[staabm]

as requested in #4823 (comment)

[ondrejmirtes]

Also the performance should be observed with special care. Does anything gets slower because of this?

[phpstan-bot]

This pull request has been marked as ready for review.

[staabm]

Does anything gets slower because of this?

before PR:

➜  phpstan-src git:(2.1.x) ✗ hyperfine 'vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php' --runs=5 
Benchmark 1: vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php
  Time (mean ± σ):     17.367 s ±  0.437 s    [User: 15.445 s, System: 0.599 s]
  Range (min … max):   16.896 s … 18.083 s    5 runs

with this PR 8c8b65a:

➜  phpstan-src git:(dis-arr) ✗ hyperfine 'vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php' --runs=5 
Benchmark 1: vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php
  Time (mean ± σ):     17.080 s ±  0.285 s    [User: 14.936 s, System: 0.631 s]
  Range (min … max):   16.744 s … 17.378 s    5 runs

I think its within the margin of error (because my company laptop has varying performance because of anti-virus stuff)

[ondrejmirtes]

I'm still worried this can break stuff. I imagine that some $this->degradeToGeneralArray = true; assignments should still be allowed to happen but some should be guarded with !$this->disableArrayDegradation. Instead of your current way of changing conditions into if (!$this->degradeToGeneralArray || $this->disableArrayDegradation) {.

One example: we want $this->degradeToGeneralArray = true; to happen and be respected at the bottom of setOffsetValueType method. Because when someone assigns general IntegerType, it doesn't make sense to keep the array shape.

[staabm]

ok, how do you feel about it now?

with this PR 71303cc:

➜  phpstan-src git:(dis-arr) ✗ hyperfine 'vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php' --runs=5 
Benchmark 1: vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php
  Time (mean ± σ):     16.712 s ±  0.157 s    [User: 15.144 s, System: 0.609 s]
  Range (min … max):   16.459 s … 16.848 s    5 runs

[ondrejmirtes]

This makes sense to me now :) Thank you!