Coordination: Prepare for students code using PHP 8.3

[mk-mxp]

For exercism/php#652 we need to have the representer ready to accept PHP 8.3 code from students. The representer itself may run on a different PHP version.

@homersimpsons Can you outline the required work to do here? An actionable list of tasks would be perfect, but any hints are welcome.

[homersimpsons]

@homersimpsons Can you outline the required work to do here? An actionable list of tasks would be perfect, but any hints are welcome.

To me there is nothing to do:

• https://github.com/nikic/PHP-Parser already supports PHP 8.3
• https://www.php.net/releases/8.3/en.php there does not seem to be anything important to normalize, things that may be great to normalize are:
  • Remove the #[\Override] attribute
  • Remove the type for class constant

That said I could also update the base docker image of the representer so everything runs on PHP8.3

[mk-mxp]

I will mark the presenter as ready for PHP8.3, as your additions most likely won't affect any solutions soon.

For running on PHP 8.3: Use XDEBUG_MODE=off for performance reasons. I found at many places, that there is a huge impact even for small programs against having the default mode.

[homersimpsons]

For the record, I did upgrade the runtime for the representer to PHP8.3.

I did not try to change XDEBUG settings nor looked into the 2 normalizations I did suggest above. So I will let this issue open for now.

[homersimpsons]

For running on PHP 8.3: Use XDEBUG_MODE=off for performance reasons. I found at many places, that there is a huge impact even for small programs against having the default mode.

I tried doing this, and it didn't lead to any significant change. Maybe I did it wrong though. My steps were:

1. Copy-paste a solution multiple times in a file in order to create a big php file (~30_000 lines)
2. Build a container and run it normally or with XDEBUG_MODE=off
3. Used hyperfine to measure changes. Note that I set --cpus 0.5 for both commands to reduce computer noise (I believe)

❯ ./hyperfine 'docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output' 'docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp --env XDEBUG_MODE=off representer-base "toto" /solution /output'
Benchmark 1: docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output
  Time (mean ± σ):      1.740 s ±  0.103 s    [User: 0.012 s, System: 0.013 s]
  Range (min … max):    1.598 s …  1.961 s    10 runs
 
Benchmark 2: docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp --env XDEBUG_MODE=off representer-base "toto" /solution /output
  Time (mean ± σ):      1.737 s ±  0.098 s    [User: 0.012 s, System: 0.013 s]
  Range (min … max):    1.620 s …  1.938 s    10 runs
 
Summary
  docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp --env XDEBUG_MODE=off representer-base "toto" /solution /output ran
    1.00 ± 0.08 times faster than docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output

Note that running with XDEBUG_MODE=debug,coverage does not really change the results.

❯ ./hyperfine 'docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output' 'docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp --env XDEBUG_MODE=debug,coverage representer-base "toto" /solution /output'
Benchmark 1: docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output
  Time (mean ± σ):      1.536 s ±  0.021 s    [User: 0.011 s, System: 0.013 s]
  Range (min … max):    1.508 s …  1.573 s    10 runs
 
Benchmark 2: docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp --env XDEBUG_MODE=debug,coverage representer-base "toto" /solution /output
  Time (mean ± σ):      1.536 s ±  0.033 s    [User: 0.012 s, System: 0.013 s]
  Range (min … max):    1.491 s …  1.604 s    10 runs
 
Summary
  docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output ran
    1.00 ± 0.03 times faster than docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp --env XDEBUG_MODE=debug,coverage representer-base "toto" /solution /output

I also tried with adding a RUN mv "$PHP_INI_DIR/php.ini-production" "$PHP_INI_DIR/php.ini" in the Dockerfile:

❯ ./hyperfine 'docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output' 'docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-prod "toto" /solution /output'
Benchmark 1: docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output
  Time (mean ± σ):      1.559 s ±  0.019 s    [User: 0.013 s, System: 0.011 s]
  Range (min … max):    1.527 s …  1.585 s    10 runs
 
Benchmark 2: docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-prod "toto" /solution /output
  Time (mean ± σ):      1.550 s ±  0.033 s    [User: 0.013 s, System: 0.011 s]
  Range (min … max):    1.504 s …  1.602 s    10 runs
 
Summary
  docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-prod "toto" /solution /output ran
    1.01 ± 0.02 times faster than docker run --rm --network none --read-only --cpus 0.5 --mount type=bind,source="/home/foo/php-representer/solutions/bench-simple-cipher",destination=/solution --mount type=tmpfs,destination=/output --mount type=tmpfs,destination=/tmp representer-base "toto" /solution /output

I believe that this is because we are in CLI and hence those settings have close to 0 impact.

I also tried to run those with poop to get other metrics, but it always failed to run (I believe because of the quotes or the special --mount argument).

[mk-mxp]

It's been a year since I was hit by the performance degradation and haven't looked into it lately. Seems to be gone now, that's good news!

[homersimpsons]

I'm closing this after #299:

• "Remove the #[\Override] attribute", while probably relevant, it is unlikely that we see it in the wild yet.
• For the XDEBUG settings, see the more lengthy discussion above.