Add Alpine 3.22 (remove Alpine 3.20)

[tianon]

No description provided.

[yosifkit]

😞 😮

In file included from /usr/src/php/ext/opcache/jit/zend_jit.c:48:
ext/opcache/jit/zend_jit_trace.c: In function 'zend_jit_trace':
ext/opcache/jit/zend_jit_trace.c:6337:70: warning: 'gen_handler' may be used uninitialized [-Wmaybe-uninitialized]
 6337 |                                                         (gen_handler || type == IS_UNKNOWN || !ra || !ra[ssa_op->result_def]));
ext/opcache/jit/zend_jit_internal.h:552:21: note: in definition of macro 'SET_STACK_TYPE'
  552 |                 if (_set_mem_type) { \
      |                     ^~~~~~~~~~~~~
In file included from /usr/src/php/ext/opcache/jit/zend_jit.c:4460:
ext/opcache/jit/zend_jit_trace.c:4288:30: note: 'gen_handler' was declared here
 4288 |                         bool gen_handler;
      |                              ^~~~~~~~~~~
/usr/src/php/ext/opcache/jit/zend_jit_x86.dasc: Assembler messages:
/usr/src/php/ext/opcache/jit/zend_jit_x86.dasc:2976: Error: @TLSGD operator requires `%rdi' as dest register
make: *** [Makefile:1303: ext/opcache/jit/zend_jit.lo] Error 1

[tianon]

_{(in which I yet again consider dropping zts variants entirely)}

[yosifkit]

🤔 👀 ❓ I found where the error is coming from, I think, but I don't know why or how to fix it.
https://sourceware.org/pipermail/binutils-cvs/2024-September/065790.html

[arnaud-lb]

See PHP issue: php/php-src#18743 (comment)

[nielsdos]

See PHP issue: php/php-src#18743 (comment)

Follow-up on this: the fix got merged upstream in php-src, and the fix will be released with PHP 8.3.23 / PHP 8.4.9.

[yosifkit]

Awesome, thanks for the quick fix @nielsdos! ❤️ 🙏

@tianon, should we backport the fix? (php/php-src@b3c8afe)

[tianon]

(that's just a rebase for starters - we also need to decide what to do with 8.1 and 8.2 that didn't get a backported fix as far as I can see)

[tianon]

Oh, and we have the patching issue that the code sticks around -- I'm guessing patching this during the build of php itself is enough, but if this codepath ever ends up being compiled as part of a module build then we have to figure out a clean way to inject a patch into those docker-php-* scripts (which isn't something I'm too keen on doing).

[nielsdos]

we also need to decide what to do with 8.1 and 8.2 that didn't get a backported fix as far as I can see

8.1 & 8.2 are out of bugfix support. If the patch doesn't apply/work there, then I can provide a patchfile for those versions that you can apply (ping me in that case).

[yosifkit]

Since 8.1 and 8.2 are out of bugfix support, we'll have to apply the backport ourselves until their end of life. As for extension time in docker-php-ext-* scripts, we can just leave the patches out (like the current patches on 8.1) and see what to do if there is a problem where any of them are needed for an extension that we don't build in.

[tianon]

@nielsdos looks like the upstream patch will work as-is for 8.2 and 8.3 (and the merge commit version in 8.4 - I'll push that change soon), but 8.1 needs something slightly different; would php/php-src@f171d5f be correct? did I miss anything?

[tianon]

FWIW, that patch does appear to build successfully 😅

[nielsdos]

The patch to make it build on 8.1 seems right.
The index multiplier used is 16 but I believe it should be 8. Probably the index is 0 which would explain why it works anyway. I'm on my phone now so I can't double check.

[tianon]

Interesting, yeah -- in 8.2+ it appears that this was split specifically in part to set that value to 8 for __MUSL__ and __FreeBSD__, but still 16 for everything else?

https://github.com/php/php-src/blob/b3c8afe272a6919248986c703c2e1defc73ff707/ext/opcache/jit/zend_jit_x86.dasc#L2908-L2933

Should I perhaps update my patch to backport that __MUSL__ block too? (I'll openly admit that I don't know what any of this means/does, just that the syntax is pretty easy to read and intuit some amount of intent from. 😄)

[tianon]

php/php-src@6b105d4 is what that looks like -- I copied the #else block and updated 8 to 16 and then staged that and copied in the newer block from 8.2+ just to be absolutely sure I wasn't letting my eyes trick me on the block similarity and it was identical. 😄

(The #if above this already includes && !defined(__MUSL__) 👍)

[tianon]

I've verified locally that using that new patch also appears to work fine and compile successfully, so if it's more technically correct I'm happy to commit that swap. 👍 ❤️

[tianon]

in 8.2+ it appears that this was split specifically in part to set that value to 8 for __MUSL__

php/php-src@94ba883 is where this happened, specifically 👍 👀

[tianon]

(I'm now convinced enough that this is correct -- happy to hear dissent/feedback, though! ❤️)

[nielsdos]

php/php-src@6b105d4 is what that looks like -- I copied the #else block and updated 8 to 16 and then staged that and copied in the newer block from 8.2+ just to be absolutely sure I wasn't letting my eyes trick me on the block similarity and it was identical. 😄

(The #if above this already includes && !defined(__MUSL__) 👍)

I checked this and it is right. It's indeed technically more correct.

[tianon]

Perfect, thank you so much, @nielsdos! ❤️