[SanderMuller]

This adds an optional $withOpcache parameter to TurboProcessRestarter::restartIfSuitable(). With false, the restart still loads Turbo and disarms xdebug, pcov and uopz, but it does not turn on OPcache. The default does not change.

This is for Rector, where a warm run is slower with OPcache on. In a warm run, the main process compiles each file it loads once and then exits. Forked workers turn OPcache off, so only the main process uses it.

PHPStan keeps OPcache, because it helps a cold run. I measured PHPStan 2.3.2 (with #6834) on GitHub runners, on 1252 files of Rector's own code at level 5. The only change between the two arms is the opcache restart reason. A cold run has no result cache, and a run without tmpDir also builds the container. Medians:

	OPcache on	OPcache off
warm, x86	1.40 s	1.03 s
warm, arm64	1.26 s	0.99 s
cold, x86 (wall / CPU)	10.62 / 36.7 s	10.70 / 38.0 s
cold, arm64	8.87 / 29.9 s	9.12 / 31.7 s
no tmpDir, x86	12.34 / 41.0 s	12.80 / 43.9 s
no tmpDir, arm64	10.14 / 32.8 s	10.42 / 34.5 s

#6634 tries to keep the cold gain without the warm cost. I have not measured it on cold runs yet.

Both arms report the same errors. The new e2e row runs PHPStan through a wrapper that restarts with $withOpcache false. It checks that one restart disarms pcov and that OPcache stays off. I ran the same check locally with xdebug as the restart reason. Without the change, it fails on restarted-for and opcache-cli-on.

🤖 Generated with Claude Code

[ondrejmirtes]

I feel like we don't need to give this option. Cold runs are faster with OPCache, and we might be able to make warm runs faster later with a daemon process.

It's not like we have the data about whether the run will be warm or cold at that point to decide "do we want OPCache or not?".