use a separate resultCache per project config file by staabm · Pull Request #1469 · phpstan/phpstan-src

staabm · 2022-06-23T12:30:27Z

closes phpstan/phpstan#7379
closes phpstan/phpstan#7601

With this patch we see analysis time dropped to ~5 seconds from previous 2-3 minutes in mono-repo projects, which contain several phpstan.neon files.

staabm · 2022-06-23T12:33:34Z

while thinking about the idea suggested in phpstan/phpstan#7379 (comment) I came to this - in my eyes - even simpler solution to the problem.

instead of fiddling with all the different path-types involved we just use a separate resultCache-name per project-configuration file.

tbh the solutions seems to be to simple, so I think it might have some serious problems somewhere ;-).

~~in case we can proceed with this idea, I would check the current build errors and see how to actually test this change...~~ at least the local testing on my project suggests that it is now using the result-cache where it didn't before.. so it seems to solve my problem at hand.

any feedback is welcome.

The $resultCacheName parameter of restore method shouldn't be misused for this, its usage is reserved only for PHPStan Pro purposes. It puts the result cache file in a subdirectory, I don't want to complicate it like that.

But hey - the logic you're doing here depends on $projectConfigArray and that is entirely available in restore() method too, so you can do the logic based on that there. A lot of useful values that can be used in the hash computation logic are already in the constructor/can be asked for there so I think this idea has wings.

adjusted per feedack and manually tested locally.

do we need a automatic test for the case? if so where and what to test (might be a bit hard because of random file-names involved)?

btw: in the docs there is no resultCachePath mentioned.. should it be added?

https://phpstan.org/config-reference#caching

ondrejmirtes · 2022-06-23T15:30:01Z

The $resultCacheName parameter of restore method shouldn't be misused for this, its usage is reserved only for PHPStan Pro purposes. It puts the result cache file in a subdirectory, I don't want to complicate it like that.

But hey - the logic you're doing here depends on $projectConfigArray and that is entirely available in restore() method too, so you can do the logic based on that there. A lot of useful values that can be used in the hash computation logic are already in the constructor/can be asked for there so I think this idea has wings.

staabm · 2022-06-30T15:20:15Z

added a test-case. should be good to go.

staabm · 2022-06-30T15:21:23Z

since checking exactly which hashes are generated is a bit too much IMO, I just verified phpstan will create result-cache files including some suffix (aka. the hash), which it did not do before this change.

staabm · 2022-06-30T15:22:56Z

the tmpDir case is already covered by an existing test which uses a tmpDir

staabm · 2022-06-30T15:23:17Z

this condition is covered by #1475

staabm · 2022-06-30T15:29:59Z

only thing I am not 100% sure: what needs to be here in this hash.

maybe just the $this->analysedPaths are already enough, since these are the absolute paths within the system..?

staabm · 2022-07-10T14:49:01Z

@ondrejmirtes anything I can do to get this merged?

refs phpstan/phpstan#7601 (comment)

herndlm · 2022-08-31T11:19:30Z

I'm looking forward to this feature too. I should be able to get rid of some local config entries most likely :)

herndlm · 2022-12-19T12:59:31Z

sorry, for the "bump", but I just thought about this PR and was wondering about its state and/or if it was forgotten maybe :)

ondrejmirtes · 2022-12-19T14:10:52Z

I'm a bit afraid to touch this area because it's hard to test it.

SanderMuller · 2026-09-03T07:43:26Z

Still live on current 2.2.x, and a bit broader than the issues describe. Two configs in one project, run alternately: every run after the first reports metadata do not match: projectConfig, analysedPaths, so neither config ever gets a warm cache. And since the default tmpDir is sys_get_temp_dir() . '/phpstan', two unrelated projects on one machine do the same to each other whenever neither sets tmpDir - I reproduced that with two throwaway projects, 6 runs, 5 full re-analyses.

On the "hard to test" worry: it is much easier to pin now than in 2022. The result-cache e2e blocks are plain shell, so alternating two configs and asserting Result cache restored. 0 files will be reanalysed. is a few lines, and a block can clear the machine temp dir first for the default-tmpDir case. Happy to contribute that fixture, as a failing test first if you prefer that shape.

Two things I noticed in the patch while reproducing:

the per-config path only applies when neither tmpDir nor resultCachePath is set, so a project that keeps its cache in the repo for CI (tmpDir: .cache/phpstan, several configs) keeps thrashing;
the name is an md5 of analysedPaths plus bootstrapFiles, so two configs over the same paths that differ only in level still land on one file and still invalidate each other - that was my first reproduction.

Deriving the name from the same projectConfig array the metadata already builds would cover both.

ondrejmirtes · 2026-09-03T09:10:59Z

Hey, this is very close to my current interest and headspace so I can think about it clearly. Can you please open a new issue and describe what we learnt meanwhile, what use cases should this solve and how does the recent relativization of paths in the result cache relate to this? (It might solve some of the use cases.)

Also I think the git worktrees might be relevant here.

What I have to say about this now is that thanks to relative paths, even if we don't set a local tmpDir and just use the system one, if we copy a project to a different directory on disk, PHPStan should be able to reuse the same result cache without invalidation (when the projects are very similar).

But of course if something in the meta changes we need to have a separate result cache so that's where the discussion should start.

Feel free to mention these points in the description too, we will build on them too.

SanderMuller · 2026-09-03T09:35:43Z

Opened as phpstan/phpstan#15163, with your points in the description.

Your hunch about relative paths checks out in the strongest form: a project copied to a different directory, PHPStan in its own vendor, no local tmpDir, reuses the cache with 0 files will be reanalysed. It stops applying when PHPStan lives outside the project, since the anchor is %rootDir% and then it does not move with the copy.

