type with element typeT
, the number of elements in the identifier-list shall be equal to the number of elements ofE
. Eachvi
is the name of an lvalue that refers to the element i of the array and whose type isT
; the referenced type isT
. [Note: The top-level cv-qualifiers ofT
are cv. —end note] [Example:—end example]auto f() -> int(&)[2]; auto [ x, y ] = f(); // x and y refer to elements in a copy of the array return value auto& [ xr, yr ] = f(); // xr and yr refer to elements in the array referred to by f's return value- Otherwise, if the qualified-id
std::tuple_size<E>
names a complete class type with a member namedvalue
, the expressionstd::tuple_size<E>::value
shall be a well-formed integral constant expression and the number of elements in the identifier-list shall be equal to the value of that expression. Leti
be an index prvalue of typestd::size_t
corresponding tovi
. The unqualified-idget
is looked up in the scope ofE
by class member access lookup (6.5.5), and if that finds at least one declaration that is a function template whose first template parameter is a non-type parameter, the initializer ise.get<i>()
. Otherwise, the initializer isget<i>(e)
, whereget
is looked up in the associated namespaces (6.5.2). In either case,get<i>
is interpreted as a template-id. [Note: Ordinary unqualified lookup (6.5.1) is not performed. —end note] In either case,e
is an lvalue if the type of the entitye
is an lvalue reference and an xvalue otherwise. Given the typeTi
designated bystd::tuple_element<i, E>::type
and the typeUi
designated by eitherTi&
orTi&&
, whereUi
is an lvalue reference if the initializer is an lvalue and an rvalue reference otherwise, variables are introduced with unique namesri
as follows:SEach vi is the name of an lvalue of typeUi ri =
initializer;
Ti
that refers to the object bound tori
; the referenced type isTi
.- Otherwise, all of
E
's non-static data members shall be direct members of E or of the same base class ofE
, well-formed when named ase.name
in the context of the structured binding,E
shall not have an anonymous union member, and the number of elements in the identifier-list shall be equal to the number of non-static data members ofE
. Designating the non-static data members of E asm0
,m1
,m2
, . . . (in declaration order), eachvi
is the name of an lvalue that refers to the membermi
ofe
and whose type is cvTi
, whereTi
is the declared type of that member; the referenced type is cvTi
. The lvalue is a bit-field if that member is a bit-field. [Example:The type of the id-expressionstruct S { int x1 : 2; volatile double y1; }; S f(); const auto [ x, y ] = f();x
is "const int
", the type of the id-expressiony
is "const volatile double
". —end example]Note that the use ofvolatile
in this example remains valid for structured bindings, even after the removal of the deprecated parts.12.7 Built-in operators [over.built]
In the remainder of this subclause, vq represents eithervolatile
or no cv-qualifier.- For every
pair (arithmetic type other thanT
, vq), whereT
is anbool
, there exist candidate operator functions of the formvqT & operator++(vqT &); T operator++(vqT &, int);- For every
pair (arithmetic type other thanT
, vq), whereT
is anbool
, there exist candidate operator functions of the formvqT & operator--(vqT &); T operator--(vqT &, int);- For every
pair (typeT
, vq)T
, whereT
is a cv-qualified or cv-unqualified object type, there exist candidate operator functions of the formT*vq& operator++(T*vq&); T*vq& operator--(T*vq&); T* operator++(T*vq&, int); T* operator--(T*vq&, int);
- For every
triple (pair (L
, vq,R
)L
,R
), whereL
is an arithmetic type, andR
is a floating-point or promoted integral type, there exist candidate operator functions of the formvqL & operator=(vqL &, R); void operator=(volatile L &, R);vqL & operator*=(vqL &, R);vqL & operator/=(vqL &, R);vqL & operator+=(vqL &, R);vqL & operator-=(vqL &, R);- For every
pair (typeT
, vq)T
, there exist candidate operator functions of the formT*vq& operator=(T*vq&, T*); void operator=(T* volatile &, T*);- For every
pair (enumeration or pointer-to-member type, there exist candidate operator functions of the formT
, vq), whereT
is anvqT & operator=(vqT &, T); void operator=(volatile T &, T);- For every
pair (typeT
, vq)T
, whereT
is a cv-qualified or cv-unqualified object type, there exist candidate operator functions of the formT*vq& operator+=(T*vq&, std::ptrdiff_t); T*vq& operator-=(T*vq&, std::ptrdiff_t);- For every
triple (pair (L
, vq,R
)L
,R
), whereL
is an integral type, andR
is a promoted integral type, there exist candidate operator functions of the formvqL & operator%=(vqL &, R);vqL & operator<<=(vqL &, R);vqL & operator>>=(vqL &, R);vqL & operator&=(vqL &, R);vqL & operator^=(vqL &, R);vqL & operator|=(vqL &, R);
D.5 Deprecated volatile types [depr.volatile.type]
Postfix++
and--
expressions (7.6.1.5) and prefix++
and--
expressions (7.6.2.2) ofvolatile
-qualified arithmetic and pointer types are deprecated.[Example:volatile int velociraptor;++velociraptor; // deprecated—end example]Certain assignments where the left operand is avolatile
-qualified non-class type are deprecated; see 7.6.19.[Example:int neck, tail;volatile int brachiosaur;brachiosaur = neck; // OKtail = brachiosaur; // OKtail = brachiosaur = neck; // deprecatedbrachiosaur += neck; // deprecatedbrachiosaur = brachiosaur + neck; // OK—end example]A function type (9.3.3.5) with a parameter withvolatile
-qualified type or with avolatile
-qualified return type is deprecated.[Example:volatile struct amber jurassic(); // deprecatedvoid trex(volatile short left_arm, volatile short right_arm); // deprecatedvoid fly(volatile struct pterosaur* pteranodon); // OK—end example]A structured binding (9.6) of avolatile
-qualified type is deprecated.[Example:struct linhenykus { short forelimb; };void park(linhenykus alvarezsauroid) {volatile auto [what_is_this] = alvarezsauroid; // deprecated// ...}—end example]
Draft compatibility note for Annex C.
EWGI Review: To be determined...
First deprecated: C++17
Warning since: (none)
No warnings: Clang EDG gcc MSVC
Static constexpr data members were added to the language in C++11, as part of the initial constexpr feature. However, they still required a definition of the member outside the class. When inline variables were added in C++17 (P0386R2) then static constexpr data members became implicitly inline, and the external definition became redundant, so was deprecated.
When we considered this feature for removal in C++20, it was thought the change was too recent for users to have time to adjust, and recommended we reconsider removal in C++23 (or later). By the time we publish C++23, this feature will have been deprecated for exactly as long as it was non-deprecated, 6 years. The suggested fix to make code confirming again is simple: remove the offending definition as it is redundant; no other changes are necessary. For code that wishes to be compatible across a range of standard dialects, the definition can be guarded by the preprocessor:
#if __cplusplus < 201703 constexpr Type Class::Member; #endif
If we were to remove this feature early in the C++23 cycle, and compiler vendors expose this through their experimental C++23 support, we would get sufficient feedback if removal of this feature were a step too far, too soon. The strong recommendation is to remove this feature from C++23. Note, however, that at the time of publishing this paper, no current compiler warns of this deprecation, in either C++17 or C++20 build modes.
The feature seems relatively small and harmless; it is not clear that there is a huge advantage to removing it immediately from C++23, although it would be one less corner case to teach. The weak recommendation is to retain this feature, advocating removal in C++26, by which time the community will have had most of a decade to clean up old C++11/14 code.
Strong recommendation: remove this facility from C++23.
6.1 Declarations and definitions [basic.def]
- A declaration is a definition unless
- — it declares a function without specifying the function's body (11.4),
- — it contains the
extern
specifier (9.2.1) or a linkage-specification22 (9.11) and neither an initializer nor a function-body,- — it declares a non-inline static data member in a class definition (11.4, 11.4.8),
— it declares a static data member outside a class definition and the variable was defined within the class with theconstexpr
specifier (this usage is deprecated; see D.6),- — it is introduced by an elaborated-type-specifier (11.3),
- — it is an opaque-enum-declaration (9.7.1),
- — it is a template-parameter (13.2),
- — it is a parameter-declaration (9.3.3.5) in a function declarator that is not the declarator of a function-definition,
- — it is a
typedef
declaration (9.2.3),- — it is an alias-declaration (9.2.3),
- — it is a using-declaration (9.9),
- — it is a deduction-guide (13.7.1.2),
- — it is a static_assert-declaration ((9.1),
- — it is an attribute-declaration ((9.1),
- — it is an empty-declaration (9.1),
- — it is a using-directive (9.8.3),
- — it is a using-enum-declaration (9.7.2),
- — it is a template-declaration (13.1) whose template-head is not followed by either a concept-definition or a declaration that defines a function, a class, a variable, or a static data member.
- — it is an explicit instantiation declaration (13.9.2), or
- — it is an explicit specialization (13.9.3) whose declaration is not a definition.
11.4.8.2 Static data members [class.static.data]
- If a non-volatile non-inline
const
static data member is of integral or enumeration type, its declaration in the class definition can specify a brace-or-equal-initializer in which every initializer-clause that is an assignment-expression is a constant expression (7.7). The member shall still be defined in a namespace scope if it is odr-used (6.3) in the program and the namespace scope definition shall not contain an initializer. An inline static data member may be defined in the class definition and may specify a brace-or-equal-initializer.If the member is declared with theDeclarations of other static data members shall not specify a brace-or-equal-initializer.constexpr
specifier, it may be redeclared in namespace scope with no initializer (this usage is deprecated; see D.6).
D.6 Redeclaration of static constexpr data members [depr.static.constexpr]
For compatibility with prior C++ International Standards, aconstexpr
static data member may be redundantly redeclared outside the class with no initializer. This usage is deprecated. [ Example:— end example ]struct A {static constexpr int n = 5; // definition (declaration in C++ 2014)};constexpr int A::n; // redundant declaration (definition in C++ 2014)
No impact on library wording is anticipated by this removal.
Draft compatibility note for Annex C.
Weak recommendation: Take no action yet, consider again for C++26.
No change to draft.
EWGI Review: To be determined...
First deprecated: C++20
Warning since: (none)
No warnings: Clang EDG gcc MSVC
This feature was deprecated as part of the effort to cleanly introduce modules into the language, and was adopted by paper P1815R2. It potentially impacts on code written against C++98 and later standards.
This feature was deprecated only at the final meeting of the C++20 development cycle, and at the time this paper is written, there is no experience with how much code has been impacted by this deprecation. As such, it is too early to make any recommendation for a change with respect to this feature.
Strong recommendation: take no action.
No change to draft.
Weak recommendation: take no action.
No change to draft.
EWGI Review: To be determined...
First deprecated: C++11
Warning since: Clang 10, gcc 9
No warnings: MSVC
Not tested: EDG
Relevant issues: Core #2132
Core issue #2132 was discussed on the EWG telecon of 2020/04/29. The consensus on the call was that this is a contentious issue that should be handled by a paper in its own right.
This feature was deprecated towards the end of the C++11 development cycle by paper N3203 and was heavily relied on prior to that. That said, the deprecated features are implicitly deleted if any move operations are declared, and users have become familiar with this idiom over the last decade, and may now find the older deprecated behavior jarring.
When we reviewed this facility for C++20, there was something of an impasse. There was concern that removing this feature ever would break too much code to ever consider it. Conversely, there was no way to achieve a consensus to undeprecate these features. Finally, it was noted that no compiler had ever issued a deprecation warning for these constructs, and all compiler implementers present opined that they would never implement such a warning, as it would flag far too much code to be a usable warning.
Since then, gcc has indeed implemented the warning, and incorporated into
its -Wall
setting, the flag to turn on all the warnings that have
a vanishingly small false-positive rate. Similarly, Clang 10 has released with
the same warning enabled since R0 of this document was published. The feedback
from that exercise was that coupling the destructor into the rule to delete
copy operations was indeed far too noisy, and unlikely to be realizable in
production code. Legitimate use case to implement a destructor unrelated to
copy operation occur too frequently in practice, such as a simple act of
logging or asserting invariants at the end of an object's lifetime. However,
the coupling of the two copy operations produced a much lower (and tractable)
number of warnings, and found real bugs along with a number of false positives.
The workaround for the issue in the vast majority of cases is to simply declare
the other copy member in the class definition as = default
so the
fix is (mostly) simple, easily hinted, and modern tools can even offer to apply
the fix for you as part of the warning. There may be a concern that the fix
relies on a C++11 feature, and so requires more work for code maintaining C++03
compatibility. This is true, but for C++03 there is no functional difference
between generating the default and writing it by hand. For C++11 the potential
difference is that a user-provided copy operation is no longer trivial, but the
distinction of decomposing PODs into trivial operations and standard layout
types matters only for C++11 onwards, as user-providing either operation causes
the type to not be a POD in C++03.
One unusual case for consideration is an idiom for defining iterator types
where the const_iterator
can be implicitly constructed from the
plain iterator
. One way to achieve this is for the two iterators
to be instantiations of the same class template, but with a const
qualified type argument for the const_iterator
, and use a
metafunction to produce an alias for the const_iterator
in both
cases. An implicit constructor declared with this aliased name will be a copy
constructor for the plain iterator
, or an additional overloaded
constructor for the const_iterator
. However, we cannot declare
both constructors, as in the case of the plain iterator
that would
be a duplicate declaration, so for the case of const_iterator
we
must rely on the implicit copy constructor declaration. As the declared
constructor is not the copy constructor for all instantiations, it cannot be
defaulted. In a common implementation of this idiom we similarly rely on the
implicitly declared copy-assignment operator, which would now have to be
user-declared. However, due to the declaration matching the signature of the
copy assignment operator for only the iterator
instantiations, we
cannot simply default this new declaration, and must explicitly provide a
definition too.
template<class T> class MyIterator { T * d_ptr = nullptr; public: using base_iterator = MyIterator<std::remove_const_t<T>>; MyIterator() = default; explicit MyIterator(T * ptr) : d_ptr{ptr} {} MyIterator(base_iterator const & it) : d_ptr{it.operator->()} {} // This is the copy constructor if 'T' is not 'const'-qualified, // otherwise the copy constructor is implicitly declared. MyIterator& operator=(base_iterator const & it) { // This is the copy-assignment operator if 'T' is not 'const'-qualified, // otherwise the copy-assignment operator is implicitly declared. d_ptr = it.operator->(); return *this; } T * operator->() const noexcept { return d_ptr; } // ... };
The strong recommendation of this paper is mixed: it is time to enforce the rule coupling the copy operations in the same manner as the move operations are coupled, and remove the legacy auto-generation if just one is user-defined. However, at the same time, we should uncouple the destructor from this rule, undeprecating its impact on copy operations. We might consider uncoupling the destructor from move operations for consistency, but that would be a separate paper.
The weak recommendation is to continue another standard cycle with this feature deprecated as-is, but consider what actions and research might demonstrate we are ready to remove this coupling for C++26.
Strong recommendation: take decisive action early in the C++23 development cycle, so early adopters can shake out the remaining cost of updating old code. Note that this would be both an API and ABI breaking change, and it would be good to set that precedent early if we wish to allow such breakage in C++23.
11.4.4.2 Copy/move constructors [class.copy.ctor]
If the class definition does not explicitly declare a copy constructor, a non-explicit one is declared implicitly. If the class definition declares a move constructor or move assignment operator, the implicitly declared copy constructor is defined as deleted; otherwise, it is defined as defaulted (9.5). The latter case is deprecated if the class has a user-declared copy assignment operator or a user-declared destructor.If the definition of a class
X
does not explicitly declare a copy constructor, a non-explicit one will be implicitly declared as defaulted if and only if
- —
X
does not have a user-declared copy assignment operator,- —
X
does not have a user-declared move constructor, and- —
X
does not have a user-declared move assignment operator.User declared destructors are deliberately omitted from this list. This wording undeprecates that concern - we could insert additional wording to retain the deprecated status when the class has a user-declared destructor. Note that we probably want the term "user-provided" as a user-declared= default
destructor should not be a concern.11.4.5 Copy/move assignment operator [class.copy.assign]
If the class definition does not explicitly declare a copy assignment operator, one is declared implicitly. If the class definition declares a move constructor or move assignment operator, the implicitly declared copy assignment operator is defined as deleted; otherwise, it is defined as defaulted (9.5). The latter case is deprecated if the class has a user-declared copy constructor or a user-declared destructor.If the definition of a class
X
does not explicitly declare a copy assignment operator, one will be implicitly declared as defaulted if and only if
- —
X
does not have a user-declared copy constructor,- —
X
does not have a user-declared move constructor, and- —
X
does not have a user-declared move assignment operator.User declared destructors are deliberately omitted from this list. This wording undeprecates that concern - we could insert additional wording to retain the deprecated status when the class has a user-declared destructor. Note that we probably want the term "user-provided" as a user-declared= default
destructor should not be a concern.- The implicitly-declared copy assignment operator for a class
X
will have the form...
D.8 Implicit declaration of copy functions [depr.impldec]
The implicit definition of a copy constructor as defaulted is deprecated if the class has a user-declared copy assignment operator or a user-declared destructor. The implicit definition of a copy assignment operator as defaulted is deprecated if the class has a user-declared copy constructor or a user-declared destructor (15.4, 15.8). In a future revision of this International Standard, these implicit definitions could become deleted (11.4).
Draft compatibility note for Annex C.
Weak recommendation: take no action now, and plan to remove this feature in the next revision of the standard that will permit both ABI and API breakage.
No change to draft.
EWGI Review: To be determined...
First deprecated: C++98
The basic C library headers are an essential compatibility feature, and not going anywhere anytime soon. However, there are certain C++ specific counterparts that do not bring value, particularly where the corresponding C header's job is to supply macros that masquerade as keywords already present in the C++ language.
One possibility to be more aggressive here, following the decision to not adopt
all C11 headers as part of the C++ mapping to C, would be to remove
those same C headers from the subset that must be shipped with a C++ compiler.
The corresponding change for the C++ <cstd*>
was accepted for
C++20. This would not prevent those headers being supplied, as part of a full
C implementation, but would indicate that they have no value to a C++ system.
Finally, it seems clear that the C headers will be retained essentially forever, as a vital compatibility layer with C and POSIX. After two decades of deprecation, not a single compiler provides a deprecation warning on including these headers, and it is difficult to imagine circumstances where they would, as any use of a third party C library by a C++ project is almost guaranteed to include one of these headers, at least indirectly. Therefore, it seems appropriate to undeprecate the headers, and find a home for them as a compatibility layer in the main standard. It is also possible that we will want to explore a different approach in the future once modules are part of C++, and the library is updated to properly take advantage of the new language feature.
The strong recommendation of this paper is to undeprecate the C library as an
essential compatibility layer with the rest of the world, optionally removing
the vacuous headers. A note in favor of removing the vacuous headers is that
the C standard itself is moving in the direction of adopting the
"regular" keywords, and hollowing out and deprecating the headers
<stdalign.h>
and <stdbool.h>
for C2X. See
WG14:N2457
and
WG14:N2458
for more details.
The weak recommendation is to remove these entirely from C++23: these headers
are already owned and specified by at least two other ISO committees, and while
C compatibility is an essential feature for any C++ tool chain sitting on POSIX,
it may be over-specification demanding what should be platform implementation
details for other environments - as noted when C++17 chose to not require the
C <atomic.h>
.
Strong recommendation(A): Remove the vacuous headers, then undeprecate the remaining [depr.c.headers] and move directly into 16.5.5.2 [res.on.headers].
16.5.5.2.1 C standard library headers [c.headers]
- For compatibility with the C standard library, the C++ standard library provides the C headers shown in Table 147.
Table 147 — C headers [tab:depr.c.headers] <assert.h>
<inttypes.h>
<signal.h>
<stdio.h>
<wchar.h>
<complex.h>
<iso646.h>
<stdalign.h><stdlib.h>
<wctype.h>
<ctype.h>
<limits.h>
<stdarg.h>
<string.h>
<errno.h>
<locale.h>
<stdbool.h>
<tgmath.h><fenv.h>
<math.h>
<stddef.h>
<time.h>
<float.h>
<setjmp.h>
<stdint.h>
<uchar.h>
- Every C header, each of which has a name of the form
<name.h>
, behaves as if each name placed in the standard library namespace by the corresponding<cname>
header is placed within the global namespace scope, except for the functions described in 26.8.6, the declarations ofstd::nullptr_t
(17.2.3) andstd::byte
(17.2.5), and the functions and function templates described in 17.2.5. It is unspecified whether these names are first declared or defined within namespace scope (6.4.6) of the namespacestd
and are then injected into the global namespace scope by explicit using-declarations (9.9).- [Example: The header
<cstdlib>
assuredly provides its declarations and definitions within the namespacestd
. It may also provide these names within the global namespace. The header<stdlib.h>
assuredly provides the same declarations and definitions within the global namespace, much as in the C Standard. It may also provide these names within the namespacestd
. —end example]C.6.1 Modifications to headers [diff.mods.to.headers]
- For compatibility with the C standard library, the C++ standard library provides the C headers enumerated in
D.9, but their use is deprecated in C++[c.headers].- There are no C++ headers for the C headers
<complex.h>
,<iso646.h>
,<stdalign.h>
,<stdatomic.h>
,<stdbool.h>
,<stdnoreturn.h>
,<tgmath.h>
, and<threads.h>
, nor are these headers from the C standard library headers themselves part of C++.The C headers<complex.h>
and<tgmath.h>
do not contain any of the content from the C standard library and instead merely include other headers from the C++ standard library.
