[HttpKernel] Make Kernel::getShareDir() nullable

[nicolas-grekas]

Q	A
Branch?	7.4
Bug fix?	no
New feature?	no
Deprecations?	no
Issues	#62202
License	MIT

This allows specifying explicitly when the app has no share dir.

If the method returns null, the kernel.share_dir parameter is not defined.
This gives a nice way to spot where references to this parameter remain with You have requested a non-existent parameter "kernel.share_dir" exceptions as a hint.

Setting the APP_SHARE_DIR env var to any falsy value (off, false, 0, etc as supported by filter_var()) will lead to configuring the app with no share dir.

[GromNaN]

This allows specifying explicitly when the app has no share dir.

In that case, app cache directory have to be set back to %kernel.cache_dir%/pools/app.
Using a non-shared for app cache is not always an issue.

[xabbuh]

the changes from #62191 also need to be addressed

[nicolas-grekas]

In that case, app cache directory have to be set back to %kernel.cache_dir%/pools/app.

nope: that's exactly the kind of config you want to spot early on: if you go with no share dir, your app cache pools have to be configured to use eg Redis by default (or you might be fine with using the cache-dir, but it shouldn't be our call by default)

[nicolas-grekas]

the changes from #62191 also need to be addressed

indeed, PR updated

[GromNaN]

nope: that's exactly the kind of config you want to spot early on: if you go with no share dir, your app cache pools have to be configured to use eg Redis by default (or you might be fine with using the cache-dir, but it shouldn't be our call by default)

Ok, I understand the intention better now. The fallback on cache_dir is a bit strange in that we only get an error only if we explicitly indicate that we don't have a share_dir, which is the default for multi-server deployments. There is no way to know if we are running on a single server or a cluster of servers without nfs.

[nicolas-grekas]

Sure. We don't want to disrupt the defaults.
What's nice about this is that it's opt-in: ppl that care will now be able to spot when some subsytem relies on a share dir and need to be reconfigured.

[alexander-schranz]

should this not be now return null; ?

[nicolas-grekas]

nope: we want to null to be opt-in

[xabbuh]

Is this needed? PHPUnit does restore the $_SERVER array after each test if the backup globale feature isn’t disabled, doesn’t it?

[nicolas-grekas]

that's quite possible 😬

[xabbuh]

FTR, it's the opposite: #62273

[Kocal]

Hey!

I think this PR broke UX test-suite on based Symfony 8: https://github.com/symfony/ux/actions/runs/18986901497/job/54232436997?pr=3154

PHP Fatal error: Declaration of Symfony\UX\LiveComponent\Tests\Fixtures\Kernel::getShareDir(): ?string must be compatible with Symfony\Component\HttpKernel\Kernel::getShareDir(): string in /home/runner/work/ux/ux/src/LiveComponent/vendor/symfony/framework-bundle/Kernel/MicroKernelTrait.php on line 127

I'm not sure, but maybe a Composer conflict must be configured?

[nicolas-grekas]

@Kocal it's just that the change wasn't merged yet into 8.0. I did it now.

[Kocal]

Thanks, that's perfect now :)