Sequential hexadecimal digits

Contents

1. Introduction

Among other guarantees, [lex.charset] paragraph 5 ensures the following:

The code unit value of each decimal digit character after the digit 0 (U+0030) is one greater than the value of the previous.

The guarantee is useful because it allows

• testing whether a character c is an ASCII digit using the expression c >= '0' && c <= '9', and
• computing the integer digit value using c - '0'.

Unfortunately, no similar guarantee is provided for other characters, which makes it non-portable to e.g. test whether a character c is a lower-case letter using c >= 'a' && c <= 'z'. This test only works for ASCII-compatible encodings such as UTF-8, and C++ supports encodings such as EBCDIC, where letters are not contiguous.

However, [N3192] Sequential Hexdigits observed that even in EBCDIC, there are blocks of 8 or 9 contiguous letters in the invariant subset. That is, 'a'..'i', 'j'..'r', and 's'..'z' are contiguous blocks; this is analogous for upper-case letters. [N3192] has been merged into the C2y draft, providing the guarantee that 'a'..'f' is a contiguous block, which is not quite as strong as EBCDIC would allow. At least the C2y guarantee should be provided in C++ as well.

2. Motivation

Having the guarantee that the letters 'a'..'f' form a contiguous block is mainly useful for working with hexadecimal digits.

There is likely a large amount of C++ code which relies on the contiguity of that range already, possibly unaware or disinterested in the lack of portability. Hexadecimal digits letters are uniquely interesting because of how frequently they are used and how obviously useful contiguity is.

There is substantially less motivation for the other two EBCDIC letter blocks. While the guarantee allows implementing a test for whether a char is a lowercase letter using three range checks, such an implementation would likely perform worse than a bitset lookup anyway. Locale-specific character tests should either be done using standard library functions, or the user should static_assert that their encoding is ASCII-based.

3. Design

The proposed change is to guarantee that the blocks of letters 'a'..'f' and 'A'..'F' are contiguous, similar to '0'..'9'.

Due to the lack of motivation mentioned above, and out of caution not to provide more guarantees than C2y, no guarantee for other blocks of letters is proposed, despite EBCDIC seemingly allowing for a stronger guarantee.

4. Impact

To my understanding and to the understanding of WG14, there is no change in behavior to existing code, nor is any implementation affected. The proposed change is for all intents and purposes on paper.

However, the proposed change makes it impossible to create a hypothetical future C++ implementation where 'a'..'f' is not contiguous. This seems like an acceptable sacrifice, especially considering that such a C++ implementation would be incompatible with C.

5. Wording

The changes are relative to [N5032].

Change [lex.charset] paragraph 5 as follows:

6. References