D.9 C headers [depr.c.headers]
For compatibility with the C standard library, the C++ standard library provides the C headers shown in Table 147.
Table 147 — C headers [tab:depr.c.headers]
<assert.h>
<inttypes.h>
<signal.h>
<stdio.h>
<wchar.h>
<complex.h>
<iso646.h>
<stdalign.h>
<stdlib.h>
<wctype.h>
<ctype.h>
<limits.h>
<stdarg.h>
<string.h>
<errno.h>
<locale.h>
<stdbool.h>
<tgmath.h>
<fenv.h>
<math.h>
<stddef.h>
<time.h>
<float.h>
<setjmp.h>
<stdint.h>
<uchar.h>
D.9.1 Header<complex.h>
synopsis [depr.complex.h.syn]#include<complex>
The header<complex.h>
behaves as if it simply includes the header<complex>
(26.4.1).[Note: Names introduced by<complex>
in namespacestd
are not placed into the global namespace scope by<complex.h>
. —end note]
D.9.2 Header<iso646.h>
synopsis [depr.iso646.h.syn]
The C++ header<iso646.h>
is empty. [Note:and
,and_eq
,bitand
,bitor
,compl
,not_eq
,not
,or
,or_eq
,xor
, andxor_eq
are keywords in this International Standard (5.11). —end note]
D.9.3 Header<stdalign.h>
synopsis [depr.stdalign.h.syn]#define __alignas_is_defined 1
The contents of the C++ header<stdalign.h>
are the same as the C standard library header<stdalign.h>
, with the following changes: The header<stdalign.h>
does not define a macro named alignas.
See also: ISO C 7.15
D.9.4 Header<stdbool.h>
synopsis [depr.stdbool.h.syn]#define __bool_true_false_are_defined 1
The contents of the C++ header<stdbool.h>
are the same as the C standard library header<stdbool.h>
, with the following changes: The header<stdbool.h>
does not define macros namedbool
,true
, orfalse
.
See also: ISO C 7.18
D.9.5 Header<tgmath.h>
synopsis [depr.tgmath.h.syn]#include<cmath>
#include<complex>
The header<tgmath.h>
behaves as if it simply includes the headers<cmath>
(26.8.1) and<complex>
(26.4.1).[Note: The overloads provided in C by type-generic macros are already provided in<complex>
and<cmath>
by “sufficient” additional overloads. —end note][Note: Names introduced by<cmath>
or<complex>
in namespace std are not placed into the global namespace scope by<tgmath.h>
. —end note]
D.9.6 Other C headers [depr.c.headers.other]
Every C header other than<complex.h>
(D.9.1),<iso646.h>
(D.9.2),<stdalign.h>
(D.9.3),<stdbool.h>
(D.9.4), and<tgmath.h>
(D.9.5), each of which has a name of the form<name.h>
, behaves as if each name placed in the standard library namespace by the corresponding<cname>
header is placed within the global namespace scope, except for the functions described in 26.8.6, the declaration ofstd::byte
(17.2.1), and the functions and function templates described in 17.2.5. It is unspecified whether these names are first declared or defined within namespace scope (6.4.6) of the namespacestd
and are then injected into the global namespace scope by explicit using-declarations (9.9).[Example: The header<cstdlib>
assuredly provides its declarations and definitions within the namespacestd
. It may also provide these names within the global namespace. The header<stdlib.h>
assuredly provides the same declarations and definitions within the global namespace, much as in the C Standard. It may also provide these names within the namespacestd
. —end example]
strong recommendation (B): Undeprecate the remaining [depr.c.headers] and move directly into 16.5.5.2 [res.on.headers].
16.5.5.2 Headers [res.on.headers]
- A C++ header may include other C++ headers. A C++ header shall provide the declarations and definitions that appear in its synopsis. A C++ header shown in its synopsis as including other C++ headers shall provide the declarations and definitions that appear in the synopses of those other headers.
- Certain types and macros are defined in more than one header. Every such entity shall be defined such that any header that defines it may be included after any other header that also defines it (6.3).
The C standard library headers (D.9) shall include only their corresponding C++ standard library header, as described in 16.5.1.2.Actually, probably want to merge this into 16.5.1.2 Headers [headers]. Also, fix up cross-references to list stable labels.16.5.5.2.1 C headers [c.headers]
- For compatibility with the C standard library, the C++ standard library provides the C headers shown in Table 147.
Table 147 — C headers [tab:depr.c.headers] <assert.h>
<inttypes.h>
<signal.h>
<stdio.h>
<wchar.h>
<complex.h>
<iso646.h>
<stdalign.h>
<stdlib.h>
<wctype.h>
<ctype.h>
<limits.h>
<stdarg.h>
<string.h>
<errno.h>
<locale.h>
<stdbool.h>
<tgmath.h>
<fenv.h>
<math.h>
<stddef.h>
<time.h>
<float.h>
<setjmp.h>
<stdint.h>
<uchar.h>
16.5.5.2.1.1 Header
<complex.h>
synopsis [complex.h.syn]#include<complex>
- The header
<complex.h>
behaves as if it simply includes the header<complex>
(26.4.1).- [Note: Names introduced by
<complex>
in namespacestd
are not placed into the global namespace scope by<complex.h>
. —end note]16.5.5.2.1.2 Header
<iso646.h>
synopsis [iso646.h.syn]
- The C++ header
<iso646.h>
is empty. [Note:and
,and_eq
,bitand
,bitor
,compl
,not_eq
,not
,or
,or_eq
,xor
, andxor_eq
are keywords in this International Standard (5.11). —end note]16.5.5.2.1.3 Header
<stdalign.h>
synopsis [stdalign.h.syn]#define __alignas_is_defined 1
- The contents of the C++ header
<stdalign.h>
are the same as the C standard library header<stdalign.h>
, with the following changes: The header<stdalign.h>
does not define a macro named alignas.See also: ISO C 7.15
16.5.5.2.1.4 Header
<stdbool.h>
synopsis [stdbool.h.syn]#define __bool_true_false_are_defined 1
- The contents of the C++ header
<stdbool.h>
are the same as the C standard library header<stdbool.h>
, with the following changes: The header<stdbool.h>
does not define macros namedbool
,true
, orfalse
.See also: ISO C 7.18
16.5.5.2.1.5 Header
<tgmath.h>
synopsis [tgmath.h.syn]#include<cmath>
#include<complex>
- The header
<tgmath.h>
behaves as if it simply includes the headers<cmath>
(26.8.1) and<complex>
(26.4.1).- [Note: The overloads provided in C by type-generic macros are already provided in
<complex>
and<cmath>
by "sufficient" additional overloads. —end note]- [Note: Names introduced by
<cmath>
or<complex>
in namespace std are not placed into the global namespace scope by<tgmath.h>
. —end note]16.5.5.2.1.6 Other C headers [c.headers.other]
- Every C header other than
<complex.h>
(D.9.1),<iso646.h>
(D.9.2),<stdalign.h>
(D.9.3),<stdbool.h>
(D.9.4), and<tgmath.h>
(D.9.5), each of which has a name of the form<name.h>
, behaves as if each name placed in the standard library namespace by the corresponding<cname>
header is placed within the global namespace scope, except for the functions described in 26.8.6, the declaration ofstd::byte
(17.2.1), and the functions and function templates described in 17.2.5. It is unspecified whether these names are first declared or defined within namespace scope (6.4.6) of the namespacestd
and are then injected into the global namespace scope by explicit using-declarations (9.9).- [Example: The header
<cstdlib>
assuredly provides its declarations and definitions within the namespacestd
. It may also provide these names within the global namespace. The header<stdlib.h>
assuredly provides the same declarations and definitions within the global namespace, much as in the C Standard. It may also provide these names within the namespacestd
. —end example]
D.9 C headers [depr.c.headers]
For compatibility with the C standard library, the C++ standard library provides the C headers shown in Table 147.
Table 147 — C headers [tab:depr.c.headers]
<assert.h>
<inttypes.h>
<signal.h>
<stdio.h>
<wchar.h>
<complex.h>
<iso646.h>
<stdalign.h>
<stdlib.h>
<wctype.h>
<ctype.h>
<limits.h>
<stdarg.h>
<string.h>
<errno.h>
<locale.h>
<stdbool.h>
<tgmath.h>
<fenv.h>
<math.h>
<stddef.h>
<time.h>
<float.h>
<setjmp.h>
<stdint.h>
<uchar.h>
D.9.1 Header<complex.h>
synopsis [depr.complex.h.syn]#include<complex>
The header<complex.h>
behaves as if it simply includes the header<complex>
(26.4.1).[Note: Names introduced by<complex>
in namespacestd
are not placed into the global namespace scope by<complex.h>
. —end note]
D.9.2 Header<iso646.h>
synopsis [depr.iso646.h.syn]
The C++ header<iso646.h>
is empty. [Note:and
,and_eq
,bitand
,bitor
,compl
,not_eq
,not
,or
,or_eq
,xor
, andxor_eq
are keywords in this International Standard (5.11). —end note]
D.9.3 Header<stdalign.h>
synopsis [depr.stdalign.h.syn]#define __alignas_is_defined 1
The contents of the C++ header<stdalign.h>
are the same as the C standard library header<stdalign.h>
, with the following changes: The header<stdalign.h>
does not define a macro named alignas.
See also: ISO C 7.15
D.9.4 Header<stdbool.h>
synopsis [depr.stdbool.h.syn]#define __bool_true_false_are_defined 1
The contents of the C++ header<stdbool.h>
are the same as the C standard library header<stdbool.h>
, with the following changes: The header<stdbool.h>
does not define macros namedbool
,true
, orfalse
.
See also: ISO C 7.18
D.9.5 Header<tgmath.h>
synopsis [depr.tgmath.h.syn]#include<cmath>
#include<complex>
The header<tgmath.h>
behaves as if it simply includes the headers<cmath>
(26.8.1) and<complex>
(26.4.1).[Note: The overloads provided in C by type-generic macros are already provided in<complex>
and<cmath>
by "sufficient" additional overloads. —end note][Note: Names introduced by<cmath>
or<complex>
in namespace std are not placed into the global namespace scope by<tgmath.h>
. —end note]
D.9.6 Other C headers [depr.c.headers.other]
Every C header other than<complex.h>
(D.9.1),<iso646.h>
(D.9.2),<stdalign.h>
(D.9.3),<stdbool.h>
(D.9.4), and<tgmath.h>
(D.9.5), each of which has a name of the form<name.h>
, behaves as if each name placed in the standard library namespace by the corresponding<cname>
header is placed within the global namespace scope, except for the functions described in 26.8.6, the declaration ofstd::byte
(17.2.1), and the functions and function templates described in 17.2.5. It is unspecified whether these names are first declared or defined within namespace scope (6.4.6) of the namespacestd
and are then injected into the global namespace scope by explicit using-declarations (9.9).[Example: The header<cstdlib>
assuredly provides its declarations and definitions within the namespacestd
. It may also provide these names within the global namespace. The header<stdlib.h>
assuredly provides the same declarations and definitions within the global namespace, much as in the C Standard. It may also provide these names within the namespacestd
. —end example]
Weak recommendation: Remove entirely from C++23.
C.6.1 Modifications to headers [diff.mods.to.headers]
- For compatibility with the C standard library, the C++ standard library provides the C headers enumerated in
D.9, but their use is deprecated in C++[c.headers].There are no C++ headers for the C headers<stdatomic.h>
,<stdnoreturn.h>
, and<threads.h>
, nor are the C headers themselves part of C++.The C headers<complex.h>
and<tgmath.h>
do not contain any of the content from the C standard library and instead merely include other headers from the C++ standard library.Revise compatibility note for Annex C. This wording needs a little more thought.
D.9 C headers [depr.c.headers]
For compatibility with the C standard library, the C++ standard library provides the C headers shown in Table 147.
Table 147 — C headers [tab:depr.c.headers]
<assert.h>
<inttypes.h>
<signal.h>
<stdio.h>
<wchar.h>
<complex.h>
<iso646.h>
<stdalign.h>
<stdlib.h>
<wctype.h>
<ctype.h>
<limits.h>
<stdarg.h>
<string.h>
<errno.h>
<locale.h>
<stdbool.h>
<tgmath.h>
<fenv.h>
<math.h>
<stddef.h>
<time.h>
<float.h>
<setjmp.h>
<stdint.h>
<uchar.h>
D.9.1 Header<complex.h>
synopsis [depr.complex.h.syn]#include<complex>
The header<complex.h>
behaves as if it simply includes the header<complex>
(26.4.1).[Note: Names introduced by<complex>
in namespacestd
are not placed into the global namespace scope by<complex.h>
. —end note]
D.9.2 Header<iso646.h>
synopsis [depr.iso646.h.syn]
The C++ header<iso646.h>
is empty. [Note:and
,and_eq
,bitand
,bitor
,compl
,not_eq
,not
,or
,or_eq
,xor
, andxor_eq
are keywords in this International Standard (5.11). —end note]
D.9.3 Header<stdalign.h>
synopsis [depr.stdalign.h.syn]#define __alignas_is_defined 1
The contents of the C++ header<stdalign.h>
are the same as the C standard library header<stdalign.h>
, with the following changes: The header<stdalign.h>
does not define a macro named alignas.
See also: ISO C 7.15
D.9.4 Header<stdbool.h>
synopsis [depr.stdbool.h.syn]#define __bool_true_false_are_defined 1
The contents of the C++ header<stdbool.h>
are the same as the C standard library header<stdbool.h>
, with the following changes: The header<stdbool.h>
does not define macros namedbool
,true
, orfalse
.
See also: ISO C 7.18
D.9.5 Header<tgmath.h>
synopsis [depr.tgmath.h.syn]#include<cmath>
#include<complex>
The header<tgmath.h>
behaves as if it simply includes the headers<cmath>
(26.8.1) and<complex>
(26.4.1).[Note: The overloads provided in C by type-generic macros are already provided in<complex>
and<cmath>
by "sufficient" additional overloads. —end note][Note: Names introduced by<cmath>
or<complex>
in namespace std are not placed into the global namespace scope by<tgmath.h>
. —end note]
D.9.6 Other C headers [depr.c.headers.other]
Every C header other than<complex.h>
(D.9.1),<iso646.h>
(D.9.2),<stdalign.h>
(D.9.3),<stdbool.h>
(D.9.4), and<tgmath.h>
(D.9.5), each of which has a name of the form<name.h>
, behaves as if each name placed in the standard library namespace by the corresponding<cname>
header is placed within the global namespace scope, except for the functions described in 26.8.6, the declaration ofstd::byte
(17.2.1), and the functions and function templates described in 17.2.5. It is unspecified whether these names are first declared or defined within namespace scope (6.4.6) of the namespacestd
and are then injected into the global namespace scope by explicit using-declarations (9.9).[Example: The header<cstdlib>
assuredly provides its declarations and definitions within the namespacestd
. It may also provide these names within the global namespace. The header<stdlib.h>
assuredly provides the same declarations and definitions within the global namespace, much as in the C Standard. It may also provide these names within the namespacestd
. —end example]
There was general agreement that deprecated does not adequately capture the intent here. The need for C compatibility means these headers are not going away, yet some have strong concerns about undeprecating these headers implying that these are perfectly good for C++ use. Those concerns are not reduced by folks like the author believing that is indeed the case. There was also the observation that such arguments about best practice are more suited to education material than the standard proper. One final observation is that perhaps these headers could move into Annex C, although that is a non-normative annex, so the status of such a move might be unclear.
Consensus on the call was that this is a big enough topic to merit a new paper in its own right, and no action should be taken as part of this omnibus proposal.
LEWGI Review: Take not action now, and follow up with a specific paper.
First deprecated: C++20
This style of documentation was deprecated editorially following the application of a sequence of papers to update each main library clause, consistently following the new conventions established by paper P0788R3
Note that no review of this section will be needed if the options to remove all of D.11, D.13, D.18, D.21, and D.23 are accepted, as there would be nothing left to review.
As library clauses continue to be deprecated and migrate to Annex D, use of this deprecated convention will become internally less consistent, so the strong recommendation is to apply the new tags consistently through the parts of Annex D that are retained. The weak recommendation is to let this term remain until the last clause using it has been removed (or undeprecated and updated) in a future standard.
Strong recommendation: Remove this feature from C++23.
D.10 Requires paragraph [depr.res.on.required]
In addition to the elements specified in 16.4.1.4, descriptions of function semantics may also contain a Requires: element to denote the preconditions for calling a function.Violation of any preconditions specified in a function's Requires: element results in undefined behavior unless the function's Throws: element specifies throwing an exception when the precondition is violated.
D.11 Relational operators [depr.relops]
The Cpp17OldConcept requirements contain a semantic component, so the most appropriate replacement for a Requires clause is a Preconditions clause.
- The header
<utility>
(20.2.1) has the following additions:namespace std::rel_ops { template<class T> bool operator!=(const T&, const T&); template<class T> bool operator> (const T&, const T&); template<class T> bool operator<=(const T&, const T&); template<class T> bool operator>=(const T&, const T&); }- To avoid redundant definitions of
operator!=
out ofoperator==
and operators>
,<=
, and>=
out ofoperator<
, the library provides the following:template<class T> bool operator!=(const T& x, const T& y);Requires: TypePreconditions:T
is Cpp17EqualityComparable (Table 25).T
meets the Cpp17EqualityComparable requirements (Table 25).- Returns:
!(x == y)
.template<class T> bool operator>(const T& x, const T& y);Requires: TypePreconditions:T
is Cpp17LessThanComparable (Table 26).T
meets the Cpp17LessThanComparable requirements (Table 26).- Returns:
y < x
.template<class T> bool operator<=(const T& x, const T& y);Requires: TypePreconditions:T
is Cpp17LessThanComparable (Table 26).T
meets the Cpp17LessThanComparable requirements (Table 26).- Returns:
!(y < x)
.template<class T> bool operator>=(const T& x, const T& y);Requires: TypePreconditions:T
is Cpp17LessThanComparable (Table 26).T
meets the Cpp17LessThanComparable requirements (Table 26).- Returns:
!(x < y)
.D.13 Deprecated type traits [depr.meta.types]
- The header
<type_traits>
(20.15.2) has the following additions:namespace std { template<class T> struct is_pod; template<class T> inline constexpr bool is_pod_v = is_pod<T>::value; }- The behavior of a program that adds specializations for any of the templates defined in this subclause is undefined, unless explicitly permitted by the specification of the corresponding template.
template<class T> struct is_pod;RequiresMandates:remove_all_extents_t<T>
shall be a complete type or cvvoid
.- Remarks:
is_pod<T>
is a Cpp17UnaryTypeTrait (20.15.1) with a base characteristic oftrue_type
ifT
is a POD type, andfalse_type
otherwise. A POD class is a class that is both a trivial class and a standard-layout class, and has no non-static data members of type non-POD class (or array thereof). A POD type is a scalar type, a POD class, an array of such a type, or a cv-qualified version of one of these types.- [Note: It is unspecified whether a closure type (7.5.5.1) is a POD type. —end note]
D.18 Deprecated
shared_ptr
atomic access [depr.util.smartptr.shared.atomic]
- The header
<memory>
(20.10.2) has the following additions:namespace std { template <class T> bool atomic_is_lock_free(const shared_ptr<T>* p); template <class T> shared_ptr<T> atomic_load(const shared_ptr<T>* p); template <class T> shared_ptr<T> atomic_load_explicit(const shared_ptr<T>* p, memory_order mo); template <class T> void atomic_store(shared_ptr<T>* p, shared_ptr<T> r); template <class T> void atomic_store_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo); template <class T> shared_ptr<T> atomic_exchange(shared_ptr<T>* p, shared_ptr<T> r); template <class T> shared_ptr<T> atomic_exchange_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo); template <class T> bool atomic_compare_exchange_weak( shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w); template <class T> bool atomic_compare_exchange_strong( shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w); template <class T> bool atomic_compare_exchange_weak_explicit( shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w, memory_order success, memory_order failure); template <class T> bool atomic_compare_exchange_strong_explicit( shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w, memory_order success, memory_order failure); }- Concurrent access to a
shared_ptr
object from multiple threads does not introduce a data race if the access is done exclusively via the functions in this section and the instance is passed as their first argument.- The meaning of the arguments of type
memory_order
is explained in 31.4.template<class T> bool atomic_is_lock_free(const shared_ptr<T>* p);RequiresPreconditions:p
shall not beis not null.- Returns:
true
if atomic access to*p
is lock-free,false
otherwise.- Throws: Nothing.
template<class T> shared_ptr<T> atomic_load(const shared_ptr<T>* p);RequiresPreconditions:p
shall not beis not null.- Returns:
atomic_load_explicit(p, memory_order::seq_cst)
.- Throws: Nothing.
template<class T> shared_ptr<T> atomic_load_explicit(const shared_ptr<T>* p, memory_order mo);RequiresPreconditions:p
shall not beis not null.RequiresPreconditions:mo
shall not beis neithermemory_order::release
normemory_order::acq_rel
.- Returns:
*p
.- Throws: Nothing.
template<class T> void atomic_store(shared_ptr<T>* p, shared_ptr<T> r);RequiresPreconditions:p
shall not beis not null.- Effects: As if by
atomic_store_explicit(p, r, memory_order::seq_cst)
.- Throws: Nothing.
template<class T> void atomic_store_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo);RequiresPreconditions:p
shall not beis not null.- Requires:
mo
shall not bememory_order::acquire
ormemory_order::acq_rel
.- Effects: As if by
p->swap(r)
.- Throws: Nothing.
template<class T> shared_ptr<T> atomic_exchange(shared_ptr<T>* p, shared_ptr<T> r);RequiresPreconditions:p
shall not beis not null.- Returns:
atomic_exchange_explicit(p, r, memory_order::seq_cst)
.- Throws: Nothing.
template<class T> shared_ptr<T> atomic_exchange_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo);RequiresPreconditions:p
shall not beis not null.- Effects: As if by
p->swap(r)
.- Returns: The previous value of
*p
.- Throws: Nothing.
template<class T> bool atomic_compare_exchange_weak(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w);RequiresPreconditions:p
shall not beis not null.- Returns:
atomic_compare_exchange_weak_explicit(p, v, w, memory_order::seq_cst, memory_order::seq_cst)
.- Throws: Nothing.
template<class T> bool atomic_compare_exchange_strong(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w);- Returns:
atomic_compare_exchange_strong_explicit(p, v, w, memory_order::seq_cst, memory_order::seq_cst)
.template <class T> bool atomic_compare_exchange_weak_explicit( shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w, memory_order success, memory_order failure); template <class T> bool atomic_compare_exchange_strong_explicit( shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w, memory_order success, memory_order failure);RequiresPreconditions:p
shall not beis not andv
shall not beis not null. Thefailure
argumentshall not beis neithermemory_order::release
normemory_order::acq_rel
.- Effects: If
*p
is equivalent to*v
, assignsw
to*p
and has synchronization semantics corresponding to the value ofsuccess
, otherwise assigns*p
to*v
and has synchronization semantics corresponding to the value offailure
.- Returns:
true
if*p
was equivalent to*v
,false
otherwise.- Throws: Nothing.
