[Messenger] Add AMPHP SQL messenger transport

[fabpot]

Q	A
Branch?	8.2
Bug fix?	no
New feature?	yes
Deprecations?	no
Issues	n/a
License	MIT

Docs: symfony/symfony-docs#22457

Adds a Symfony Messenger transport backed by asynchronous AMP SQL that allows to use Messenger without blocking an Amp event loop.

[nicolas-grekas]

Two things I noticed that have not come up yet.

splitsh.json has no entry for the bridge, so the package would never be split or published. Nothing guards that in CI, since package-tests skips component_bridge. The precedent is 0f5b9b6039d, which retrofitted Prelude.

The SQLite DSN reads three slashes as a relative path. amp-sqlite:///tmp/messages.db resolves to tmp/messages.db, and an absolute path needs amp-sqlite:////tmp/messages.db. Everywhere else three slashes means absolute, including sqlite:/// in the PDO cache and lock stores. testSupports and the Messenger TransportFactoryTest fixture both use the three-slash form, so the tests currently lock the behaviour in.

Smaller: createTransport() catches \Throwable and rethrows a bare "DSN is invalid" with no previous exception, so a bad sslmode or a typo in an option loses its cause. And the two [CI] commits reworking static-analysis.yml look unrelated to the bridge. They are also why PHPStan and Psalm are red here: the refresh step wipes composer.lock and reinstalls from the PR's own composer.json, so the base and the PR are no longer analysed against the same tree. Neither job reports anything in an AmpSql file.

[fabpot]

Two things I noticed that have not come up yet.

splitsh.json has no entry for the bridge, so the package would never be split or published. Nothing guards that in CI, since package-tests skips component_bridge. The precedent is 0f5b9b6039d, which retrofitted Prelude.

The SQLite DSN reads three slashes as a relative path. amp-sqlite:///tmp/messages.db resolves to tmp/messages.db, and an absolute path needs amp-sqlite:////tmp/messages.db. Everywhere else three slashes means absolute, including sqlite:/// in the PDO cache and lock stores. testSupports and the Messenger TransportFactoryTest fixture both use the three-slash form, so the tests currently lock the behaviour in.

Smaller: createTransport() catches \Throwable and rethrows a bare "DSN is invalid" with no previous exception, so a bad sslmode or a typo in an option loses its cause. And the two [CI] commits reworking static-analysis.yml look unrelated to the bridge. They are also why PHPStan and Psalm are red here: the refresh step wipes composer.lock and reinstalls from the PR's own composer.json, so the base and the PR are no longer analysed against the same tree. Neither job reports anything in an AmpSql file.

All fixed