swoole hooks can be compiled if pgsql/sqlite are not compiled in statically

[henderkes]

What does this PR do?

enable swoole features as possible

Checklist before merging

If your PR involves the changes mentioned below and completed the action, please tick the corresponding option.
If a modification is not involved, please skip it directly.

• Features working
  • pgsql
  • mysql
  • sqlite
  • threads
  • liburing
  • odbc (add SWOOLE_ODBC_LIBS variable to config.m4 swoole/swoole-src#5847 to get rid of patch)

[jingjingxyk]

new feature:

• `--enable-swoole-stdext'
• stdext enhance

[henderkes]

new feature:

• `--enable-swoole-stdext'
• stdext enhance

What is swoole-stdext?

Edit: Oh damn, I wished for that for a long time. Too bad I can't personally use swoole, we're making use of too many caddy features.

[henderkes]

edit: it's v6.1+ only, the stable release is still v6.0.2, so that will need a feature request that we can do later

[henderkes]

Looks like the swoole fails are not introduced by this pr (they still happen when I disable the new --enable options), but by Swoole src not checking for extended posix support before using it? Hard to track down without a mac.

[crazywhalecc]

I'll check it tomorrow evening.

[henderkes]

@crazywhalecc good for merge?

[crazywhalecc]

Haven't tested on linux yet, after my day job is finished I'll look into it. I've been a bit busy at work lately.

[crazywhalecc]

The macOS bug reason:

• macOS doesn't implement pthread_barrier_* function in pthread.
• The pg_pthread has pthread_barrier port.
• When configuring pthread_barrier_init, this includes pg libs for conftest, and it will pass the check.
• Making does not contain the pg libs, that is, the posix thread barrier obviously not implemented and got failed build.

The root cause is that the configure built in spc will bring other libraries into a separate AC_CHECK_LIB additional library, which leads to this problem.

[crazywhalecc]

And for linux liburing, I dont' see anything wrong, but I recommend triggering full extension tests and all versions of bulk build tests with swoole's hooks at the end.

[henderkes]

Ah, it's pgsql again...

[henderkes]

And for linux liburing, I dont' see anything wrong, but I recommend triggering full extension tests and all versions of bulk build tests with swoole's hooks at the end.

https://github.com/static-php/spc-packages/actions/runs/17260452046

Already tested!

[henderkes]

I think we need to disable opcache jit with zts builds for php < 8.4 on alpine. It was fixed, but apparently not backported.

[crazywhalecc]

I added missing notes and updated the cli generator, good to me now. But I don't think we should add pkg-config things in this PR.

Actually I also prefer hosting pkg-config musl-static binaries in dl.static-php.dev without building in doctor. The pkg-config build actions would be in static-php-cli-hosted.

[henderkes]

Yes, you can change that in the other branch. I don't have access to the dl.static-php so I can't upload a prebuilt version of it.

[henderkes]

Oh, as a side note, what is supposed to happen if swoole-pgsql and pdo_pgsql extension are both loaded shared? Does one need to initialise first to get it working correctly?

[crazywhalecc]

Oh, as a side note, what is supposed to happen if swoole-pgsql and pdo_pgsql extension are both loaded shared? Does one need to initialise first to get it working correctly?

I haven't tested shared build for swoole extension yet, but theoretically it should work together.

[henderkes]

I've tested it and it works fine, but only the general pdo_pgsql. I have never worked with swoole.