- Remarks: Two
shared_ptr
objects are equivalent if they store the same pointer value and share ownership. The weak form may fail spuriously. See 31.8.1.D.21 Deprecated convenience conversion interfaces [depr.conversions]
D.21.1 Class template wstring_convert [depr.conversions.string]
- Class template
wstring_convert
performs conversions between a wide string and a byte string. It lets you specify a code conversion facet (like class templatecodecvt
) to perform the conversions, without affecting any streams or locales. [ Example: If you want to use the code conversion facetcodecvt_utf8
to output tocout
a UTF-8 multibyte sequence corresponding to a wide string, but you don’t want to alter the locale forcout
, you can write something like:— end example ]std::wstring_convert<std::codecvt_utf8<wchar_t>> myconv; std::string mbstring = myconv.to_bytes(L"Hello\n"); std::cout << mbstring;namespace std { template <class Codecvt, class Elem = wchar_t, class WideAlloc = allocator<Elem>, class ByteAlloc = allocator<char>> class wstring_convert { public: using byte_string = basic_string<char, char_traits<char>, ByteAlloc>; using wide_string = basic_string<Elem, char_traits<Elem>, WideAlloc>; using state_type = typename Codecvt::state_type; using int_type = typename wide_string::traits_type::int_type; wstring_convert() : wstring_convert(new Codecvt) {} explicit wstring_convert(Codecvt* pcvt = new Codecvt); wstring_convert(Codecvt* pcvt, state_type state); explicit wstring_convert(const byte_string& byte_err, const wide_string& wide_err = wide_string()); ~wstring_convert(); wstring_convert(const wstring_convert&) = delete; wstring_convert& operator=(const wstring_convert&) = delete; wide_string from_bytes(char byte); wide_string from_bytes(const char* ptr); wide_string from_bytes(const byte_string& str); wide_string from_bytes(const char* first, const char* last); byte_string to_bytes(Elem wchar); byte_string to_bytes(const Elem* wptr); byte_string to_bytes(const wide_string& wstr); byte_string to_bytes(const Elem* first, const Elem* last); size_t converted() const noexcept; state_type state() const; private: byte_string byte_err_string; // Exposition only wide_string wide_err_string; // Exposition only Codecvt* cvtptr; // Exposition only state_type cvtstate; // Exposition only size_t cvtcount; // Exposition only }; }- The class template describes an object that controls conversions between wide string objects of class
basic_string<Elem, char_traits<Elem>, WideAlloc>
and byte string objects of classbasic_string<char, char_traits<char>, ByteAlloc>
. The class template defines the typeswide_string
andbyte_string
as synonyms for these two types. Conversion between a sequence ofElem
values (stored in awide_string
object) and multibyte sequences (stored in abyte_string
object) is performed by an object of classCodecvt
, which meets the requirements of the standard code-conversion facetcodecvt<Elem, char, mbstate_t>
.- An object of this class template stores:
- —
byte_err_string
— a byte string to display on errors- —
wide_err_string
— a wide string to display on errors- —
cvtptr
— a pointer to the allocated conversion object (which is freed when thewstring_convert
object is destroyed)- —
cvtstate
— a conversion state object- —
cvtcount
— a conversion countusing byte_string = basic_string<char, char_traits<char>, ByteAlloc>;- The type shall be a synonym for
basic_string<char, char_traits<char>, ByteAlloc>
size_t converted() const noexcept;- Returns:
cvtcount
.wide_string from_bytes(char byte); wide_string from_bytes(const char* ptr); wide_string from_bytes(const byte_string& str); wide_string from_bytes(const char* first, const char* last);- Effects: The first member function shall convert the single-element sequence
byte
to a wide string. The second member function shall convert the null-terminated sequence beginning atptr
to a wide string. The third member function shall convert the sequence stored in str to a wide string. The fourth member function shall convert the sequence defined by the range[first, last)
to a wide string.- In all cases:
- — If the
cvtstate
object was not constructed with an explicit value, it shall be set to its default value (the initial conversion state) before the conversion begins. Otherwise it shall be left unchanged.- — The number of input elements successfully converted shall be stored in
cvtcount
.- Returns: If no conversion error occurs, the member function shall return the converted wide string. Otherwise, if the object was constructed with a wide-error string, the member function shall return the wide-error string. Otherwise, the member function throws an object of class
range_error
.using int_type = typename wide_string::traits_type::int_type;- The type shall be a synonym for
wide_string::traits_type::int_type
.state_type state() const;- returns
cvtstate
.using state_type = typename Codecvt::state_type;- The type shall be a synonym for
Codecvt::state_type
.byte_string to_bytes(Elem wchar); byte_string to_bytes(const Elem* wptr); byte_string to_bytes(const wide_string& wstr); byte_string to_bytes(const Elem* first, const Elem* last);- Effects: The first member function shall convert the single-element sequence
wchar
to a byte string. The second member function shall convert the null-terminated sequence beginning atwptr
to a byte string. The third member function shall convert the sequence stored in wstr to a byte string. The fourth member function shall convert the sequence defined by the range[first, last)
to a byte string.- In all cases:
- — If the
cvtstate
object was not constructed with an explicit value, it shall be set to its default value (the initial conversion state) before the conversion begins. Otherwise it shall be left unchanged.- — The number of input elements successfully converted shall be stored in
cvtcount
.- Returns: If no conversion error occurs, the member function shall return the converted byte string. Otherwise, if the object was constructed with a byte-error string, the member function shall return the byte-error string. Otherwise, the member function shall throw an object of class
range_error
.using wide_string = basic_string<Elem, char_traits<Elem>, WideAlloc>;- The type shall be a synonym for
basic_string<Elem, char_traits<Elem>, WideAlloc>
.explicit wstring_convert(Codecvt* pcvt = new Codecvt); wstring_convert(Codecvt* pcvt, state_type state); explicit wstring_convert(const byte_string& byte_err, const wide_string& wide_err = wide_string());RequiresPreconditions: For the first and second constructors,pcvt != nullptr
.- Effects: The first constructor shall store
pcvt
incvtptr
and default values incvtstate
,byte_err_string
, andwide_err_string
. The second constructor shall storepcvt
incvtptr
,state
incvtstate
, and default values inbyte_err_string
andwide_err_string
; moreover the stored state shall be retained between calls tofrom_bytes
andto_bytes
. The third constructor shall storenew Codecvt
incvtptr
,state_type()
incvtstate
,byte_err
inbyte_err_string
, andwide_err
inwide_err_string
.~wstring_convert();- Effects: The destructor shall delete
cvtptr
.D.21.2 Class template wbuffer_convert [depr.conversions.buffer]
- Class template
wbuffer_convert
looks like a wide stream buffer, but performs all its I/O through an underlying byte stream buffer that you specify when you construct it. Like class templatewstring_convert
, it lets you specify a code conversion facet to perform the conversions, without affecting any streams or locales.namespace std { template <class Codecvt, class Elem = wchar_t, class Tr = char_traits<Elem>> class wbuffer_convert : public basic_streambuf<Elem, Tr> { public: using state_type = typename Codecvt::state_type; wbuffer_convert() : wbuffer_convert(nullptr) {} explicit wbuffer_convert(streambuf* bytebuf = 0, Codecvt* pcvt = new Codecvt, state_type state = state_type()); ~wbuffer_convert(); wbuffer_convert(const wbuffer_convert&) = delete; wbuffer_convert& operator=(const wbuffer_convert&) = delete; streambuf* rdbuf() const; streambuf* rdbuf(streambuf* bytebuf); state_type state() const; private: streambuf* bufptr; // exposition only Codecvt* cvtptr; // exposition only state_type cvtstate; // exposition only }; }- The class template describes a stream buffer that controls the transmission of elements of type
Elem
, whose character traits are described by the classTr
, to and from a byte stream buffer of typestreambuf
. Conversion between a sequence ofElem
values and multibyte sequences is performed by an object of classCodecvt
, which shall meet the requirements of the standard code-conversion facetcodecvt<Elem, char, mbstate_t>
.- An object of this class template stores:
- —
bufptr
— a pointer to its underlying byte stream buffer- —
cvtptr
— a pointer to the allocated conversion object (which is freed when thewbuffer_convert
object is destroyed)- —
cvtstate
— a conversion state objectstate_type state() const;- Returns:
cvtstate
.streambuf* rdbuf() const;- Returns:
bufptr
.streambuf* rdbuf(streambuf* bytebuf);- Effects: Stores
bytebuf
inbufptr
.- Returns: The previous value of
bufptr
.using state_type = typename Codecvt::state_type;- The type shall be a synonym for
Codecvt::state_type
.explicit wbuffer_convert(streambuf* bytebuf = 0, Codecvt* pcvt = new Codecvt, state_type state = state_type());RequiresPreconditions:pcvt != nullptr
.- Effects: The constructor constructs a stream buffer object, initializes
bufptr
tobytebuf
, initializescvtptr
topcvt
, and initializescvtstate
tostate
.~wbuffer_convert();- Effects: The destructor shall delete
cvtptr
.
D.23 Deprecated filesystem path factory functions [depr.fs.path.factory]
The Requires clause here covers both compile-time and run-time conditions, so is split into separate Mandates and Preconditions clauses. A drive-by fix further catches that several typedefs intending to refer to members ofpath
are actually meaningless for a non-member factory function. The current form of wording was copy/pasted from the specification of constructors where thepath::
would be implicit.
- The header
<filesystem>
(29.11.5) has the following additions:template<class Source> path u8path(const Source& source); template<class InputIterator> path u8path(InputIterator first, InputIterator last);Requires: Thesource
and[first, last)
sequences are UTF-8 encoded. The value type ofSource
andInputIterator
ischar
orchar8_t
.Source
meets the requirements specified in 29.11.7.3.- Mandates: The value type of
Source
andInputIterator
ischar
orchar8_t
.- Preconditions: The
source
and[first, last)
sequences are UTF-8 encoded.Source
meets the requirements specified in 29.11.7.3.- Returns:
- — If
path::value_type
ischar
and the current native narrow encoding (29.11.7.2.2) is UTF-8, returnpath(source)
orpath(first, last)
; otherwise,- — if
path::value_type
iswchar_t
and the native wide encoding is UTF-16, or ifpath::value_type
ischar16_t
orchar32_t
, convertsource
or[first,last)
to a temporary,tmp
, of typepath::string_type
and returnpath(tmp)
; otherwise,- — convert
source
or[first, last)
to a temporary,tmp
, of typeu32string
and returnpath(tmp)
.- Remarks: Argument format conversion (29.11.7.2.1) applies to the arguments for these functions. How Unicode encoding conversions are performed is unspecified.
- [Example: A string is to be read from a database that is encoded in UTF-8, and used to create a directory using the native encoding for filenames:
For POSIX-based operating systems with the native narrow encoding set to UTF-8, no encoding or type conversion occurs.namespace fs = std::filesystem; std::string utf8_string = read_utf8_data(); fs::create_directory(fs::u8path(utf8_string));For POSIX-based operating systems with the native narrow encoding not set to UTF-8, a conversion to UTF-32 occurs, followed by a conversion to the current native narrow encoding. Some Unicode characters may have no native character set representation.
For Windows-based operating systems a conversion from UTF-8 to UTF-16 occurs. —end example] [Note: The example above is representative of a historical use offilesystem::u8path
. Passing astd::u8string
topath
’s constructor is preferred for an indication of UTF-8 encoding more consistent withpath
’s handling of other encodings. —end note]
Weak recommendation: take no action.
No change to draft.
LEWGI Review: Defer to LWG
First deprecated: C++20
This feature was deprecated with the introduction of support for the 3-way comparison "spaceship" operator, by paper P0768R1.
The std::rel_ops
namespace was introduced in the original C++
standard, so its removal would potentially impact on code written against C++98
and later standards. However, this paper will still recommend strongly for its
removal from the C++23 standard. The feature relies on the antipattern of
using
a namespace in a namespace scope in order to provide
comparison operators for my own type, typically in a header. Such usage is not
recommended as the using
will have an impact on code in other
headers, and potentially introduce header order dependencies. A more targeted
use within a function may serve as an adaption for 3rd party libraries, but
again, by providing these operators for all types whether desired or not. As
there are no viable base classes in this namespace, it was never possible to
arrange for these operators to be discovered by ADL, so there was no better
workaround to tame these functions.
The better solution appeared in C++20 with the introduction of the spaceship
operator, and rules for synthesizing comparisons from a few primitive operators
now built into the language. The typical remedy for much code will be to
simply remove the offending using namespace std::rel_ops;
from
their code, and it should continue to work. As added insurance, the name
rel_ops
will be added to the list of zombie names reserved for
previous standardization so that standard library vendors can continue to offer
these as a compatibility shim for their customers for as long as they wish, so
the impact on the user community of removing the specification of this feature
from the standard is expected to be minimal.
The weak recommendation is that this feature was deprecated just one standard prior, so the user community may take a while to catch up and remove the dependencies from their code. Holding the feature in Annex D does little harm, but the wording should be updated to the latest library documentation style for C++23. We do not see any reason to consider undeprecation of this feature.
As a historical note, the standard library specification itself used to rely on
std::rel_ops
to provide the specification for any comparison
operator if the necessary wording were missing. However, as part of C++20, it
was confirmed that no current wording relies on that legacy fall-back, and the
corresponding as-if wording has already been removed.
Strong recommendation: Remove this feature from C++23.
16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
rel_ops
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.
D.11 Relational operators [depr.relops]
The header<utility>
(20.2.1) has the following additions:>namespace std::rel_ops {template<class T> bool operator!=(const T&, const T&);template<class T> bool operator> (const T&, const T&);template<class T> bool operator<=(const T&, const T&);template<class T> bool operator>=(const T&, const T&);>}To avoid redundant definitions ofoperator!=
out ofoperator==
and operators>
,<=
, and>=
out ofoperator<
, the library provides the following:template<class T> bool operator!=(const T& x, const T& y);Requires: TypeT
is Cpp17EqualityComparable (Table 23).Returns:!(x == y)
.template<class T> bool operator>(const T& x, const T& y);Requires: TypeT
is Cpp17LessThanComparable (Table 24).Returns:y < x
.template<class T> bool operator<=(const T& x, const T& y);Requires: TypeT
is Cpp17LessThanComparable (Table 24).Returns:!(y < x)
.template<class T> bool operator>=(const T& x, const T& y);Requires: TypeT
is Cpp17LessThanComparable (Table 24).Returns:!(x < y)
.
Draft compatibility note for Annex C.
Weak recommendation: Clean up wording.
Update wording to replace Requires clauses (see D.10 Requires paragraph).
Concerns were raised about the lack of research into how much code is likely to break with the removal of this API. We would like to see more numbers and analysis on publicly available code, such as across all of Github. Given that implementers are likely to provide an implementation (through zombie names freedom) for some time after removal, there was consensus to proceed with removal, assuming the requested research does not reveal major concerns before the main LEWG review to follow.
LEWGI Review: Remove this feature from C++23.
char*
streams [depr.str.strstreams]First deprecated: C++98
The char*
streams were provided, pre-deprecated, in C++98 and have
been considered for removal before. The underlying principle of not removing
them until a suitable replacement is available still holds, so there should be
nothing further to discuss at this point.
However, it should be noted that the spanstream
library
(P0448R2)
passed LEWG review for C++20, but ran out of LWG working time to integrate into
the final standard. This library is expected to be the replacement facility we
point users to in the future, so it may be reasonable to again consider the
classic strstreams facility for removal in C++26.
There remains the alternative position that this facility has been a supported shipping part of the C++ standard for around 25 years when C++23 ships. If we have not made serious moves to remove the library in all that time, maybe we should consider undeprecating, and taking away the shadow of doubt over any code that reaches for this facility today.
Strong recommendation: take no action.
No change to draft.
Weak recommendation:
Undeprecate the char*
streams.
Text moved and renumbered without any review to modernising the wording. Do observe that the paragraph types are consistent with current library practice though, so we do not need a "mandates review".
29.11
char*
streams [str.strstreams]29.11.1 Header
<strstream>
synopsis [strstream.syn]
- The header
<strstream>
defines types that associate stream buffers with character array objects and assist reading and writing such objects.Could use a better description indicating how these stream classes are distinct from thestringstream
classes.namespace std { class strstreambuf; class istrstream; class ostrstream; class strstream; }29.11.2 Class
strstreambuf
[strstreambuf]namespace std { class strstreambuf : public basic_streambuf<char> { public: strstreambuf() : strstreambuf(0) {} explicit strstreambuf(streamsize alsize_arg); strstreambuf(void* (*palloc_arg)(size_t), void (*pfree_arg)(void*)); strstreambuf(char* gnext_arg, streamsize n, char* pbeg_arg = nullptr); strstreambuf(const char* gnext_arg, streamsize n); strstreambuf(signed char* gnext_arg, streamsize n, signed char* pbeg_arg = nullptr); strstreambuf(const signed char* gnext_arg, streamsize n); strstreambuf(unsigned char* gnext_arg, streamsize n, unsigned char* pbeg_arg = nullptr); strstreambuf(const unsigned char* gnext_arg, streamsize n); virtual ~strstreambuf(); void freeze(bool freezefl = true); char* str(); int pcount(); protected: int_type overflow (int_type c = EOF) override; int_type pbackfail(int_type c = EOF) override; int_type underflow() override; pos_type seekoff(off_type off, ios_base::seekdir way, ios_base::openmode which = ios_base::in | ios_base::out) override; pos_type seekpos(pos_type sp, ios_base::openmode which = ios_base::in | ios_base::out) override; streambuf* setbuf(char* s, streamsize n) override; private: using strstate = T1; // exposition only static const strstate allocated; // exposition only static const strstate constant; // exposition only static const strstate dynamic; // exposition only static const strstate frozen; // exposition only strstate strmode; // exposition only streamsize alsize; // exposition only void* (*palloc)(size_t); // exposition only void (*pfree)(void*); // exposition only }; }
- The class
strstreambuf
associates the input sequence, and possibly the output sequence, with an object of some character array type, whose elements store arbitrary values. The array object has several attributes.- [Note: For the sake of exposition, these are represented as elements of a bitmask type (indicated here as
T1
) calledstrstate
. The elements are:—end note]
- —
allocated
, set when a dynamic array object has been allocated, and hence should be freed by the destructor for thestrstreambuf
object;- —
constant
, set when the array object hasconst
elements, so the output sequence cannot be written;- —
dynamic
, set when the array object is allocated (or reallocated) as necessary to hold a character sequence that can change in length;- —
frozen
, set when the program has requested that the array object not be altered, reallocated, or freed.- [Note: For the sake of exposition, the maintained data is presented here as:
—end note]
- —
strstate strmode
, the attributes of the array object associated with thestrstreambuf
object;- —
int alsize
, the suggested minimum size for a dynamic array object;- —
void* (*palloc)(size_t)
, points to the function to call to allocate a dynamic array object;- —
void (*pfree)(void*)
, points to the function to call to free a dynamic array object.- Each object of class
strstreambuf
has a seekable area, delimited by the pointersseeklow
andseekhigh
. Ifgnext
is a null pointer, the seekable area is undefined. Otherwise,seeklow
equalsgbeg
andseekhigh
is eitherpend
, ifpend
is not a null pointer, orgend
.29.11.2.1
strstreambuf
constructors [strstreambuf.cons]
explicit strstreambuf(streamsize alsize_arg);- Effects: Initializes the base class with
streambuf()
. The postconditions of this function are indicated in Table 148.
Table 148: strstreambuf(streamsize)
effects [tab:depr.strstreambuf.cons.sz]Element Value strmode
dynamic
alsize
alsize_arg
palloc
a null pointer pfree
a null pointer
strstreambuf(void* (*palloc_arg)(size_t), void (*pfree_arg)(void*));- Effects: Initializes the base class with
streambuf()
. The postconditions of this function are indicated in Table 149.
