CODING_STANDARDS.md: establish C99 as the implementation language #10631
Conversation
PHP 8 switched to C99, but did not documented that anywhere. After @derickr rejected a pull request on timelib (derickr/timelib#141 (comment)) because my suggested change removed compile-time checks for fixed-width integer types, pointing out that they are optional in the C99 standard, @nikic disagreed with using `uint_least8_t` instead (which is guaranteed to be available), stating that "We already make extensive use of uint8_t, you can assume it exists" (php#10621 (review)). In order to avoid such confusion in the future, let's document this architecture requirement.
|
Just as a side node, timelib is a library independent from php-src, so both can have different requirements regarding the supported C standard and compiler support. (As long as the php-src requirement >= the timelib requirement.) That said, the only reasonable environment in which the uintN_t types may not be available is an environment where CHAR_BITS is not 8. Clang doesn't even support such targets in the first place :) |
|
Of course, we all know the world has settled on machines where all of those are available and nobody bothers to support theoretical machines where it isn't, but that assumption should be documented. Prior to PHP8, these configure-/compile-time checks existed not to support those theoretical machines, but because C89 did not have any fixed-width integers. The removal of those checks does not break PHP (or timelib) on theoretical machines; it was already broken before, and those checks did not help with that. |
PHP 8 switched to C99, but did not documented that anywhere.
After @derickr rejected a pull request on timelib
(derickr/timelib#141 (comment)) because my suggested change removed compile-time checks for fixed-width integer types, pointing out that they are optional in the C99 standard, @nikic disagreed with using
uint_least8_tinstead (which is guaranteed to be available), stating that "We already make extensive use of uint8_t, you can assume it exists" (#10621 (review)).In order to avoid such confusion in the future, let's document this architecture requirement.