The worktree measurements are in there too, along with the two remaining thrash cases and a suggestion for where the file name could come from.

staabm commented Jun 23, 2022

View reviewed changes

staabm marked this pull request as ready for review June 23, 2022 14:30

ondrejmirtes requested changes Jun 23, 2022

View reviewed changes

staabm force-pushed the separate-caches branch from c6d5ecf to d6af959 Compare June 23, 2022 19:05

staabm marked this pull request as draft June 23, 2022 19:41

staabm force-pushed the separate-caches branch from fff3d06 to b235560 Compare June 23, 2022 19:47

clxmstaab force-pushed the separate-caches branch from b235560 to 1f38eba Compare June 24, 2022 07:49

staabm force-pushed the separate-caches branch from 1f38eba to 7ae6a7e Compare June 24, 2022 07:59

staabm marked this pull request as ready for review June 24, 2022 08:01

clxmstaab force-pushed the separate-caches branch 4 times, most recently from 2c1a994 to 90d9217 Compare June 24, 2022 08:12

staabm mentioned this pull request Jun 24, 2022

e2e test resultCachePath #1475

Merged

staabm force-pushed the separate-caches branch 3 times, most recently from c8658a0 to 67e8c4b Compare June 27, 2022 12:02

clxmstaab force-pushed the separate-caches branch from 67e8c4b to b6dda79 Compare June 29, 2022 14:41

staabm changed the base branch from 1.7.x to 1.8.x June 29, 2022 14:41

staabm force-pushed the separate-caches branch from b6dda79 to 3d85276 Compare June 29, 2022 14:41

staabm marked this pull request as draft June 29, 2022 14:57

clxmstaab force-pushed the separate-caches branch from 3af7989 to 3c8cace Compare June 29, 2022 15:26

staabm mentioned this pull request Jun 30, 2022

Name resultCache.php file more uniquely in global tmpDir phpstan/phpstan#7379

Open

clxmstaab force-pushed the separate-caches branch 2 times, most recently from ea060a9 to c711f0d Compare June 30, 2022 15:17

staabm marked this pull request as ready for review June 30, 2022 15:19

clxmstaab force-pushed the separate-caches branch from c711f0d to 7bebfdf Compare June 30, 2022 15:20

staabm commented Jun 30, 2022

View reviewed changes

Comment thread src/Analyser/ResultCache/ResultCacheManager.php Outdated

staabm Jun 30, 2022

Copy link
Copy Markdown

Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this condition is covered by #1475

staabm commented Jun 30, 2022

View reviewed changes

staabm force-pushed the separate-caches branch 2 times, most recently from 286d1f6 to 3ac86ba Compare July 10, 2022 12:34

staabm force-pushed the separate-caches branch from 3ac86ba to 7ff840b Compare July 18, 2022 18:31

ondrejmirtes force-pushed the 1.8.x branch from b69949f to 5511155 Compare July 19, 2022 12:52

staabm force-pushed the separate-caches branch from 46831b1 to 9e98a7e Compare August 30, 2022 06:57

staabm marked this pull request as draft August 30, 2022 07:04

clxmstaab and others added 6 commits August 31, 2022 13:00

use a separate resultCache per project config file

ccf2063

cs

a633809

test separate cache files are generated

bab9ff0

use phpstan-version in result-cache hash

3e83afc

refs phpstan/phpstan#7601 (comment)

fixed method case

1f90cc2

fix after rebase

e24b20e

staabm force-pushed the separate-caches branch from 9e98a7e to e24b20e Compare August 31, 2022 11:04

staabm marked this pull request as ready for review August 31, 2022 11:18

staabm mentioned this pull request Sep 9, 2022

Allow configuring tmpDir using PHPSTAN_TMPDIR environment variable #1717

Closed

SanderMuller mentioned this pull request Sep 3, 2026

Do not invalidate the result cache when tmpDir changes #6359

Merged

SanderMuller mentioned this pull request Sep 3, 2026

Result cache: when should two analyses share a cache file, and when should they have their own? phpstan/phpstan#15163

Open

staabm closed this Sep 3, 2026

staabm deleted the separate-caches branch September 3, 2026 12:33

SanderMuller mentioned this pull request Sep 14, 2026

Two config files in one project never get a warm result cache phpstan/phpstan#15232

Open

Conversation

staabm commented Jun 23, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

staabm Jun 23, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ondrejmirtes Jun 23, 2022

Choose a reason for hiding this comment

Uh oh!

staabm Jun 24, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

staabm Jun 24, 2022

Choose a reason for hiding this comment

Uh oh!

ondrejmirtes Jun 23, 2022

Choose a reason for hiding this comment

Uh oh!

staabm commented Jun 30, 2022

Uh oh!

staabm Jun 30, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

staabm Jun 30, 2022

Choose a reason for hiding this comment

Uh oh!

staabm Jun 30, 2022

Choose a reason for hiding this comment

Uh oh!

staabm Jun 30, 2022

Choose a reason for hiding this comment

Uh oh!

staabm commented Jul 10, 2022

Uh oh!

herndlm commented Aug 31, 2022

Uh oh!

herndlm commented Dec 19, 2022

Uh oh!

ondrejmirtes commented Dec 19, 2022

Uh oh!

SanderMuller commented Sep 3, 2026

Uh oh!

ondrejmirtes commented Sep 3, 2026

Uh oh!

SanderMuller commented Sep 3, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

staabm commented Jun 23, 2022 •

edited

Loading

staabm Jun 23, 2022 •

edited

Loading

staabm Jun 24, 2022 •

edited

Loading

staabm Jun 30, 2022 •

edited

Loading