Table 149: strstreambuf(void* (*)(size_t), void (*)(void*))
effects [tab:depr.strstreambuf.cons.alloc]Element Value strmode
dynamic
alsize
an unspecified value palloc
palloc_arg
pfree
pfree_arg
strstreambuf(char* gnext_arg, streamsize n, char* pbeg_arg = nullptr); strstreambuf(signed char* gnext_arg, streamsize n, signed char* pbeg_arg = nullptr); strstreambuf(unsigned char* gnext_arg, streamsize n, unsigned char* pbeg_arg = nullptr);- Effects: Initializes the base class with
streambuf()
. The postconditions of this function are indicated in Table 150.
Table 150: strstreambuf(charT*, streamsize, charT*)
effects [tab:depr.strstreambuf.cons.ptr]Element Value strmode
0 alsize
an unspecified value palloc
a null pointer pfree
a null pointer
gnext_arg
shall point to the first element of an array object whose number of elementsN
is determined as follows:
- — If
n > 0
,N
isn
.- — If
n == 0
,N
isstd::strlen(gnext_arg)
.- — If
n < 0
,N
isINT_MAX
.327- If
pbeg_arg
is a null pointer, the function executes:setg(gnext_arg, gnext_arg, gnext_arg + N);- Otherwise, the function executes:
setg(gnext_arg, gnext_arg, pbeg_arg); setp(pbeg_arg, pbeg_arg + N);strstreambuf(const char* gnext_arg, streamsize n); strstreambuf(const signed char* gnext_arg, streamsize n); strstreambuf(const unsigned char* gnext_arg, streamsize n);- Effects: Behaves the same as
strstreambuf((char*)gnext_arg,n)
, except that the constructor also setsconstant
instrmode
.virtual ~strstreambuf();- Effects: Destroys an object of class
strstreambuf
. The function frees the dynamically allocated array object only if(strmode & allocated) != 0
and(strmode & frozen) == 0
. (29.11.2.3 describes how a dynamically allocated array object is freed.)29.11.2.2 Member functions [strstreambuf.members]
void freeze(bool freezefl = true);- Effects: If
strmode & dynamic
is nonzero, alters the freeze status of the dynamic array object as follows:
- — If
freezefl
istrue
, the function setsfrozen
instrmode
.- — Otherwise, it clears
frozen
instrmode
.char* str();- Effects: Calls
freeze()
, then returns the beginning pointer for the input sequence,gbeg
.- Remarks: The return value can be a null pointer.
int pcount() const;- Effects: If the next pointer for the output sequence,
pnext
, is a null pointer, returns zero. Otherwise, returns the current effective length of the array object as the next pointer minus the beginning pointer for the output sequence,pnext - pbeg
.29.11.2.3
strstreambuf
overridden virtual functions [strstreambuf.virtuals]
int_type overflow(int_type c = EOF) override;- Effects: Appends the character designated by
c
to the output sequence, if possible, in one of two ways:
- — If
c != EOF
and if either the output sequence has a write position available or the function makes a write position available (as described below), assignsc
to*pnext++
. Returns(unsigned char)c
.- — If
c == EOF
, there is no character to append. Returns a value other thanEOF
.- Returns
EOF
to indicate failure.- Remarks: The function can alter the number of write positions available as a result of any call.
- To make a write position available, the function reallocates (or initially allocates) an array object with a sufficient number of elements
n
to hold the current array object (if any), plus at least one additional write position. How many additional write positions are made available is otherwise unspecified.328 Ifpalloc
is not a null pointer, the function calls(*palloc)(n)
to allocate the new dynamic array object. Otherwise, it evaluates the expressionnew charT[n]
. In either case, if the allocation fails, the function returnsEOF
. Otherwise, it setsallocated
instrmode
.- To free a previously existing dynamic array object whose first element address is
p
: Ifpfree
is not a null pointer, the function calls(*pfree)(p)
. Otherwise, it evaluates the expressiondelete[]p
.- If
(strmode & dynamic) == 0
, or if(strmode & frozen) != 0
, the function cannot extend the array (reallocate it with greater length) to make a write position available.int_type pbackfail(int_type c = EOF) override;- Puts back the character designated by
c
to the input sequence, if possible, in one of three ways:
- — If
c != EOF
, if the input sequence has a putback position available, and if(char)c == gnext[-1]
, assignsgnext - 1 to gnext
. Returnsc
.- — If
c != EOF
, if the input sequence has a putback position available, and ifstrmode & constant
is zero, assignsc
to*--gnext
. Returnsc
.- — If
c == EOF
and if the input sequence has a putback position available, assignsgnext - 1
tognext
.- Returns a value other than
EOF
. ReturnsEOF
to indicate failure.- Remarks: If the function can succeed in more than one of these ways, it is unspecified which way is chosen. The function can alter the number of putback positions available as a result of any call.
int_type underflow() override;- Effects: Reads a character from the input sequence, if possible, without moving the stream position past it, as follows:
- — If the input sequence has a read position available, the function signals success by returning
(unsigned char)*gnext
.- — Otherwise, if the current write next pointer
pnext
is not a null pointer and is greater than the current read end pointergend
, makes a read position available by assigning togend
a value greater thangnext
and no greater thanpnext
. Returns(unsigned char)*gnext
.- Returns
EOF
to indicate failure.- Remarks: The function can alter the number of read positions available as a result of any call.
pos_type seekoff(off_type off, seekdir way, openmode which = in | out) override;- Effects: Alters the stream position within one of the controlled sequences, if possible, as indicated in Table 151.
Table 151: seekoff positioning [tab:depr.strstreambuf.seekoff.pos] Conditions Result (which & ios::in) != 0
positions the input sequence (which & ios::out) != 0
positions the output sequence (which & (ios::in | ios::out)) == (ios::in | ios::out)
and eitherway == ios::beg
orway == ios::end
positions both the input and the output sequences Otherwise the positioning operation fails. - For a sequence to be positioned, if its next pointer is a null pointer, the positioning operation fails. Otherwise, the function determines
newoff
as indicated in Table 152.
Table 152: newoff values [tab:depr.strstreambuf.seekoff.newoff] Condition newoff
Valueway == ios::beg
0 way == ios::cur
the next pointer minus the beginning pointer ( xnext - xbeg
).way == ios::end
seekhigh
minus the beginning pointer (seekhigh - xbeg
).- If
(newoff + off) < (seeklow - xbeg)
or(seekhigh - xbeg) < (newoff + off)
, the positioning operation fails. Otherwise, the function assignsxbeg + newoff + off
to the next pointerxnext
.- Returns:
pos_type(newoff)
, constructed from the resultant offsetnewoff
(of typeoff_type)
, that stores the resultant stream position, if possible. If the positioning operation fails, or if the constructed object cannot represent the resultant stream position, the return value ispos_type(off_type(-1))
.pos_type seekpos(pos_type sp, ios_base::openmode which = ios_base::in | ios_base::out) override;- Effects: Alters the stream position within one of the controlled sequences, if possible, to correspond to the stream position stored in
sp
(as described below).
- — If
(which & ios::in) != 0
, positions the input sequence.- — If
(which & ios::out) != 0
, positions the output sequence.- — If the function positions neither sequence, the positioning operation fails.
- For a sequence to be positioned, if its next pointer is a null pointer, the positioning operation fails. Otherwise, the function determines
newoff
fromsp.offset()
:
- — If
newoff
is an invalid stream position, has a negative value, or has a value greater than(seekhigh - seeklow)
, the positioning operation fails- — Otherwise, the function adds
newoff
to the beginning pointerxbeg
and stores the result in the next pointerxnext
.- Returns:
pos_type(newoff)
, constructed from the resultant offsetnewoff
(of typeoff_type
), that stores the resultant stream position, if possible. If the positioning operation fails, or if the constructed object cannot represent the resultant stream position, the return value ispos_type(off_type(-1))
.streambuf<char>* setbuf(char* s, streamsize n) override;- Effects: Behavior is implementation-defined, except that
setbuf(0, 0)
has no effect.29.11.3 Class
istrstream
[istrstream]namespace std { class istrstream : public basic_istream<char> { public: explicit istrstream(const char* s); explicit istrstream(char* s); istrstream(const char* s, streamsize n); istrstream(char* s, streamsize n); virtual ~istrstream(); strstreambuf* rdbuf() const; char* str(); private: strstreambuf sb; // exposition only }; }
- The class
istrstream
supports the reading of objects of classstrstreambuf
. It supplies astrstreambuf
object to control the associated array object. For the sake of exposition, the maintained data is presented here as:
- —
sb
, thestrstreambuf
object.29.11.3.1
istrstream
constructors [istrstream.cons]
explicit istrstream(const char* s); explicit istrstream(char* s);- Effects: Initializes the base class with
istream(&sb)
andsb
withstrstreambuf(s, 0)
.s
shall designate the first element of an NTBS.istrstream(const char* s, streamsize n); istrstream(char* s, streamsize n);- Effects: Initializes the base class with
istream(&sb)
andsb
withstrstreambuf(s, n)
.s
shall designate the first element of an array whose length isn
elements, andn
shall be greater than zero.29.11.3.2 Member functions [istrstream.members]
strstreambuf* rdbuf() const;- Returns:
const_cast<strstreambuf*>(&sb)
.char* str();- Returns:
rdbuf()->str()
.12.4 Class
ostrstream
[ostrstream]namespace std { class ostrstream : public basic_ostream<char> { public: ostrstream(); ostrstream(char* s, int n, ios_base::openmode mode = ios_base::out); virtual ~ostrstream(); strstreambuf* rdbuf() const; void freeze(bool freezefl = true); char* str(); int pcount() const; private: strstreambuf sb; // exposition only }; }
- The class
ostrstream
supports the writing of objects of classstrstreambuf
. It supplies astrstreambuf
object to control the associated array object. For the sake of exposition, the maintained data is presented here as:
- —
sb
, thestrstreambuf
object.29.11.4.1
ostrstream
constructors [ostrstream.cons]
ostrstream();- Effects: Initializes the base class with
ostream(&sb)
andsb
withstrstreambuf()
.ostrstream(char* s, int n, ios_base::openmode mode = ios_base::out);- Effects: Initializes the base class with
ostream(&sb)
, andsb
with one of two constructors:
- — If
(mode & app) == 0
, thens
shall designate the first element of an array ofn
elements. The constructor isstrstreambuf(s, n, s)
.- — If
(mode & app) != 0
, thens
shall designate the first element of an array ofn
elements that contains an NTBS whose first element is designated bys
. The constructor isstrstreambuf(s, n, s + std::strlen(s))
.32929.11.4.2 Member functions [ostrstream.members]
strstreambuf* rdbuf() const;- Returns:
(strstreambuf*)&sb
.void freeze(bool freezefl = true);- Effects: Calls
rdbuf()->freeze(freezefl)
.char* str();- Returns:
rdbuf()->str()
.int pcount() const;- Returns:
rdbuf()->pcount()
.29.11.5 Class
strstream
[strstream]namespace std { class strstream : public basic_iostream<char> { public: // types using char_type = char; using int_type = char_traits<char>::int_type; using pos_type = char_traits<char>::pos_type; using off_type = char_traits<char>::off_type; // constructors/destructor strstream(); strstream(char* s, int n, ios_base::openmode mode = ios_base::in|ios_base::out); virtual ~strstream(); // members strstreambuf* rdbuf() const; void freeze(bool freezefl = true); int pcount() const; char* str(); private: strstreambuf sb; // exposition only }; }
- The class
strstream
supports reading and writing from objects of classstrstreambuf
. It supplies astrstreambuf
object to control the associated array object. For the sake of exposition, the maintained data is presented here as:
- —
sb
, thestrstreambuf
object.29.11.5.1
strstream
constructors [strstream.cons]
strstream();- Effects: Initializes the base class with
iostream(&sb)
.strstream(char* s, int n, ios_base::openmode mode = ios_base::in|ios_base::out);- Effects: Initializes the base class with
iostream(&sb)
, andsb
with one of the two constructors:
- — If
(mode & app) == 0
, thens
shall designate the first element of an array ofn
elements. The constructor isstrstreambuf(s,n,s)
.- — If
(mode & app) != 0
, thens
shall designate the first element of an array ofn
elements that contains an NTBS whose first element is designated bys
. The constructor isstrstreambuf(s,n,s + std::strlen(s))
.29.11.5.2
strstream
destructor [strstream.dest]
virtual ~strstream();- Effects: Destroys an object of class
strstream
.29.11.5.3
strstream
operations [strstream.oper]
strstreambuf* rdbuf() const;- Returns:
const_cast<strstreambuf*>(&sb)
.void freeze(bool freezefl = true);- Effects: Calls
rdbuf()->freeze(freezefl)
.char* str();- Returns:
rdbuf()->str()
.int pcount() const;- Returns:
rdbuf()->pcount()
.
D.12char*
streams [depr.str.strstreams]
D.12.1 Header<strstream>
synopsis [depr.strstream.syn]
The header<strstream>
defines types that associate stream buffers with character array objects and assist reading and writing such objects.namespace std { class strstreambuf; class istrstream; class ostrstream; class strstream; }
D.12.2 Classstrstreambuf
[depr.strstreambuf]namespace std { class strstreambuf : public basic_streambuf<char> { public: strstreambuf() : strstreambuf(0) {} explicit strstreambuf(streamsize alsize_arg); strstreambuf(void* (*palloc_arg)(size_t), void (*pfree_arg)(void*)); strstreambuf(char* gnext_arg, streamsize n, char* pbeg_arg = nullptr); strstreambuf(const char* gnext_arg, streamsize n); strstreambuf(signed char* gnext_arg, streamsize n, signed char* pbeg_arg = nullptr); strstreambuf(const signed char* gnext_arg, streamsize n); strstreambuf(unsigned char* gnext_arg, streamsize n, unsigned char* pbeg_arg = nullptr); strstreambuf(const unsigned char* gnext_arg, streamsize n); virtual ~strstreambuf(); void freeze(bool freezefl = true); char* str(); int pcount(); protected: int_type overflow (int_type c = EOF) override; int_type pbackfail(int_type c = EOF) override; int_type underflow() override; pos_type seekoff(off_type off, ios_base::seekdir way, ios_base::openmode which = ios_base::in | ios_base::out) override; pos_type seekpos(pos_type sp, ios_base::openmode which = ios_base::in | ios_base::out) override; streambuf* setbuf(char* s, streamsize n) override; private: using strstate = T1; // exposition only static const strstate allocated; // exposition only static const strstate constant; // exposition only static const strstate dynamic; // exposition only static const strstate frozen; // exposition only strstate strmode; // exposition only streamsize alsize; // exposition only void* (*palloc)(size_t); // exposition only void (*pfree)(void*); // exposition only }; }
The classstrstreambuf
associates the input sequence, and possibly the output sequence, with an object of some character array type, whose elements store arbitrary values. The array object has several attributes.[Note: For the sake of exposition, these are represented as elements of a bitmask type (indicated here asT1
) calledstrstate
. The elements are:—end note]
—allocated
, set when a dynamic array object has been allocated, and hence should be freed by the destructor for thestrstreambuf
object;—constant
, set when the array object hasconst
elements, so the output sequence cannot be written;—dynamic
, set when the array object is allocated (or reallocated) as necessary to hold a character sequence that can change in length;—frozen
, set when the program has requested that the array object not be altered, reallocated, or freed.[Note: For the sake of exposition, the maintained data is presented here as:—end note]
—strstate strmode
, the attributes of the array object associated with thestrstreambuf
object;—int alsize
, the suggested minimum size for a dynamic array object;—void* (*palloc)(size_t)
, points to the function to call to allocate a dynamic array object;—void (*pfree)(void*)
, points to the function to call to free a dynamic array object.Each object of classstrstreambuf
has a seekable area, delimited by the pointersseeklow
andseekhigh
. Ifgnext
is a null pointer, the seekable area is undefined. Otherwise,seeklow
equalsgbeg
andseekhigh
is eitherpend
, ifpend
is not a null pointer, orgend
.
D.12.2.1strstreambuf
constructors [depr.strstreambuf.cons]
explicit strstreambuf(streamsize alsize_arg);Effects: Initializes the base class withstreambuf()
. The postconditions of this function are indicated in Table 148.
Table 148:strstreambuf(streamsize)
effects [tab:depr.strstreambuf.cons.sz]ElementValuestrmode
dynamic
alsize
alsize_arg
palloc
a null pointerpfree
a null pointer
strstreambuf(void* (*palloc_arg)(size_t), void (*pfree_arg)(void*));Effects: Initializes the base class withstreambuf()
. The postconditions of this function are indicated in Table 149.
Table 149:strstreambuf(void* (*)(size_t), void (*)(void*))
effects [tab:depr.strstreambuf.cons.alloc]ElementValuestrmode
dynamic
alsize
an unspecified valuepalloc
palloc_arg
pfree
pfree_arg
strstreambuf(char* gnext_arg, streamsize n, char* pbeg_arg = nullptr); strstreambuf(signed char* gnext_arg, streamsize n, signed char* pbeg_arg = nullptr); strstreambuf(unsigned char* gnext_arg, streamsize n, unsigned char* pbeg_arg = nullptr);Effects: Initializes the base class withstreambuf()
. The postconditions of this function are indicated in Table 150.
Table 150:strstreambuf(charT*, streamsize, charT*)
effects [tab:depr.strstreambuf.cons.ptr]ElementValuestrmode
0alsize
an unspecified valuepalloc
a null pointerpfree
a null pointer
gnext_arg
shall point to the first element of an array object whose number of elementsN
is determined as follows:
— Ifn > 0
,N
isn
.— Ifn == 0
,N
isstd::strlen(gnext_arg)
.— Ifn < 0
,N
isINT_MAX
.327Ifpbeg_arg
is a null pointer, the function executes:setg(gnext_arg, gnext_arg, gnext_arg + N);Otherwise, the function executes:setg(gnext_arg, gnext_arg, pbeg_arg); setp(pbeg_arg, pbeg_arg + N);strstreambuf(const char* gnext_arg, streamsize n); strstreambuf(const signed char* gnext_arg, streamsize n); strstreambuf(const unsigned char* gnext_arg, streamsize n);Effects: Behaves the same asstrstreambuf((char*)gnext_arg,n)
, except that the constructor also setsconstant
instrmode
.virtual ~strstreambuf();Effects: Destroys an object of classstrstreambuf
. The function frees the dynamically allocated array object only if(strmode & allocated) != 0
and(strmode & frozen) == 0
. (D.12.2.3 describes how a dynamically allocated array object is freed.)
D.12.2.2 Member functions [depr.strstreambuf.members]
void freeze(bool freezefl = true);Effects: Ifstrmode & dynamic
is nonzero, alters the freeze status of the dynamic array object as follows:
— Iffreezefl
istrue
, the function setsfrozen
instrmode
.— Otherwise, it clearsfrozen
instrmode
.char* str();Effects: Callsfreeze()
, then returns the beginning pointer for the input sequence,gbeg
.Remarks: The return value can be a null pointer.int pcount() const;Effects: If the next pointer for the output sequence,pnext
, is a null pointer, returns zero. Otherwise, returns the current effective length of the array object as the next pointer minus the beginning pointer for the output sequence,pnext - pbeg
.
D.12.2.3strstreambuf
overridden virtual functions [depr.strstreambuf.virtuals]
int_type overflow(int_type c = EOF) override;Effects: Appends the character designated byc
to the output sequence, if possible, in one of two ways:
— Ifc != EOF
and if either the output sequence has a write position available or the function makes a write position available (as described below), assignsc
to*pnext++
. Returns(unsigned char)c
.— Ifc == EOF
, there is no character to append. Returns a value other thanEOF
.ReturnsEOF
to indicate failure.Remarks: The function can alter the number of write positions available as a result of any call.To make a write position available, the function reallocates (or initially allocates) an array object with a sufficient number of elementsn
to hold the current array object (if any), plus at least one additional write position. How many additional write positions are made available is otherwise unspecified.328 Ifpalloc
is not a null pointer, the function calls(*palloc)(n)
to allocate the new dynamic array object. Otherwise, it evaluates the expressionnew charT[n]
. In either case, if the allocation fails, the function returnsEOF
. Otherwise, it setsallocated
instrmode
.To free a previously existing dynamic array object whose first element address isp
: Ifpfree
is not a null pointer, the function calls(*pfree)(p)
. Otherwise, it evaluates the expressiondelete[]p
.If(strmode & dynamic) == 0
, or if(strmode & frozen) != 0
, the function cannot extend the array (reallocate it with greater length) to make a write position available.int_type pbackfail(int_type c = EOF) override;Puts back the character designated byc
to the input sequence, if possible, in one of three ways:
— Ifc != EOF
, if the input sequence has a putback position available, and if(char)c == gnext[-1]
, assignsgnext - 1 to gnext
. Returnsc
.— Ifc != EOF
, if the input sequence has a putback position available, and ifstrmode & constant
is zero, assignsc
to*--gnext
. Returnsc
.— Ifc == EOF
and if the input sequence has a putback position available, assignsgnext - 1
tognext
.Returns a value other thanEOF
. ReturnsEOF
to indicate failure.Remarks: If the function can succeed in more than one of these ways, it is unspecified which way is chosen. The function can alter the number of putback positions available as a result of any call.int_type underflow() override;Effects: Reads a character from the input sequence, if possible, without moving the stream position past it, as follows:
— If the input sequence has a read position available, the function signals success by returning(unsigned char)*gnext
.— Otherwise, if the current write next pointerpnext
is not a null pointer and is greater than the current read end pointergend
, makes a read position available by assigning togend
a value greater thangnext
and no greater thanpnext
. Returns(unsigned char)*gnext
.ReturnsEOF
to indicate failure.Remarks: The function can alter the number of read positions available as a result of any call.pos_type seekoff(off_type off, seekdir way, openmode which = in | out) override;Effects: Alters the stream position within one of the controlled sequences, if possible, as indicated in Table 151.For a sequence to be positioned, if its next pointer is a null pointer, the positioning operation fails. Otherwise, the function determinesnewoff
as indicated in Table 152.If(newoff + off) < (seeklow - xbeg)
or(seekhigh - xbeg) < (newoff + off)
, the positioning operation fails. Otherwise, the function assignsxbeg + newoff + off
to the next pointerxnext
.Returns:pos_type(newoff)
, constructed from the resultant offsetnewoff
(of typeoff_type)
, that stores the resultant stream position, if possible. If the positioning operation fails, or if the constructed object cannot represent the resultant stream position, the return value ispos_type(off_type(-1))
.
