On Sun, Aug 4, 2024 at 10:05 AM Gina P. Banyard <internals@gpb.moe> wrote:
On Friday, 2 August 2024 at 14:33, Giovanni Giacobbi <giovanni@giacobbi.net> wrote:
On Thu, 1 Aug 2024 at 23:57, Ilija Tovilo <tovilo.ilija@gmail.com> wrote:
[...]
I started fixing these in a PR [1] which required more changes than
expected. After a short discourse, we were wondering whether it might
be better to switch to a newer C standard instead. Our coding
standards [2] currently specify that compiling php-src requires C99.
The Unix installation page on php.net [3] claims it is ANSI C, which
is certainly outdated. There have been suggestions to require C11 for
a while, which should be well supported by all compilers shipped with
maintained distributions.
It feels wrong to raise such an important requirement that might affect a lot of people, including maintainers of extensions, for just one specific problem, and 99% of the codebase would still be C99 compliant.
I quickly put together an alternative PR (#15202) with a slightly different approach, just as a proof of concept. The idea is to move all the global typedefs in a new include header "zend_types_defs.h" (but also zend_portability.h can be reused for this purpose, as all the relevant files already include it).
While putting together that PR, I had the feeling that this typedef redefinition problem is in reality hiding some smelly design of header files. So maybe, rather than requiring a compiler more tolerant to poor code, we should rather focus on getting the design right.
I would like to see actual factual data rather than "vibes" as to how this would cause issues with third-party extensions.
php-src is not C99 pedeantic compliant and we do use GCC extenstions.
I do agree that maybe we should go back to fixing headers, but considering the drama this caused 18 months ago I do not know who here has the motivation to do this.
Ideally I feel we should target C17 as it is a "bug fix" release of C11, but this might be impossible due to old compilers still being the default on LTS Distributions.
The main benefit of C11/17 in the long run are atomics, that we kinda use already anyway.
As a person who wrote some of those atomics, I would definitely prefer
to move to standard C11/C17 atomics at some point in the future and
remove all other fallbacks. The blocker right now is Microsoft Visual
Studio. Although they added C11/C17 support a few versions back, they
did not add atomics at that time. It's available now, but it is still
experimental as far as I can tell.
Maybe for PHP 9.0 we can address header issues and also by then VS
will have a general release we can move to for atomics.
One thing I feel strongly about is that we should not add flags like
`-std=c11`. If we check, we ought to just ensure that whatever flags
the user provided, we have certain C11/C17 features. This allows for
extensions to use a newer version, or a relaxed version like
`-std=gnu11`.