Table 151: seekoff positioning [tab:depr.strstreambuf.seekoff.pos]ConditionsResult(which & ios::in) != 0
positions the input sequence(which & ios::out) != 0
positions the output sequence(which & (ios::in | ios::out)) == (ios::in | ios::out)
and eitherway == ios::beg
orway == ios::end
positions both the input and the output sequencesOtherwisethe positioning operation fails.
Table 152: newoff values [tab:depr.strstreambuf.seekoff.newoff]Conditionnewoff
Valueway == ios::beg
0way == ios::cur
the next pointer minus the beginning pointer (xnext - xbeg
).way == ios::end
seekhigh
minus the beginning pointer (seekhigh - xbeg
).pos_type seekpos(pos_type sp, ios_base::openmode which = ios_base::in | ios_base::out) override;Effects: Alters the stream position within one of the controlled sequences, if possible, to correspond to the stream position stored insp
(as described below).
— If(which & ios::in) != 0
, positions the input sequence.— If(which & ios::out) != 0
, positions the output sequence.— If the function positions neither sequence, the positioning operation fails.For a sequence to be positioned, if its next pointer is a null pointer, the positioning operation fails. Otherwise, the function determinesnewoff
fromsp.offset()
:
— Ifnewoff
is an invalid stream position, has a negative value, or has a value greater than(seekhigh - seeklow)
, the positioning operation fails— Otherwise, the function addsnewoff
to the beginning pointerxbeg
and stores the result in the next pointerxnext
.Returns:pos_type(newoff)
, constructed from the resultant offsetnewoff
(of typeoff_type
), that stores the resultant stream position, if possible. If the positioning operation fails, or if the constructed object cannot represent the resultant stream position, the return value ispos_type(off_type(-1))
.streambuf<char>* setbuf(char* s, streamsize n) override;Effects: Behavior is implementation-defined, except thatsetbuf(0, 0)
has no effect.
D.12.3 Classistrstream
[depr.istrstream]namespace std { class istrstream : public basic_istream<char> { public: explicit istrstream(const char* s); explicit istrstream(char* s); istrstream(const char* s, streamsize n); istrstream(char* s, streamsize n); virtual ~istrstream(); strstreambuf* rdbuf() const; char* str(); private: strstreambuf sb; // exposition only }; }
The classistrstream
supports the reading of objects of classstrstreambuf
. It supplies astrstreambuf
object to control the associated array object. For the sake of exposition, the maintained data is presented here as:
—sb
, thestrstreambuf
object.
D.12.3.1istrstream
constructors [depr.istrstream.cons]
explicit istrstream(const char* s); explicit istrstream(char* s);Effects: Initializes the base class withistream(&sb)
andsb
withstrstreambuf(s, 0)
.s
shall designate the first element of an NTBS.istrstream(const char* s, streamsize n); istrstream(char* s, streamsize n);Effects: Initializes the base class withistream(&sb)
andsb
withstrstreambuf(s, n)
.s
shall designate the first element of an array whose length isn
elements, andn
shall be greater than zero.
D.12.3.2 Member functions [depr.istrstream.members]
strstreambuf* rdbuf() const;Returns:const_cast<strstreambuf*>(&sb)
.char* str();Returns:rdbuf()->str()
.
12.4 Classostrstream
[depr.ostrstream]namespace std { class ostrstream : public basic_ostream<char> { public: ostrstream(); ostrstream(char* s, int n, ios_base::openmode mode = ios_base::out); virtual ~ostrstream(); strstreambuf* rdbuf() const; void freeze(bool freezefl = true); char* str(); int pcount() const; private: strstreambuf sb; // exposition only }; }
The classostrstream
supports the writing of objects of classstrstreambuf
. It supplies astrstreambuf
object to control the associated array object. For the sake of exposition, the maintained data is presented here as:
—sb
, thestrstreambuf
object.
D.12.4.1ostrstream
constructors [depr.ostrstream.cons]
ostrstream();Effects: Initializes the base class withostream(&sb)
andsb
withstrstreambuf()
.ostrstream(char* s, int n, ios_base::openmode mode = ios_base::out);Effects: Initializes the base class withostream(&sb)
, andsb
with one of two constructors:
— If(mode & app) == 0
, thens
shall designate the first element of an array ofn
elements. The constructor isstrstreambuf(s, n, s)
.— If(mode & app) != 0
, thens
shall designate the first element of an array ofn
elements that contains an NTBS whose first element is designated bys
. The constructor isstrstreambuf(s, n, s + std::strlen(s))
.329
D.12.4.2 Member functions [depr.ostrstream.members]
strstreambuf* rdbuf() const;Returns:(strstreambuf*)&sb
.void freeze(bool freezefl = true);Effects: Callsrdbuf()->freeze(freezefl)
.char* str();Returns:rdbuf()->str()
.int pcount() const;Returns:rdbuf()->pcount()
.
D.12.5 Classstrstream
[depr.strstream]namespace std { class strstream : public basic_iostream<char> { public: // types using char_type = char; using int_type = char_traits<char>::int_type; using pos_type = char_traits<char>::pos_type; using off_type = char_traits<char>::off_type; // constructors/destructor strstream(); strstream(char* s, int n, ios_base::openmode mode = ios_base::in|ios_base::out); virtual ~strstream(); // members strstreambuf* rdbuf() const; void freeze(bool freezefl = true); int pcount() const; char* str(); private: strstreambuf sb; // exposition only }; }
The classstrstream
supports reading and writing from objects of classstrstreambuf
. It supplies astrstreambuf
object to control the associated array object. For the sake of exposition, the maintained data is presented here as:
—sb
, thestrstreambuf
object.
D.12.5.1strstream
constructors [depr.strstream.cons]
strstream();Effects: Initializes the base class withiostream(&sb)
.strstream(char* s, int n, ios_base::openmode mode = ios_base::in|ios_base::out);Effects: Initializes the base class withiostream(&sb)
, andsb
with one of the two constructors:
— If(mode & app) == 0
, thens
shall designate the first element of an array ofn
elements. The constructor isstrstreambuf(s,n,s)
.— If(mode & app) != 0
, thens
shall designate the first element of an array ofn
elements that contains an NTBS whose first element is designated bys
. The constructor isstrstreambuf(s,n,s + std::strlen(s))
.
D.12.5.2strstream
destructor [depr.strstream.dest]
virtual ~strstream();Effects: Destroys an object of classstrstream
.
D.12.5.3strstream
operations [depr.strstream.oper]
strstreambuf* rdbuf() const;Returns:const_cast<strstreambuf*>(&sb)
.void freeze(bool freezefl = true);Effects: Callsrdbuf()->freeze(freezefl)
.char* str();Returns:rdbuf()->str()
.int pcount() const;Returns:rdbuf()->pcount()
.
The main benefit of strstreams is that they provide efficient management of a
shared buffer, reducing much unnecessary copying of data, at the expense of
more user responsibility for managing memory. Concerns remain about removing
this facility without providing a replacement, such as Peter Sommerlad's
spanstream
library that is currently awaiting LWG review for
C++23. There was no enthusiasm for undeprecating the library at this point,
with clear desire to follow the various paths towards finally removing this
feature after the intended replacement has landed. It is not clear whether we
would welcome removal in C++23 if the replacement lands in this release, or in
a subsequent standard. That discussion is deferred to discussions of the
replacement facility itself.
Consensus to take no further action on this topic within this omnibus paper.
LEWGI Review: Take no action at this time.
First deprecated: C++20While traits have been deprecated in earlier versions of the standard, C++20 removed all traits deprecated in C++17 and earlier, so the traits that remain were all deprecated by C++20.
The is_pod
trait was deprecated by paper
P0767R1
as part of removing the POD vocabulary from the C++ standard, both core and
library. The term had changed meaning so frequently that it no longer served
as useful vocabulary. The type trait was extracted to Annex D, and now itself
provides the only definition in the standard for a POD. Client code is
encouraged to use the more specific traits for trivial and standard layout
types to better describe their need. Note that the related term POF, for Plain
Old Function, was removed from C++17 by paper
P0270R3.
The is_pod
trait was first supplied as part of C++11, so its
removal would potentially impact programs written against the C++11 standard or
later. Users have had up to three years of implementations warning on use of
the deprecated trait, so we could consider removal, with the usual proviso that
the name is preserved as a zombie for previous standardization. As the case
for removal is not urgent, the weak recommendation of this paper is to remove
the trait from C++23.
However, the current wording does not follow library best practices, and should be updated to better specify the Requires clauses with our modern vocabulary. The strong recommendation is to retain it as deprecated in Annex D and revise the wording accordingly.
Strong recommendation: Clean up wording.
Update wording to replace Requires clauses (see D.10 Requires paragraph).
Weak recommendation: Remove the traits that can live on as zombies.
16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
is_pod
,- —
is_pod_v
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.
D.13 Deprecated Type Traits [depr.meta.type]
The header<type_traits>
(20.15.2) has the following additions:namespace std {template<class T> struct is_pod;template<class T> inline constexpr bool is_pod_v = is_pod<T>::value;}The behavior of a program that adds specializations for any of the templates defined in this subclause is undefined, unless explicitly permitted by the specification of the corresponding template.template<class T> struct is_pod;Requires:remove_all_extents_t<T>
shall be a complete type or cvvoid
.is_pod<T>
is a Cpp17UnaryTypeTrait (20.15.1) with a base characteristic oftrue_type
ifT
is a POD type, andfalse_type
otherwise. A POD class is a class that is both a trivial class and a standard-layout class, and has no non-static data members of type non-POD class (or array thereof). A POD type is a scalar type, a POD class, an array of such a type, or a cv-qualified version of one of these types.[Note: It is unspecified whether a closure type (7.5.5.1) is a POD type. —end note]
Draft compatibility note for Annex C.
Concerns were raised about the lack of research into how much code is likely to break with the removal of this API. We would like to see more numbers and analysis on publicly available code, such as across all of Github. Given that implementers are likely to provide an implementation (through zombie names freedom) for some time after removal, there was consensus to proceed with removal, assuming the requested research does not reveal major concerns before the main LEWG review to follow.
LEWGI Review: Remove this feature from C++23
There is a preference for removing replaced facilities from the standard at the earliest opportunity, and letting the vendors remove the zombie implementations at a time of their own choosing, assessing their own customer demand.
First deprecated: C++20
This feature was deprecated as part of the effort to clean up the use and
semantics of volatile
in the language, and was adopted by paper
P1831R1.
It potentially impacts on code written against C++11 and later standards.
This feature was deprecated only at the final meeting of the C++20 development cycle, and at the time this paper is written, there is no experience with how much code has been impacted by this deprecation. As such, it is too early to make any recommendation for a change with respect to this feature.
It is further noted that there is a coupling between the deprecated tuple
traits API, and the deprecated support for volatile structured bindings (D.5).
First, note that volatile tuple
itself does not support structured
bindings (nor do any pair
and array
) as there are no
overloads for the get
function taking references to
volatile
qualified objects. However, it is still possible for a
user to customize their own type to support such get
calls. If we
remove the volatile
support from the tuple traits by default, then
the user would have to provide their own specializations for
tuple_size<volatile TYPE>
and
tuple_element<volatile TYPE>
, and similarly for the
const volatile
qualifier. Alternatively, we could ensure we
remove the deprecated support for volatile
structured bindings at
the same time that we remove the tuple traits volatile
API.
Strong recommendation: take no action.
No change to draft.
Weak recommendation: take no action.
No change to draft.
None of the standard library uses these traits, they are not user customizable,
and volatile
qualified tuples have no usable member functions or
accessors, so it seems safe to remove these partial specializations at this
point. The only concern is that the structured bindings core language facility
relies on these specializations for the deprecated volatile
structured bindings feature. Consensus is that we would like to remove this
feature, contingent on EWG also removing the core support for
volatile
structured bindings.
LEWGI Review:
Remove this feature from C++23 if volatile
structured bindings are
removed by EWG.
First deprecated: C++20
This feature was deprecated as part of the effort to clean up the use and
semantics of volatile
in the language, and was adopted by paper
P1831R1.
It potentially impacts on code written against C++17 and later standards.
This feature was deprecated only at the final meeting of the C++20 development
cycle, and at the time this paper is written, there is no experience with how
much code has been impacted by this deprecation. However, variant
has been in the language for a much shorter time than the similarly impacted
tuple
, and so it is likely that much less code would be impacted
by its removal. Secondly, as variant
has no volatile
qualified member functions, nor external accessors like get
accepting volatile
variants, the scope for reasonable use of a
volatile
variant is vanishingly small. Therefore, the strong
recommendation of this paper is to remove directly from C++23, and the weak
recommendation is to hold it over for one more standard cycle, allowing more
time for any vestigial usage to be reworked.
Strong recommendation: remove this feature from C++23.
D.15 Variant [depr.variant]
The header<variant>
(20.7.2) has the following additions:namespace std {template<class T> struct variant_size<volatile T>;template<class T> struct variant_size<const volatile T>;template<size_t I, class T> struct variant_alternative<I, volatile T>;template<size_t I, class T> struct variant_alternative<I, const volatile T>;}template<class T> class variant_size<volatile T>;template<class T> class variant_size<const volatile T>;LetVS
denotevariant_size<T>
of the cv-unqualified typeT
. Then specializations of each of the two templates meet the Cpp17UnaryTypeTrait requirements with a base characteristic ofintegral_constant<size_t, VS::value>
.template<size_t I, class T> class variant_alternative<I, volatile T>;template<size_t I, class T> class variant_alternative<I, const volatile T>;LetVA
denotevariant_alternative<I, T>
of the cv-unqualified typeT
. Then specializations of each of the two templates meet the Cpp17TransformationTrait requirements with a member typedef type that names the following type:
— for the first specialization,add_volatile_t<VA::type>
, and— for the second specialization,add_cv_t<VA::type>
.
Draft compatibility note for Annex C.
Weak recommendation: take no action.
No change to draft.
None of the standard library uses these traits, they are not user customizable,
and volatile
qualified variant
s have no usable member
functions or accessors. Consensus to remove these partial specializations from
C++23.
LEWGI Review: Remove this feature from C++23
First deprecated: C++17
The class template iterator
was first deprecated in C++17 by the paper
P0174R2.
The concern was that providing the needed support for iterator typenames
through a templated base class, determining which name maps to which type
purely by parameter order, was less clear than simply providing the needed
names. Further, there were corner cases in usage that fell out of template
syntax that made this tool hard to recommend as a simpler way of providing the
type names, yet that was its whole reason to exist.
When this facility was reviewed for removal in C++20, it was noted that there
were valid uses that relied on the default template arguments to deduce at
least a few of the needed type names. Subsequent work on iterators and ranges
(P0896R4)
that landed in C++20 now means that work is also done by the primary
iterator_traits
template, and so the remaining use case (for new
code) is also covered, making this class template strictly redundant.
The main concern that remains is breaking old code by removing this code from
the standard libraries. That risk is ameliorated by the zombie names clause in
the standard, allowing vendors to maintain their own support for as long as
their customers demand. By the time C++23 ships, those customers will already
have been on 6 years notice that their code might not be supported in future
standards. However, we note the repeated use of the name iterator
as a type within many containers means we might choose to leave this name off
the zombie list. We conservatively place it there anyway, to ensure that we
are covered by the previous standardization terminology to encompass uses other
than as a container iterator typedef.
The strong recommendation of this paper is to remove this feature from the pending standard. The weak recommendation is to ask what further changes we would like to see before we could remove this feature. If there is no anticipated change of circumstance that would allow the standard to stop tracking this feature, then the recommendation should be to undeprecate, and restore this text to the main body of the standard.
Strong recommendation: remove this feature from C++23.
16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
iterator
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.
D.16 Deprecated iterator primitives [depr.iterator.primitives]
D.16.1 Basic iterator [depr.iterator.basic]
The header<iterator>
(23.2) has the following addition:namespace std {template<class Category, class T, class Distance = ptrdiff_t,class Pointer = T*, class Reference = T&>struct iterator {using iterator_category = Category;using value_type = T;using difference_type = Distance;using pointer = Pointer;using reference = Reference;};}Theiterator
template may be used as a base class to ease the definition of required types for new iterators.[ Note: If the new iterator type is a class template, then these aliases will not be visible from within the iterator class's template definition, but only to callers of that class — end note][ Example: If a C++ program wants to define a bidirectional iterator for some data structure containingdouble
and such that it works on a large memory model of the implementation, it can do so with:— end example ]class MyIterator :public iterator<bidirectional_iterator_tag, double, long, T*, T&> {// code implementing ++, etc.};
Draft compatibility note for Annex C.
Weak recommendation: Undeprecate the facility, and restore it to clause 23.
Wording to be supplied.
Concerns were raised about the lack of research into how much code is likely to
break with the removal of this API. We would like to see more numbers and
analysis on publicly available code, such as across all of Github. The better
treatment of implicit generation of iterator_traits
in C++23, and
more familiarity with a limited number of code bases that still rely on this
facility, gave more confidence in moving forward with removal than we had for
C++20. It was also noted that the name may be unfortunate with the chosen form
of concept naming adopted for C++20, and so its removal might lead to one fewer
sources of future confusion, Given that implementers are likely to provide an
implementation (through zombie names freedom) for some time after removal,
there was consensus to proceed with removal, assuming the requested research
does not reveal major concerns before the main LEWG review to follow.
LEWGI Review: Remove this feature from C++23
move_iterator
access [depr.move.iter.elem]First deprecated: C++20
This feature was deprecated for C++20 by the paper
P1252R2
highlighting the concern that for a move iterator adapter, intending to expose
its target as an rvalue (or xvalue), the arrow operator must return the
original adapted iterator, which will likely produce an lvalue when
dereferenced. The operator is not fit for purpose, and cannot be fixed. The
workaround for users is to dereference the move iterator with
operator *
and call the member they wish to access using the
familiar .
notation. This preserves the value category of the
iterator's target.
The proposal for C++20 was to deprecate this operator, with a view to removal at a later date. While it may seem early, this is the first such later date appropriate to consider that removal.
Lacking clear evidence that this issue is causing widespread bugs in practice, the strong recommendation for this paper is to hold the feature in Annex D for another standard cycle, and strongly consider its removal in C++23. The weak recommendation is that with three years (or more) experience of deprecation warnings, we have given users enough notice, and it is time to remove this misleading feature now, from C++23. Code written against the C++11 standard or later might be impacted by this change.
Strong recommendation: Take no action.
No change to draft.
Weak recommendation: Remove this feature from C++23
D.17 Deprecated move_iterator access [depr.move.iter.elem]
The following member is declared in addition to those members specified in 23.5.3.5:namespace std {template<class Iterator>class move_iterator {public:constexpr pointer operator->() const;};}constexpr pointer operator->() const;Returns:current
.
Draft compatibility note for Annex C.
Eric provided the rationale for the ranges reworking removing this overload,
which is that move iterator is supposed to return rvalues, as seen through the
operator*
overload returning an rvalue reference, but there is no
way for a pointer to return a pointer-to-rvalue distinct from a pointer-to-lvalue,
and in such cases it is better for this operation to not exist at all. That is
perfectly in keeping with the revised iterator concepts. Consensus to proceed
with removal, assuming the subsequent research does not reveal major concerns
before the main LEWG review that is to follow.
LEWGI Review: Remove this feature from C++23
shared_ptr
atomic access [depr.util.smartptr.shared.atomic]First deprecated: C++20
The legacy C-style atomic API for manipulating shared pointers provided in
C++11 is subtle, frequently misunderstood: a shared_ptr
that is to
be used with the atomic API can never be used directly, but can only be
manipulated through the atomic API (other than construction and destruction).
Its failure mode on misuse is silent undefined behavior, typically a data race.
C++20 provides a type-safe alternative that also provides support for
atomic<weak_ptr<T>>
.
Strong recommendation: Remove this feature at the earliest opportunity, ideally C++23.
D.18 Deprecatedshared_ptr
atomic access [depr.util.smartptr.shared.atomic]
The header<memory>
(20.10.2) has the following additions:namespace std {template <class T>bool atomic_is_lock_free(const shared_ptr<T>* p);template <class T>shared_ptr<T> atomic_load(const shared_ptr<T>* p);template <class T>shared_ptr<T> atomic_load_explicit(const shared_ptr<T>* p, memory_order mo);template <class T>void atomic_store(shared_ptr<T>* p, shared_ptr<T> r);template <class T>void atomic_store_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo);template <class T>shared_ptr<T> atomic_exchange(shared_ptr<T>* p, shared_ptr<T> r);template <class T>shared_ptr<T> atomic_exchange_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo);template <class T>bool atomic_compare_exchange_weak(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w);template <class T>bool atomic_compare_exchange_strong(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w);template <class T>bool atomic_compare_exchange_weak_explicit(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w,memory_order success, memory_order failure);template <class T>bool atomic_compare_exchange_strong_explicit(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w,memory_order success, memory_order failure);}Concurrent access to ashared_ptr
object from multiple threads does not introduce a data race if the access is done exclusively via the functions in this section and the instance is passed as their first argument.The meaning of the arguments of typememory_order
is explained in 31.4.template<class T>bool atomic_is_lock_free(const shared_ptr<T>* p);Requires:p
shall not be null.Returns:true
if atomic access to*p
is lock-free,false
otherwise.Throws: Nothing.template<class T>shared_ptr<T> atomic_load(const shared_ptr<T>* p);Requires:p
shall not be null.Returns:atomic_load_explicit(p, memory_order::seq_cst)
.Throws: Nothing.template<class T>shared_ptr<T> atomic_load_explicit(const shared_ptr<T>* p, memory_order mo);Requires:p
shall not be null.Requires:mo
shall not bememory_order::release
ormemory_order::acq_rel
.Returns:*p
.Throws: Nothing.template<class T>void atomic_store(shared_ptr<T>* p, shared_ptr<T> r);Requires:p
shall not be null.Effects: As if byatomic_store_explicit(p, r, memory_order::seq_cst)
.Throws: Nothing.template<class T>void atomic_store_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo);Requires:p
shall not be null.Requires:mo
shall not bememory_order::acquire
ormemory_order::acq_rel
.Effects: As if byp->swap(r)
.Throws: Nothing.template<class T>shared_ptr<T> atomic_exchange(shared_ptr<T>* p, shared_ptr<T> r);Requires:p
shall not be null.Returns:atomic_exchange_explicit(p, r, memory_order::seq_cst)
.Throws: Nothing.template<class T>shared_ptr<T> atomic_exchange_explicit(shared_ptr<T>* p, shared_ptr<T> r, memory_order mo);Requires:p
shall not be null.Effects: As if byp->swap(r)
.Returns: The previous value of*p
.Throws: Nothing.template<class T>bool atomic_compare_exchange_weak(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w);Requires:p
shall not be null.Returns:atomic_compare_exchange_weak_explicit(p, v, w, memory_order::seq_cst, memory_order::seq_cst)
.Throws: Nothing.template<class T>bool atomic_compare_exchange_strong(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w);Returns:atomic_compare_exchange_strong_explicit(p, v, w, memory_order::seq_cst, memory_order::seq_cst)
.template <class T>bool atomic_compare_exchange_weak_explicit(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w,memory_order success, memory_order failure);template <class T>bool atomic_compare_exchange_strong_explicit(shared_ptr<T>* p, shared_ptr<T>* v, shared_ptr<T> w,memory_order success, memory_order failure);Requires:p
shall not be null andv
shall not be null. Thefailure
argument shall not bememory_order::release
normemory_order::acq_rel
.Effects: If*p
is equivalent to*v
, assignsw
to*p
and has synchronization semantics corresponding to the value ofsuccess
, otherwise assigns*p
to*v
and has synchronization semantics corresponding to the value offailure
.Returns:true
if*p
was equivalent to*v
,false
otherwise.Throws: Nothing.Remarks: Twoshared_ptr
objects are equivalent if they store the same pointer value and share ownership. The weak form may fail spuriously. See 31.8.1.
Draft compatibility note for Annex C.
Weak recommendation: Remove this feature at the earliest opportunity after C++23.
Update wording to replace Requires clauses (see D.10 Requires paragraph).
This topic was immediately deferred to SG1 for initial review, before coming back to this group.
LEWGI Review: Defer to SG1
basic_string
capacity [depr.string.capacity]First deprecated: C++20
This feature was first deprecated for C++20 by the paper
P0966R1.
The deprecation was a consequence of cleaning up the behavior of the
reserve
function to no longer optionally reallocate on a request
to shrink. The original C++98 specification for basic_string
supplied a default argument of 0 for reserve
, turning a call to
reserve()
into a non-binding shrink_to_fit
request.
Note that shrink_to_fit
was added in C++11 to better support this
use case. With the removal of the potentially reallocating behavior,
reserve()
is now a redundant function overload that is guaranteed
to do nothing. Hence it was deprecated in C++20, with a view to removing it
entirely in a later standard to eliminate on more legacy source of confusion
from the standard.
As the feature was deprecated so recently, the strong recommendation of this paper is to make no changes for C++23, but strongly consider removal when it is time to review again for C++26. The weak recommendation is that this feature is small and obscure enough that it is better to remove now from C++23 than preserve for another three years into C++26.
Strong recommendation: Take no action.
No change to draft.
Weak recommendation: Remove this feature from C++23
D.19 Deprecated basic_string capacity [depr.string.capacity]
The following member is declared in addition to those members specified in 21.3.2.4:namespace std {template<class charT, class traits = char_traits<charT>,class Allocator = allocator<charT>>class basic_string {public:void reserve();};}void reserve();Effects: After this call,capacity()
has an unspecified value greater than or equal tosize()
. [Note: This is a non-binding shrink to fit request. —end note]
Draft compatibility note for Annex C.
General agreement that this member is a holdover from another time, whose replacement has been in place for some time. Consensus to remove from C++23, assuming the subsequent research does not reveal major concerns before the main LEWG review that is to follow.
LEWGI Review: Remove this feature from C++23
First deprecated: C++17
This feature was originally proposed for C++11 by paper N2007 and deprecated for C++17 by paper P0618R0. As noted at the time, the feature was underspecified and would require more work than we wished to invest to bring it up to standard. Since then SG16 has been convened and is producing a steady stream of work to bring reliable well-specified Unicode support to C++.
It should also be noted that this deprecated clause pins a dated reference to a 20 year old ISO standard (revised repeatedly over the intervening decades) purely to provide a definition of the term UCS2.
Given vendors propensity to provide ongoing support for these names under the zombie name reservations, the strong recommendation of this paper is to remove this library immediately from C++23, along with its binding reference to an obsolete Unicode standard. The weak recommendation is to do nothing at this point, until SG16 (or some other entity) produces a clean replacement for this facility.
We note that this feature was originally added at the Kona 2007 meeting so that (while motivated by likely user applications) the example in the then recently added [lib.conversions] called on standard rather than user-provided classes to illustrate use (adopted just one meeting earlier). Therefore, if we remove this library unilaterally, we should also revert that example back to its original spelling.
Strong recommendation: Remove this facility from the standard at the earliest opportunity.
2 Normative references [intro.refs]
- The following documents are referred to in the text in such a way that some or all of their content constitutes requirements of this document. For dated references, only the edition cited applies. For undated references, the latest edition of the referenced document (including any amendments) applies.
- — Ecma International, ECMAScript Language Specification, Standard Ecma-262, third edition, 1999.
- — ISO/IEC 2382 (all parts), Information technology — Vocabulary
- — ISO 8601:2004, Data elements and interchange formats — Information interchange — Representation of dates and times
- — ISO/IEC 9899:2018, Programming languages — C
- — ISO/IEC 9945:2003, Information Technology — Portable Operating System Interface (POSIX)
- — ISO/IEC 10646, Information technology — Universal Coded Character Set (UCS)
— ISO/IEC 10646-1:1993, Information technology — Universal Multiple-Octet Coded Character Set (UCS) — Part 1: Architecture and Basic Multilingual Plane- — ISO/IEC/IEEE 60559:2011, Information technology — Microprocessor Systems — Floating-Point arithmetic
- — ISO 80000-2:2009, Quantities and units — Part 2: Mathematical signs and symbols to be used in the natural sciences and technology
- The library described in Clause 7 of ISO/IEC 9899:2018 is hereinafter called the C standard library.1
- The operating system interface described in ISO/IEC 9945:2003 is hereinafter called POSIX.
- The ECMAScript Language Specification described in Standard Ecma-262 is hereinafter called ECMA-262.
[Note: References to ISO/IEC 10646-1:1993 are used only to support deprecated features (D.19). — end note ]16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
codecvt_mode
,- —
codecvt_utf16
,- —
codecvt_utf8
,- —
codecvt_utf8_utf16
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
consume_header
,- —
generate_header
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
little_endian
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.- The following names are reserved as member types for previous standardization, and may not be used as a name for object-like macros in portable code:
- —
argument_type
,- —
first_argument_type
,- —
io_state
,- —
open_mode
,- —
second_argument_type
, and- —
seek_dir
.- The name
stossc
is reserved as a member function for previous standardization, and may not be used as a name for function-like macros in portable code.- The header names
<ccomplex>
,<ciso646>
,<codecvt>
,<cstdalign>
,<cstdbool>
, and<ctgmath>
are reserved for previous standardization.
D.20 Deprecated Standard code conversion facets [depr.locale.stdcvt]
The header<codecvt>
provides code conversion facets for various character encodings.
D.20.1 Header<codecvt>
synopsis [depr.codecvt.syn]namespace std {enum codecvt_mode {consume_header = 4,generate_header = 2,little_endian = 1};template<class Elem, unsigned long Maxcode = 0x10ffff,codecvt_mode Mode = (codecvt_mode)0>class codecvt_utf8: public codecvt<Elem, char, mbstate_t> {public:explicit codecvt_utf8(size_t refs = 0);~codecvt_utf8();};template<class Elem, unsigned long Maxcode = 0x10ffff,codecvt_mode Mode = (codecvt_mode)0>class codecvt_utf16: public codecvt<Elem, char, mbstate_t> {public:explicit codecvt_utf16(size_t refs = 0);~codecvt_utf16();};template<class Elem, unsigned long Maxcode = 0x10ffff,codecvt_mode Mode = (codecvt_mode)0>class codecvt_utf8_utf16: public codecvt<Elem, char, mbstate_t> {public:explicit codecvt_utf8_utf16(size_t refs = 0);~codecvt_utf8_utf16();};}
D.20.2 Requirements [depr.locale.stdcvt.req]
For each of the three code conversion facetscodecvt_utf8
,codecvt_utf16
, andcodecvt_utf8_utf16
:
—Elem
is the wide-character type, such aswchar_t
,char16_t
, orchar32_t
.—Maxcode
is the largest wide-character code that the facet will read or write without reporting a conversion error.— If(Mode & consume_header)
, the facet shall consume an initial header sequence, if present, when reading a multibyte sequence to determine the endianness of the subsequent multibyte sequence to be read.— If(Mode & generate_header)
, the facet shall generate an initial header sequence when writing a multibyte sequence to advertise the endianness of the subsequent multibyte sequence to be written.— If(Mode & little_endian)
, the facet shall generate a multibyte sequence in little-endian order, as opposed to the default big-endian order.For the facetcodecvt_utf8
:
— The facet shall convert between UTF-8 multibyte sequences and UCS2 or UTF-32 (depending on the size ofElem
) within the program.— Endianness shall not affect how multibyte sequences are read or written.— The multibyte sequences may be written as either a text or a binary file.For the facetcodecvt_utf16
:
— The facet shall convert between UTF-16 multibyte sequences and UCS2 or UTF-32 (depending on the size ofElem
) within the program.— Multibyte sequences shall be read or written according to theMode
flag, as set out above.— The multibyte sequences may be written only as a binary file. Attempting to write to a text file produces undefined behavior.For the facetcodecvt_utf8_utf16
:
— The facet shall convert between UTF-8 multibyte sequences and UTF-16 (one or two 16-bit codes) within the program.— Endianness shall not affect how multibyte sequences are read or written.— The multibyte sequences may be written as either a text or a binary file.The encoding forms UTF-8, UTF-16, and UTF-32 are specified in ISO/IEC 10646. The encoding form UCS-2 is specified in ISO/IEC 10646-1:1993.D.21.1 Class template wstring_convert [depr.conversions.string]
- Class template
wstring_convert
performs conversions between a wide string and a byte string. It lets you specify a code conversion facet (like class templatecodecvt
) to perform the conversions, without affecting any streams or locales. [ Example: If youwant to use the code conversion facethave a code conversion facet namedcodecvt_utf8
codecvt_utf8
that you want to use to output tocout
a UTF-8 multibyte sequence corresponding to a wide string, but you don’t want to alter the locale forcout
, you can write something like:— end example ]std::wstring_convert<std::codecvt_utf8<wchar_t>> myconv; std::string mbstring = myconv.to_bytes(L"Hello\n"); std::cout << mbstring;
Draft compatibility note for Annex C.
Weak recommendation: take no action at this time.
No change to draft.
Agreement that this feature does not work well today, is underspecified, and dragging in a normative reference to withdrawn ISO documents. Any fix coming out of SG16 will be an entirely new facility, not a rehabilitation of this header. Consensus to remove this library from C++23.
SG16 agree with the urgency to remove the dependency on an obsolete standard. There was a suggestion that the removal could be more targeted to remove only the UCS2 parts, to avoid premature breakage of clients of the UTF-8/UTF-32 parts of this feature. However, there were more concerns raises about how badly underspecified the current feature is, and any attempt to rehabilitate it would amount to recreating the (needed) replacement feature - notably to properly specify error handling and behavior on malformed inputs. Consensus is to rely on the zombie names features so that client code is unlikely to break (or at least, not become more broken) on the release of C++23 implementations, but transfer responsibility for handling all of the undocumented and underspecified behavior that might impact portability onto the library vendors.
LEWGI Review: Following check with SG16 for concerns, recommend removal of this feature from C++23.
First deprecated: C++17
This feature was originally proposed for C++11 by paper N2401 and deprecated for C++17 by paper P0618R0. As noted at the time, the feature was underspecified and would require more work than we wished to invest to bring it up to standard. Since then SG16 has been convened and is producing a steady stream of work to bring reliable well-specified Unicode support to C++.
Given vendors propensity to provide ongoing support for these names under the zombie name reservations, the strong recommendation of this paper is to remove this library immediately from the C++23 standard. The weak recommendation is to do the minimal work to clean up the wording to use the more precise terms that replaced Requires clauses, waiting until SG16 (or some other entity) produces a clean replacement for this facility for users to migrate to before removal.
Strong recommendation: Remove this facility from the standard at the earliest opportunity.
16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
,and- —
unexpected_handler
.,- —
wbuffer_convert
, and- —
wstring_convert
.- The following names are reserved as member types for previous standardization, and may not be used as a name for object-like macros in portable code:
- —
argument_type
,- —
first_argument_type
,- —
io_state
,- —
open_mode
,- —
second_argument_type
, and- —
seek_dir
.The nameThe following names are reserved as member functions for previous standardization, and may not be used as a name for function-like macros in portable code:stossc
is reserved as a member function for previous standardization, and may not be used as a name for function-like macros in portable code.
- —
converted
,- —
from_bytes
,- —
stossc
, and- —
to_bytes
.- The header names
<ccomplex>
,<ciso646>
,<cstdalign>
,<cstdbool>
, and<ctgmath>
are reserved for previous standardization.
D.21 Deprecated convenience conversion interfaces [depr.conversions]
The header<locale>
(28.2) has the following additions:namespace std {template<class Codecvt, class Elem = wchar_t,class WideAlloc = allocator<Elem>,class ByteAlloc = allocator<char>>class wstring_convert;template<class Codecvt, class Elem = wchar_t,class Tr = char_traits<Elem>class wbuffer_convert;}
D.21.1 Class template wstring_convert [depr.conversions.string]
Class templatewstring_convert
performs conversions between a wide string and a byte string. It lets you specify a code conversion facet (like class templatecodecvt
) to perform the conversions, without affecting any streams or locales. [ Example: If you want to use the code conversion facetcodecvt_utf8
to output tocout
a UTF-8 multibyte sequence corresponding to a wide string, but you don’t want to alter the locale forcout
, you can write something like:— end example ]wstring_convert<std::codecvt_utf8<wchar_t>> myconv;std::string mbstring = myconv.to_bytes(L"Hello\n");std::cout << mbstring;namespace std {template <class Codecvt, class Elem = wchar_t,class WideAlloc = allocator<Elem>,class ByteAlloc = allocator<char>>class wstring_convert {public:using byte_string = basic_string<char, char_traits<char>, ByteAlloc>;using wide_string = basic_string<Elem, char_traits<Elem>, WideAlloc>;using state_type = typename Codecvt::state_type;using int_type = typename wide_string::traits_type::int_type;wstring_convert() : wstring_convert(new Codecvt) {}explicit wstring_convert(Codecvt* pcvt = new Codecvt);wstring_convert(Codecvt* pcvt, state_type state);explicit wstring_convert(const byte_string& byte_err,const wide_string& wide_err = wide_string());~wstring_convert();wstring_convert(const wstring_convert&) = delete;wstring_convert& operator=(const wstring_convert&) = delete;wide_string from_bytes(char byte);wide_string from_bytes(const char* ptr);wide_string from_bytes(const byte_string& str);wide_string from_bytes(const char* first, const char* last);byte_string to_bytes(Elem wchar);byte_string to_bytes(const Elem* wptr);byte_string to_bytes(const wide_string& wstr);byte_string to_bytes(const Elem* first, const Elem* last);size_t converted() const noexcept;state_type state() const;private:byte_string byte_err_string; // Exposition onlywide_string wide_err_string; // Exposition onlyCodecvt* cvtptr; // Exposition onlystate_type cvtstate; // Exposition onlysize_t cvtcount; // Exposition only};}The class template describes an object that controls conversions between wide string objects of classbasic_string<Elem, char_traits<Elem>, WideAlloc>
and byte string objects of classbasic_string<char, char_traits<char>, ByteAlloc>
. The class template defines the typeswide_string
andbyte_string
as synonyms for these two types. Conversion between a sequence ofElem
values (stored in awide_string
object) and multibyte sequences (stored in abyte_string
object) is performed by an object of classCodecvt
, which meets the requirements of the standard code-conversion facetcodecvt<Elem, char, mbstate_t>
.An object of this class template stores:
—byte_err_string
— a byte string to display on errors—wide_err_string
— a wide string to display on errors—cvtptr
— a pointer to the allocated conversion object (which is freed when thewstring_convert
object is destroyed)—cvtstate
— a conversion state object—cvtcount
— a conversion countusing byte_string = basic_string<char, char_traits<char>, ByteAlloc>;The type shall be a synonym forbasic_string<char, char_traits<char>, ByteAlloc>
size_t converted() const noexcept;Returns:cvtcount
.wide_string from_bytes(char byte);wide_string from_bytes(const char* ptr);wide_string from_bytes(const byte_string& str);wide_string from_bytes(const char* first, const char* last);Effects: The first member function shall convert the single-element sequencebyte
to a wide string. The second member function shall convert the null-terminated sequence beginning atptr
to a wide string. The third member function shall convert the sequence stored instr
to a wide string. The fourth member function shall convert the sequence defined by the range[first, last)
to a wide string.In all cases:
— If thecvtstate
object was not constructed with an explicit value, it shall be set to its default value (the initial conversion state) before the conversion begins. Otherwise it shall be left unchanged.— The number of input elements successfully converted shall be stored incvtcount
.Returns: If no conversion error occurs, the member function shall return the converted wide string. Otherwise, if the object was constructed with a wide-error string, the member function shall return the wide-error string. Otherwise, the member function throws an object of classrange_error
.using int_type = typename wide_string::traits_type::int_type;The type shall be a synonym forwide_string::traits_type::int_type
.state_type state() const;returnscvtstate
.using state_type = typename Codecvt::state_type;The type shall be a synonym forCodecvt::state_type
.byte_string to_bytes(Elem wchar);byte_string to_bytes(const Elem* wptr);byte_string to_bytes(const wide_string& wstr);byte_string to_bytes(const Elem* first, const Elem* last);Effects: The first member function shall convert the single-element sequencewchar
to a byte string. The second member function shall convert the null-terminated sequence beginning atwptr
to a byte string. The third member function shall convert the sequence stored inwstr
to a byte string. The fourth member function shall convert the sequence defined by the range[first, last)
to a byte string.In all cases:
— If thecvtstate
object was not constructed with an explicit value, it shall be set to its default value (the initial conversion state) before the conversion begins. Otherwise it shall be left unchanged.— The number of input elements successfully converted shall be stored incvtcount
.Returns: If no conversion error occurs, the member function shall return the converted byte string. Otherwise, if the object was constructed with a byte-error string, the member function shall return the byte-error string. Otherwise, the member function shall throw an object of classrange_error
.using wide_string = basic_string<Elem, char_traits<Elem>, WideAlloc>;The type shall be a synonym forbasic_string<Elem, char_traits<Elem>, WideAlloc>
.explicit wstring_convert(Codecvt* pcvt = new Codecvt);wstring_convert(Codecvt* pcvt, state_type state);explicit wstring_convert(const byte_string& byte_err,const wide_string& wide_err = wide_string());Requires: For the first and second constructors,pcvt != nullptr
.Effects: The first constructor shall storepcvt
incvtptr
and default values incvtstate
,byte_err_string
, andwide_err_string
. The second constructor shall storepcvt
incvtptr
,state
incvtstate
, and default values inbyte_err_string
andwide_err_string
; moreover the stored state shall be retained between calls tofrom_bytes
andto_bytes
. The third constructor shall storenew Codecvt
incvtptr
,state_type()
incvtstate
,byte_err
inbyte_err_string
, andwide_err
inwide_err_string
.~wstring_convert();Effects: The destructor shall deletecvtptr
.
D.21.2 Class template wbuffer_convert [depr.conversions.buffer]
Class templatewbuffer_convert
looks like a wide stream buffer, but performs all its I/O through an underlying byte stream buffer that you specify when you construct it. Like class templatewstring_convert
, it lets you specify a code conversion facet to perform the conversions, without affecting any streams or locales.namespace std {template <class Codecvt, class Elem = wchar_t, class Tr = char_traits<Elem>>class wbuffer_convert: public basic_streambuf<Elem, Tr> {public:using state_type = typename Codecvt::state_type;wbuffer_convert() : wbuffer_convert(nullptr) {}explicit wbuffer_convert(streambuf* bytebuf = 0,Codecvt* pcvt = new Codecvt,state_type state = state_type());~wbuffer_convert();wbuffer_convert(const wbuffer_convert&) = delete;wbuffer_convert& operator=(const wbuffer_convert&) = delete;streambuf* rdbuf() const;streambuf* rdbuf(streambuf* bytebuf);state_type state() const;private:streambuf* bufptr; // exposition onlyCodecvt* cvtptr; // exposition onlystate_type cvtstate; // exposition only};}The class template describes a stream buffer that controls the transmission of elements of typeElem
, whose character traits are described by the classTr
, to and from a byte stream buffer of typestreambuf
. Conversion between a sequence ofElem
values and multibyte sequences is performed by an object of classCodecvt
, which shall meet the requirements of the standard code-conversion facetcodecvt<Elem, char, mbstate_t>
.An object of this class template stores:
—bufptr
— a pointer to its underlying byte stream buffer—cvtptr
— a pointer to the allocated conversion object (which is freed when thewbuffer_convert
object is destroyed)—cvtstate
— a conversion state objectstate_type state() const;Returns:cvtstate
.streambuf* rdbuf() const;Returns:bufptr
.streambuf* rdbuf(streambuf* bytebuf);Effects: Storesbytebuf
inbufptr
.Returns: The previous value ofbufptr
.using state_type = typename Codecvt::state_type;The type shall be a synonym forCodecvt::state_type
.explicit wbuffer_convert(streambuf* bytebuf = 0,Codecvt* pcvt = new Codecvt, state_type state = state_type());Requires:pcvt != nullptr
.Effects: The constructor constructs a stream buffer object, initializesbufptr
tobytebuf
, initializescvtptr
topcvt
, and initializescvtstate
tostate
.~wbuffer_convert();Effects: The destructor shall deletecvtptr
.
Draft compatibility note for Annex C.
Weak recommendation: Clean up wording.
Update wording to replace Requires clauses (see D.10 Requires paragraph).
Discussion was broadly in favor of removal from the C++23 specification, and relying on library vendors to maintain source compatibility as long as needed. However, LEWGI explicitly requested to confer with SG16 in case that study group is aware of any reason to hold back on removal, before proceeding with the recommendation.
SG16 raises concerns that the original paper deprecating this feature lacked a
strong motivation, as it was simply recording a recommendation from LWG review
when deprecating the <codecvt>
header for D.20. There is
general concern that the codecvt
is not fit for purpose, notably
due to poorly specified error handling capabilities while transcoding, but this
deprecation does not address that, it is merely a convenience API for using
that underspecified library component. While removing the
<codecvt>
header might mean there would be fewer standard
codecvt
facets to use with this API, it remains just a as usable
with user provided codecvt
facets as before, as well as those in
the <locale>
header. While we would like to see a
replacement facility, there is no such proposal at this time.
Polling showed no consensus to recommend the removal for C++23, but no objection to that removal either.
LEWGI Review: Confirmed SG16 has no objection; remove this feature from C++23
First deprecated: C++20
This feature was added as part of the initial basic support for Unicode types in C++11 by paper N2238 and deprecated on the recommendation of SG16 for C++20 by paper P0482R6.
As SG16 do not report any urgent issue relating to this deprecated feature, and are still working through the process of providing clean Unicode support in the C++ standard library, and given the deprecation is as recent as C++20, both the strong and weak recommendations are to take no action on this feature at this time.
Strong recommendation: Take no action.
No change to draft.
Weak recommendation: Take no action.
No change to draft.
Discussion was broadly in favor of removal from the C++23 specification, and relying on library vendors to maintain source compatibility as long as needed. However, LEWGI explicitly requested to confer with SG16 in case that study group is aware of any reason to hold back on removal, before proceeding with the recommendation.
SG16 is concerned that code_cvt
in general has poor error handling
facilities, especially when dealing with encodings that may take multiple code
units to express a code point, and so have more cause to report on malformed
inputs. The specific facets in D.22 stand in the way of putting a minimally
useful replacement into the standard, by sitting on the good names but with
poor semantics. This means that the usual safety net of the zombie names
clause does not apply, as we will want code to not compile for at least one
release in order to introduce a replacement with the same names but more
appropriate semantics. Concern was raised about removing a feature deprecated
only as recently as the current C++20 standard.
Polling showed no consensus to recommend the removal for C++23, but no objection to that removal either.
LEWGI Review: Confirmed SG16 has no objection; remove this feature from C++23
D.22 Deprecated locale category facets [depr.locale.category]
Thectype
locale category includes the following facets as if they were specified in table Table 102 of 28.3.1.1.1.codecvt<char16_t, char, mbstate_t>codecvt<char32_t, char, mbstate_t>Thectype
locale category includes the following facets as if they were specified in table Table 103 of 28.3.1.1.1.codecvt_byname<char16_t, char, mbstate_t>codecvt_byname<char32_t, char, mbstate_t>The following class template specializations are required in addition to those specified in 28.4.1.4. The specializationcodecvt<char16_t, char, mbstate_t>
converts between the UTF-16 and UTF-8 encoding forms, and the specializationcodecvt<char32_t, char, mbstate_t>
converts between the UTF-32 and UTF-8 encoding forms.
Draft compatibility note for Annex C.
First deprecated: C++20
A factory function to create path names from UTF-8 sequences was part of the
original filesystem library adopted for C++17. However, this was the only
string-based factory function, as the preferred interface is to simply
construct a path
with a string of the corresponding type/encoding.
This factory function was deprecated in C++20 with the addition of
char8_t
and the ability to now invoke a specific constructor for
UTF-8 encoded (and typed) strings. See
P0482R6
for details.
The legacy API continues to function, but is more cumbersome than necessary. There appears to be no compelling case that the API is a risk through misuse. Therefore, given it was so recently deprecated, the strong recommendation is to retain this feature in Annex D for C++23, giving the community time to catch up, and consider removal again for C++26. However, the current wording does not follow library best practices, and should be updated to better specify the Requires clauses.
However, while it does no active harm, there is always a cost to maintaining text in the standard. The application of zombie names means that even if we remove this clause from Annex D in C++23, standard library vendors are likely to continue shipping to meet customer demand for some time to come. So the weak recommendation is to add the names to the zombie clause, and remove immediately from C++23.
Strong recommendation: Clean up wording.
Update wording to replace Requires clauses (see D.10 Requires paragraph).
Weak recommendation: Remove this feature from C++23
16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
u8path
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.
D.23 Deprecated filesystem path factory functions [depr.fs.path.factory]
template<class Source>path u8path(const Source& source);template<class InputIterator>path u8path(InputIterator first, InputIterator last);Requires: Thesource
and[first, last)
sequences are UTF-8 encoded. The value type ofSource
andInputIterator
ischar
orchar8_t
.Source
meets the requirements specified in 29.11.7.3.Returns:
— Ifvalue_type
ischar
and the current native narrow encoding (29.11.7.2.2) is UTF-8, returnpath(source)
orpath(first, last)
; otherwise,— ifvalue_type
iswchar_t
and the native wide encoding is UTF-16, or ifvalue_type
ischar16_t
orchar32_t
, convertsource
or[first, last)
to a temporary,tmp
, of typestring_type
and returnpath(tmp)
; otherwise,— convertsource
or[first, last)
to a temporary,tmp
, of typeu32string
and returnpath(tmp)
.Remarks: Argument format conversion (29.11.7.2.1) applies to the arguments for these functions. How Unicode encoding conversions are performed is unspecified.[Example: A string is to be read from a database that is encoded in UTF-8, and used to create a directory using the native encoding for filenames:namespace fs = std::filesystem;std::string utf8_string = read_utf8_data();fs::create_directory(fs::u8path(utf8_string));For POSIX-based operating systems with the native narrow encoding set to UTF-8, no encoding or type conversion occurs.
For POSIX-based operating systems with the native narrow encoding not set to UTF-8, a conversion to UTF-32 occurs, followed by a conversion to the current native narrow encoding. Some Unicode characters may have no native character set representation.For Windows-based operating systems a conversion from UTF-8 to UTF-16 occurs. —end example] [Note: The example above is representative of a historical use offilesystem::u8path
. Passing astd::u8string
topath
’s constructor is preferred for an indication of UTF-8 encoding more consistent withpath
’s handling of other encodings. —end note]
Draft compatibility note for Annex C.
Discussion was broadly in favor of removal from the C++23 specification, and relying on library vendors to maintain source compatibility as long as needed. However, LEWGI explicitly requested to confer with SG16 in case that study group is aware of any reason to hold back on removal, before proceeding with the recommendation.
SG16 is not persuaded that this is actually a text issue, as filenames in practice accept a broader definition of text, and treating them as straight text is often a source of subtle and surprising problems. There was mild concern at the idea of removing something deprecated as recently as C++20, but it was also noted that the deprecating paper was adopted into the working draft before at least one major library distribution had provided its initial implementation! No objection to the removal, if anything a mild encouragement to proceed!
LEWGI Review: Following check with SG16 for concerns, recommend removal of this feature from C++23.
First deprecated: C++20
The original API to initialize atomic variables from C++11 was deprecated
for C++20 when the atomic
template was given a default constructor to
do the right thing. See
P0883R2
for details.
The legacy API continues to function, but is more cumbersome than necessary. There appears to be no compelling case that the API is a risk through misuse. Therefore, given it was so recently deprecated, the strong recommendation is to retain this feature in Annex D for C++23, giving the community time to catch up, and consider removal again for C++26.
While it does no active harm, there is always a cost to maintaining text in the
standard. This is similarly reflected in the C standard, that initially
deprecated the ATOMIC_VAR_INT
macro (marked it as obsolescent) in
C17, and is actively looking to remove it from the C2X standard, per the paper
WG14:N2390.
It seems likely that the next C standard will ship in the timeframe of C++23,
and will no longer define this macro. Note that C provides this macro in the
<stdatomic.h>
header, which is deliberately excluded from
the set of C headers defined for use by the C++ standard. For the C++
standard, the application of zombie names means that even if we remove this
clause from Annex D in C++23, standard library vendors are likely to continue
shipping to meet customer demand for some time to come. So the weak
recommendation is to add the names to the zombie clause, and remove immediately
from C++23.
Additionally, the volatile
qualified member functions of the
atomic
class template were deprecated for C++20 by paper
P1831R1.
Both the strong and weak recommendations are to leave this feature alone until
the interaction with the non-deprecated non-member functions in the
<atomic>
header that take pointer-to-volatile-qualified type.
Possible directions would be to deprecate those non-member functions too, or to
undeprecate the volatile-qualified member functions.
Strong recommendation: Take no action. Reconsider for C++26.
No change to draft.
Weak recommendation: Remove deprecated atomic initialization feature from C++23
16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
ATOMIC_FLAG_INIT
,- —
atomic_init
,- —
ATOMIC_VAR_INIT
,- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.D.24 Deprecated atomic initialization [depr.atomics]
The header<atomics>
(31.2) has the following additions.namespace std {template<class T>void atomic_init(volatile atomic<T>*, typename atomic<T>::value_type) noexcept;template<class T>void atomic_init(atomic<T>*, typename atomic<T>::value_type) noexcept;#define ATOMIC_VAR_INIT(value) see below#define ATOMIC_FLAG_INIT see below}D.24.1 Volatile access [depr.atomics.volatile]
If an atomic specialization has one of the following overloads, then that overload participates in overload resolution even if
atomic<T>::is_always_lock_free
isfalse
:void store(T desired, memory_order order = memory_order::seq_cst) volatile noexcept; T operator=(T desired) volatile noexcept; T load(memory_order order = memory_order::seq_cst) const volatile noexcept; operator T() const volatile noexcept; T exchange(T desired, memory_order order = memory_order::seq_cst) volatile noexcept; bool compare_exchange_weak(T& expected, T desired, memory_order success, memory_order failure) volatile noexcept; bool compare_exchange_strong(T& expected, T desired, memory_order success, memory_order failure) volatile noexcept; bool compare_exchange_weak(T& expected, T desired, memory_order order = memory_order::seq_cst) volatile noexcept; bool compare_exchange_strong(T& expected, T desired, memory_order order = memory_order::seq_cst) volatile noexcept; T fetch_key(T operand, memory_order order = memory_order::seq_cst) volatile noexcept; T operator op=(T operand) volatile noexcept; T* fetch_key(ptrdiff_t operand, memory_order order = memory_order::seq_cst) volatile noexcept;
D.24.2 Non-member functions [depr.atomics.nonmembers]template<class T>void atomic_init(volatile atomic<T>* object, typename atomic<T>::value_type desired) noexcept;template<class T>void atomic_init(atomic<T>* object, typename atomic<T>::value_type desired) noexcept;
Effects: Equivalent to:atomic_store_explicit(object, desired, memory_order::relaxed);
D.24.3 Operations on atomic types [depr.atomics.types.operations]#define ATOMIC_VAR_INIT(value) see below
The macro expands to a token sequence suitable for constant initialization of an atomic variable of static storage duration of a type that is initialization-compatible withvalue
. [Note: This operation may need to initialize locks. —end note] Concurrent access to the variable being initialized, even via an atomic operation, constitutes a data race. [Example:—end example]atomic<int> v = ATOMIC_VAR_INIT(5);
D.24.4 Flag type and operations [depr.atomics.flag]#define ATOMIC_FLAG_INIT see below
Remarks: The macroATOMIC_FLAG_INIT
is defined in such a way that it can be used to initialize an object of typeatomic_flag
to the clear state. The macro can be used in the form:It is unspecified whether the macro can be used in other initialization contexts. For a complete static-duration object, that initialization shall be static.atomic_flag guard = ATOMIC_FLAG_INIT;
Draft compatibility note for Annex C.
This topic was immediately deferred to SG1 for initial review, before coming back to this group.
LEWGI Review: Defer to SG1
While practicing good housekeeping and clearing out Annex D for each release may be the preferred option, there are other approaches that may be taken.
One approach, epitomized in the Java language, is that deprecated features are discouraged for future use, but guaranteed to remain available forever, and just accumulate.
This approach is rejected by this paper for a number of reasons. First, C++ has been relatively successful in actually removing its deprecated features in the past, a tradition we would like to continue. It also undercuts the available-forever rationale, as it is not a guarantee we have given before.
A second concern is that we do not want to pay a cost to maintain deprecated components forever - restricting growth of the language for compatibility with deprecated features, or having to review the whole of Annex D and upgrade components for every new language release, in order to keep up with subtle shifts in the core language.
If something is deprecated, but later (re)discovered to have value, then it could be revitalized and restored to the main standard. For example, this is exactly what happened to static function declarations when the unnamed namespace was given internal linkage - it is merely the classical way to say the same thing, and often clearer to write.
This may be a consideration for long-term deprecated features that don't appear
to be going anywhere, such as the strstream
facility, or the C
headers. It may be appropriate to find them a home in the regular standard,
and this is called out in the specific reviews of each facility above.
This section integrates the proposed wording of the above recommendations as the various working groups approve them for review by the Core and Library Working Groups. It will apply the wording for mandating Annex D if the recommendation is to retain a facility.
Collected summary of recommendations:
Subclause | Feature | Adopted Recommendation | Action |
D.1 | Arithmetic conversion on enumerations | ||
D.2 | Implicit capture of *this by reference |
||
D.3 | Comma operator in subscript expressions | ||
D.4 | Array comparisons | ||
D.5 | Deprecated use of volatile |
||
D.6 | Reclare constexpr members |
||
D.7 | Non-local use of TU-local entities | ||
D.8 | Implicit special members | ||
D.9 | C <*.h> headers |
||
D.10 | Requires: clauses | ||
D.11 | relops |
||
D.12 | char * streams |
||
D.13 | Deprecated type traits | ||
D.14 | volatile tuple traits |
||
D.15 | volatile variant traits |
||
D.16 | std::iterator |
||
D.17 | move_iterator::operator-> |
||
D.18 | C API to use shared_ptr atomically |
||
D.19 | basic_string::reserve() |
||
D.20 | <codecvt> |
||
D.21 | wstring_convert et al. |
||
D.22 | Deprecated locale category facets | ||
D.23 | filesystem::u8path |
||
D.24 | atomic operations |
Full wording follows:
Wording will accumulate (L)EWG recommendations when approved.
2 Normative references [intro.refs]
- The following documents are referred to in the text in such a way that some or all of their content constitutes requirements of this document. For dated references, only the edition cited applies. For undated references, the latest edition of the referenced document (including any amendments) applies.
- — Ecma International, ECMAScript Language Specification, Standard Ecma-262, third edition, 1999.
- — ISO/IEC 2382 (all parts), Information technology — Vocabulary
- — ISO 8601:2004, Data elements and interchange formats — Information interchange — Representation of dates and times
- — ISO/IEC 9899:2018, Programming languages — C
- — ISO/IEC 9945:2003, Information Technology — Portable Operating System Interface (POSIX)
- — ISO/IEC 10646, Information technology — Universal Coded Character Set (UCS)
— ISO/IEC 10646-1:1993, Information technology — Universal Multiple-Octet Coded Character Set (UCS) — Part 1: Architecture and Basic Multilingual Plane- — ISO/IEC/IEEE 60559:2011, Information technology — Microprocessor Systems — Floating-Point arithmetic
- — ISO 80000-2:2009, Quantities and units — Part 2: Mathematical signs and symbols to be used in the natural sciences and technology
- The library described in Clause 7 of ISO/IEC 9899:2018 is hereinafter called the C standard library.1
- The operating system interface described in ISO/IEC 9945:2003 is hereinafter called POSIX.
- The ECMAScript Language Specification described in Standard Ecma-262 is hereinafter called ECMA-262.
[Note: References to ISO/IEC 10646-1:1993 are used only to support deprecated features (D.19). — end note ]7.4 Usual arithmetic conversions [expr.arith.conv]
- Many binary operators that expect operands of arithmetic or enumeration type cause conversions and yield result types in a similar way. The purpose is to yield a common type, which is also the type of the result. This pattern is called the usual arithmetic conversions, which are defined as follows:
- — If either operand is of scoped enumeration type (9.7.1), no conversions are performed; if the other operand does not have the same type, the expression is ill-formed.
- — Otherwise, if either operand is of unscoped enumeration type (9.7.1) and the other operand is of a floating-point type, the expression is ill-formed.
- — Otherwise, if either operand is of type
long double
, the other shall be converted tolong double
.- — Otherwise, if either operand is
double
, the other shall be converted todouble
.- — Otherwise, if either operand is
float
, the other shall be converted tofloat
.- — Otherwise, the integral promotions (7.3.6) shall be performed on both operands.56 Then the following rules shall be applied to the promoted operands:
- — If both operands have the same type, no further conversion is needed.
- — Otherwise, if both operands have signed integer types or both have unsigned integer types, the operand with the type of lesser integer conversion rank shall be converted to the type of the operand with greater rank.
- — Otherwise, if the operand that has unsigned integer type has rank greater than or equal to the rank of the type of the other operand, the operand with signed integer type shall be converted to the type of the operand with unsigned integer type.
- — Otherwise, if the type of the operand with signed integer type can represent all of the values of the type of the operand with unsigned integer type, the operand with unsigned integer type shall be converted to the type of the operand with signed integer type.
- — Otherwise, both operands shall be converted to the unsigned integer type corresponding to the type of the operand with signed integer type.
- If
oneeither operand is of unscoped enumeration type and the other operand is of a different enumeration typeor a floating-point type, this behavior is deprecated (D.1).7.6.1.1 Subscripting [expr.sub]
[Note:A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression is ill-formed.deprecated; see D.3. —end note]7.6.9 Relational operators [expr.rel]
- The relational operators group left-to-right. [Example:
a<b<c
means(a<b)<c
and not(a<b)&&(b<c)
. — end example]The lvalue-to-rvalue (7.3.1)relational-expression : compare-expression relational-expression < compare-expression relational-expression > compare-expression relational-expression <= compare-expression relational-expression >= compare-expression, array-to-pointer (7.3.2),and function-to-pointer(7.3.3) standard conversions are performed on the operands. If at least one of the operands is a pointer, array-to-pointer conversions (7.3.2) are performed.The comparison is deprecated if both operands were of array type prior to these conversions (D.4).- The converted operands shall have arithmetic, enumeration, or pointer type. The operators
<
(less than),>
(greater than),<=
(less than or equal to), and>=
(greater than or equal to) all yieldfalse
ortrue
. The type of the result isbool
.- The usual arithmetic conversions (7.4) are performed on operands of arithmetic or enumeration type. If both operands are pointers, pointer conversions (7.3.11) and qualification conversions (7.3.5) are performed to bring them to their composite pointer type (7.2.2). After conversions, the operands shall have the same type.
- The result of comparing unequal pointers to objects ...
7.6.10 Equality operators [expr.eq]
equality-expression : relational-expression equality-expression == relational-expression equality-expression != relational-expression
- The
==
(equal to) and the!=
(not equal to) operators group left-to-right. The lvalue-to-rvalue (7.3.1), array-to-pointer (7.3.2),and function-to-pointer (7.3.3) standard conversions are performed on the operands.The comparison is deprecated if both operands were of array type prior to these conversions (D.4).- If at least one of the operands is a pointer, array-to-pointer conversions (7.3.2), pointer conversions (7.3.11), function pointer conversions (7.3.13), and qualification conversions (7.3.5) are performed on both operands to bring them to their composite pointer type (7.2.2).
Comparing pointers is defined as follows:- The converted operands shall have arithmetic, enumeration, pointer, or pointer-to-member type, or type
std::nullptr_t
. The operators==
and!=
both yieldtrue
orfalse
, i.e., a result of typebool
. In each case below, the operands shall have the same type after the specified conversions have been applied.- Comparing pointers is defined as follows:
- — If one pointer represents the address of a complete object, and another pointer represents the address one past the last element of a different complete object,79 the result of the comparison is unspecified.
- — Otherwise, if the pointers are both null, both point to the same function, or both represent the same address (6.8.2), they compare equal.
- — Otherwise, the pointers compare unequal.
- If at least one of the operands is a pointer to member, ...
7.6.20 Comma operator [expr.comma]
- [Note: A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression is ill-formed (7.6.1.1)
deprecated; see D.3. —end note]16.5.4.3.1 Zombie names [zombie.names]
- In namespace
std
, the following names are reserved for previous standardization:
- —
auto_ptr
,- —
auto_ptr_ref
,- —
binary_function
,- —
binary_negate
,- —
bind1st
,- —
bind2nd
,- —
binder1st
,- —
binder2nd
,- —
codecvt_mode
,- —
codecvt_utf16
,- —
codecvt_utf8
,- —
codecvt_utf8_utf16
,- —
const_mem_fun1_ref_t
,- —
const_mem_fun1_t
,- —
const_mem_fun_ref_t
,- —
const_mem_fun_t
,- —
consume_header
,- —
generate_header
,- —
get_temporary_buffer
,- —
get_unexpected
,- —
gets
,- —
is_literal_type
,- —
is_literal_type_v
,- —
is_pod
,- —
is_pod_v
,- —
iterator
,- —
little_endian
,- —
mem_fun1_ref_t
,- —
mem_fun1_t
,- —
mem_fun_ref_t
,- —
mem_fun_ref
,- —
mem_fun_t
,- —
mem_fun
,- —
not1
,- —
not2
,- —
pointer_to_binary_function
,- —
pointer_to_unary_function
,- —
ptr_fun
,- —
random_shuffle
,- —
raw_storage_iterator
,- —
rel_ops
,- —
result_of
,- —
result_of_t
,- —
return_temporary_buffer
,- —
set_unexpected
,- —
u8path
,- —
unary_function
,- —
unary_negate
,- —
uncaught_exception
,- —
unexpected
, and- —
unexpected_handler
.- The following names are reserved as member types for previous standardization, and may not be used as a name for object-like macros in portable code:
- —
argument_type
,- —
first_argument_type
,- —
io_state
,- —
open_mode
,- —
second_argument_type
, and- —
seek_dir
.- The name
stossc
is reserved as a member function for previous standardization, and may not be used as a name for function-like macros in portable code.- The header names
<ccomplex>
,<ciso646>
,<codecvt>
,<cstdalign>
,<cstdbool>
, and<ctgmath>
are reserved for previous standardization.C.1 C++ and ISO C++ 2020
- This subclause lists the differences between C++ and ISO C++ 2020 (ISO/IEC 14882:2020, Programming Languages — C++), by the chapters of this document.
C.1.X Clause 7: Expressions [diff.cpp20.expr]
Affected subclause: 7.4 Note for editors: [expr.arith.conv]
Change: Unscoped enumerations do not implicitly promote to integral type in expressions with floating point types.
Rationale:
Provide rationaleEffect on original feature: A valid C++ 2020 involving both an unscoped enumeration and a floating point value will be rejected as ill-formed in this International Standard. Either argument could be explicitly conerted with a cast, or the enumeration could be explicitly promoted to an integer with unary operator +, for no change of meaning since C++ 2020. [Example:
enum multipliers { gigaseconds = 1'000'000 }; double century_ish = 3.14 * gigaseconds; // ill-formed; previously well-formed double century_est = 3.14 * +gigaseconds; // OK—end example]
Affected subclause: 7.6.1.1 Note for editors: [expr.sub]
Change: Cannot parse unparenthesized comma expressions as subscript expressions.
Rationale: The old behavior was surprising for users familiar with other languages. It was removed to allow a future extension to support overloading the subscript operator for multiple arguments.
Effect on original feature: A valid C++ 2020 program using a comma expression as the argument to a subscript expression will be rejected as ill-formed in this International Standard. The intended comma expression can be enclosed in parentheses for no change of meaning since C++ 2020. [Example:
void f(int *a, int b, int c) { a[b,c]; // ill-formed; previously well-formed a[(b,c)]; // OK }Add an example of a DSEL—end example]
Affected subclause: 7.6.9 and 7.6.10 Note for editors: [expr.rel] and [expr.eq]
Change: Cannot compare two objects of array type.
Rationale: The old behavior was confusing, as it did not compare the contents of the two arrays, but compare their addresses. Depending on context, this would either report whether the two arrays were the same object, or have an unspecified result.
Effect on original feature: A valid C++ 2020 program directly comparing two array objects will be rejected as ill-formed in this International Standard. [Example:
int arr1[5]; int arr2[5]; bool same = arr1 == arr2; // ill-formed; previously well-formed bool idem = arr1 == +arr2; // compare addresses, unspecified result—end example]
C.1.14 Annex D: compatibility features [diff.cpp17.depr]
Change: Remove namespace
std::rel_ops
.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove type trait
is_pod
.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove
volatile
specializations ofvariant
traits.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove
std::iterator
.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove
move_iterator::operator->
.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove
basic_string::reserve()
with no arguments.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove
<codecvt>
header.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakChange: Remove
u8_path
function.Rationale:
Provide rationaleEffect on original feature:
Provide examples of code that might breakC.5.3 Clause 7: expressions [diff.expr]
Affected subclause: 7.4 Note for editors: [expr.arith.conv]
Change: Enumerations cannot be used in expressions with floating point types.
Rationale:
Provide rationaleEffect on original feature: Deletion of semantically well-defined feature.
Difficulty of converting: Could be automated. Violations will be diagnosed by the C++ translator. The fix is to add a cast, or explicitly promote the
enum
object to an integer with unary operator+
. For example:How widely used: Common.
Affected subclause: 7.6.1.1 Note for editors: [expr.sub]
Change: Cannot parse unparenthesized comma expressions as subscript expressions.
Rationale: The old behavior was surprising for users familiar with other languages. It was removed to allow a future extension to support overloading the subscript operator for multiple arguments.
Effect on original feature: Deletion of semantically well-defined feature.
Difficulty of converting: Could be automated. Violations will be diagnosed by the C++ translator. The fix is to add parentheses. For example:
void f(int *a, int b, int c) { a[b,c]; // ill-formed; previously well-formed a[(b,c)]; // OK }How widely used: Rare.
Affected subclause: 7.6.9 and 7.6.10 Note for editors: [expr.rel] and [expr.eq]
Change: Cannot compare two objects of array type.
Rationale: The old behavior was confusing, as it did not compare the contents of the two arrays, but compare their addresses. Depending on context, this would either report whether the two arrays were the same object, or have an unspecified result.
Effect on original feature: Deletion of feature with unspecified behavior.
Difficulty of converting: Violations will be diagnosed by the C++ translator.
How widely used: Rare. In the cases where the result is well defined, it is reporting whether both arguments are the same object, using the same name.
D.1 Arithmetic conversion on enumerations [depr.arith.conv.enum]
- The ability to apply the usual arithmetic conversions (7.4) on operands where one is of unscoped enumeration type and the other is of a different enumeration type
or a floating-point typeis deprecated. [Note: Three-way comparisons (7.6.8) between such operands are ill-formed. —end note] [Example:—end example]enum E1 { e }; enum E2 { f };bool b = e <= 3.7; // deprecatedint k = f - e; // deprecated auto cmp = e <=> f; // error
D.3 Comma operator in subscript expressions [depr.comma.subscript]
A comma expression (7.6.20) appearing as the expr-or-braced-init-list of a subscripting expression (7.6.1.1) is deprecated. [Note: A parenthesized comma expression is not deprecated. —end note] [Example:void f(int *a, int b, int c) {a[b,c]; // deprecateda[(b,c)]; // OK}—end example]
D.4 Array comparisons [depr.array.comp]
Equality and relational comparisons (7.6.10, 7.6.9) between two operands of array type are deprecated. [Note: Three-way comparisons (7.6.8) between such operands are ill-formed. —end note] [Example:int arr1[5];int arr2[5];bool same = arr1 == arr2; // deprecated, same as &arr1[0] == &arr2[0],// does not compare array contentsauto cmp = arr1 <=> arr2; // error—end example]
D.11 Relational operators [depr.relops]
The header<utility>
(20.2.1) has the following additions:>namespace std::rel_ops {template<class T> bool operator!=(const T&, const T&);template<class T> bool operator> (const T&, const T&);template<class T> bool operator<=(const T&, const T&);template<class T> bool operator>=(const T&, const T&);>}To avoid redundant definitions ofoperator!=
out ofoperator==
and operators>
,<=
, and>=
out ofoperator<
, the library provides the following:template<class T> bool operator!=(const T& x, const T& y);Requires: TypeT
is Cpp17EqualityComparable (Table 23).Returns:!(x == y)
.template<class T> bool operator>(const T& x, const T& y);Requires: TypeT
is Cpp17LessThanComparable (Table 24).Returns:y < x
.template<class T> bool operator<=(const T& x, const T& y);Requires: TypeT
is Cpp17LessThanComparable (Table 24).Returns:!(y < x)
.template<class T> bool operator>=(const T& x, const T& y);Requires: TypeT
is Cpp17LessThanComparable (Table 24).Returns:!(x < y)
.
D.13 Deprecated Type Traits [depr.meta.type]
The header<type_traits>
(20.15.2) has the following additions:namespace std {template<class T> struct is_pod;template<class T> inline constexpr bool is_pod_v = is_pod<T>::value;}The behavior of a program that adds specializations for any of the templates defined in this subclause is undefined, unless explicitly permitted by the specification of the corresponding template.template<class T> struct is_pod;Requires:remove_all_extents_t<T>
shall be a complete type or cvvoid
.is_pod<T>
is a Cpp17UnaryTypeTrait (20.15.1) with a base characteristic oftrue_type
ifT
is a POD type, andfalse_type
otherwise. A POD class is a class that is both a trivial class and a standard-layout class, and has no non-static data members of type non-POD class (or array thereof). A POD type is a scalar type, a POD class, an array of such a type, or a cv-qualified version of one of these types.[Note: It is unspecified whether a closure type (7.5.5.1) is a POD type. —end note]
D.15 Variant [depr.variant]
The header<variant>
(20.7.2) has the following additions:namespace std {template<class T> struct variant_size<volatile T>;template<class T> struct variant_size<const volatile T>;template<size_t I, class T> struct variant_alternative<I, volatile T>;template<size_t I, class T> struct variant_alternative<I, const volatile T>;}template<class T> class variant_size<volatile T>;template<class T> class variant_size<const volatile T>;LetVS
denotevariant_size<T>
of the cv-unqualified typeT
. Then specializations of each of the two templates meet the Cpp17UnaryTypeTrait requirements with a base characteristic ofintegral_constant<size_t, VS::value>
.template<size_t I, class T> class variant_alternative<I, volatile T>;template<size_t I, class T> class variant_alternative<I, const volatile T>;LetVA
denotevariant_alternative<I, T>
of the cv-unqualified typeT
. Then specializations of each of the two templates meet the Cpp17TransformationTrait requirements with a member typedef type that names the following type:
— for the first specialization,add_volatile_t<VA::type>
, and— for the second specialization,add_cv_t<VA::type>
.
D.16 Deprecated iterator primitives [depr.iterator.primitives]
D.16.1 Basic iterator [depr.iterator.basic]
The header<iterator>
(23.2) has the following addition:namespace std {template<class Category, class T, class Distance = ptrdiff_t,class Pointer = T*, class Reference = T&>struct iterator {using iterator_category = Category;using value_type = T;using difference_type = Distance;using pointer = Pointer;using reference = Reference;};}Theiterator
template may be used as a base class to ease the definition of required types for new iterators.[ Note: If the new iterator type is a class template, then these aliases will not be visible from within the iterator class's template definition, but only to callers of that class — end note][ Example: If a C++ program wants to define a bidirectional iterator for some data structure containingdouble
and such that it works on a large memory model of the implementation, it can do so with:— end example ]class MyIterator :public iterator<bidirectional_iterator_tag, double, long, T*, T&> {// code implementing ++, etc.};
D.17 Deprecated move_iterator access [depr.move.iter.elem]
The following member is declared in addition to those members specified in 23.5.3.5:namespace std {template<class Iterator>class move_iterator {public:constexpr pointer operator->() const;};}constexpr pointer operator->() const;Returns:current
.
D.19 Deprecated basic_string capacity [depr.string.capacity]
The following member is declared in addition to those members specified in 21.3.2.4:namespace std {template<class charT, class traits = char_traits<charT>,class Allocator = allocator<charT>>class basic_string {public:void reserve();};}void reserve();Effects: After this call,capacity()
has an unspecified value greater than or equal tosize()
. [Note: This is a non-binding shrink to fit request. —end note]
D.20 Deprecated Standard code conversion facets [depr.locale.stdcvt]
The header<codecvt>
provides code conversion facets for various character encodings.
D.20.1 Header<codecvt>
synopsis [depr.codecvt.syn]namespace std {enum codecvt_mode {consume_header = 4,generate_header = 2,little_endian = 1};template<class Elem, unsigned long Maxcode = 0x10ffff,codecvt_mode Mode = (codecvt_mode)0>class codecvt_utf8: public codecvt<Elem, char, mbstate_t> {public:explicit codecvt_utf8(size_t refs = 0);~codecvt_utf8();};template<class Elem, unsigned long Maxcode = 0x10ffff,codecvt_mode Mode = (codecvt_mode)0>class codecvt_utf16: public codecvt<Elem, char, mbstate_t> {public:explicit codecvt_utf16(size_t refs = 0);~codecvt_utf16();};template<class Elem, unsigned long Maxcode = 0x10ffff,codecvt_mode Mode = (codecvt_mode)0>class codecvt_utf8_utf16: public codecvt<Elem, char, mbstate_t> {public:explicit codecvt_utf8_utf16(size_t refs = 0);~codecvt_utf8_utf16();};}
D.20.2 Requirements [depr.locale.stdcvt.req]
For each of the three code conversion facetscodecvt_utf8
,codecvt_utf16
, andcodecvt_utf8_utf16
:
—Elem
is the wide-character type, such aswchar_t
,char16_t
, orchar32_t
.—Maxcode
is the largest wide-character code that the facet will read or write without reporting a conversion error.— If(Mode & consume_header)
, the facet shall consume an initial header sequence, if present, when reading a multibyte sequence to determine the endianness of the subsequent multibyte sequence to be read.— If(Mode & generate_header)
, the facet shall generate an initial header sequence when writing a multibyte sequence to advertise the endianness of the subsequent multibyte sequence to be written.— If(Mode & little_endian)
, the facet shall generate a multibyte sequence in little-endian order, as opposed to the default big-endian order.For the facetcodecvt_utf8
:
— The facet shall convert between UTF-8 multibyte sequences and UCS2 or UTF-32 (depending on the size ofElem
) within the program.— Endianness shall not affect how multibyte sequences are read or written.— The multibyte sequences may be written as either a text or a binary file.For the facetcodecvt_utf16
:
— The facet shall convert between UTF-16 multibyte sequences and UCS2 or UTF-32 (depending on the size ofElem
) within the program.— Multibyte sequences shall be read or written according to theMode
flag, as set out above.— The multibyte sequences may be written only as a binary file. Attempting to write to a text file produces undefined behavior.For the facetcodecvt_utf8_utf16
:
— The facet shall convert between UTF-8 multibyte sequences and UTF-16 (one or two 16-bit codes) within the program.— Endianness shall not affect how multibyte sequences are read or written.— The multibyte sequences may be written as either a text or a binary file.The encoding forms UTF-8, UTF-16, and UTF-32 are specified in ISO/IEC 10646. The encoding form UCS-2 is specified in ISO/IEC 10646-1:1993.D.21 Deprecated convenience conversion interfaces [depr.conversions]
D.21.1 Class template wstring_convert [depr.conversions.string]
- Class template
wstring_convert
performs conversions between a wide string and a byte string. It lets you specify a code conversion facet (like class templatecodecvt
) to perform the conversions, without affecting any streams or locales. [ Example: If youwant to use the code conversion facethave a code conversion facet namedcodecvt_utf8
codecvt_utf8
that you want to use to output tocout
a UTF-8 multibyte sequence corresponding to a wide string, but you don’t want to alter the locale forcout
, you can write something like:— end example ]std::wstring_convert<std::codecvt_utf8<wchar_t>> myconv; std::string mbstring = myconv.to_bytes(L"Hello\n"); std::cout << mbstring;explicit wstring_convert(Codecvt* pcvt = new Codecvt); wstring_convert(Codecvt* pcvt, state_type state); explicit wstring_convert(const byte_string& byte_err, const wide_string& wide_err = wide_string());
RequiresPreconditions: For the first and second constructors,pcvt != nullptr
.- Effects: The first constructor shall store
pcvt
incvtptr
and default values incvtstate
,byte_err_string
, andwide_err_string
. The second constructor shall storepcvt
incvtptr
,state
incvtstate
, and default values inbyte_err_string
andwide_err_string
; moreover the stored state shall be retained between calls tofrom_bytes
andto_bytes
. The third constructor shall storenew Codecvt
incvtptr
,state_type()
incvtstate
,byte_err
inbyte_err_string
, andwide_err
inwide_err_string
.D.21.2 Class template wbuffer_convert [depr.conversions.buffer]
explicit wbuffer_convert(streambuf* bytebuf = 0, Codecvt* pcvt = new Codecvt, state_type state = state_type());
RequiresPreconditions:pcvt != nullptr
.- Effects: The constructor constructs a stream buffer object, initializes
bufptr
tobytebuf
, initializescvtptr
topcvt
, and initializescvtstate
tostate
.
D.23 Deprecated filesystem path factory functions [depr.fs.path.factory]
template<class Source>path u8path(const Source& source);template<class InputIterator>path u8path(InputIterator first, InputIterator last);Requires: Thesource
and[first, last)
sequences are UTF-8 encoded. The value type ofSource
andInputIterator
ischar
orchar8_t
.Source
meets the requirements specified in 29.11.7.3.Returns:
— Ifvalue_type
ischar
and the current native narrow encoding (29.11.7.2.2) is UTF-8, returnpath(source)
orpath(first, last)
; otherwise,— ifvalue_type
iswchar_t
and the native wide encoding is UTF-16, or ifvalue_type
ischar16_t
orchar32_t
, convertsource
or[first, last)
to a temporary,tmp
, of typestring_type
and returnpath(tmp)
; otherwise,— convertsource
or[first, last)
to a temporary,tmp
, of typeu32string
and returnpath(tmp)
.Remarks: Argument format conversion (29.11.7.2.1) applies to the arguments for these functions. How Unicode encoding conversions are performed is unspecified.[Example: A string is to be read from a database that is encoded in UTF-8, and used to create a directory using the native encoding for filenames:namespace fs = std::filesystem;std::string utf8_string = read_utf8_data();fs::create_directory(fs::u8path(utf8_string));For POSIX-based operating systems with the native narrow encoding set to UTF-8, no encoding or type conversion occurs.
For POSIX-based operating systems with the native narrow encoding not set to UTF-8, a conversion to UTF-32 occurs, followed by a conversion to the current native narrow encoding. Some Unicode characters may have no native character set representation.For Windows-based operating systems a conversion from UTF-8 to UTF-16 occurs. —end example] [Note: The example above is representative of a historical use offilesystem::u8path
. Passing astd::u8string
topath
’s constructor is preferred for an indication of UTF-8 encoding more consistent withpath
’s handling of other encodings. —end note]
Special thanks to Hal Finkel for allowing a late update of the original paper to track the latest IS wording and still make the mailing. Thanks to Stephan T. Lavavej for the early review, and catching too many spelling and grammar errors!
Thanks to everyone who worked on flagging these facilities for deprecation, let's take the next step!
span<charT>
as buffer,
Peter Sommerlad
char8_t
: A type for UTF-8 characters and strings,
Tom Honermann
<codecvt>
,
Alisdair Meredith
atomic_shared_ptr
for C++20,
Alisdair Meredith
this
via [=]
,
Thomas Köppe
string::reserve
Should Not Shrink,
Mark Zeren, Andrew Luo
volatile
,
JF Bastien
volatile
: library,
JF Bastien
ATOMIC_VAR_INIT
,
Jens Gustedt