PhpParser 5 & PhpUnit 11

[kukulich]

1. There are some little BC breaks, see changes eg. inClassForSourceStubberExpected.php

/** Comment */
define('CONSTANT_NAME', true);

Node\Expr\FuncCall is now wrapped in Node\Stmt\Expression and the comment is part of the wrapper. I've made a little hack and I copy the comment to the FuncCall node. See eg. FindReflectionsInTree

[Ocramius]

Note: didn't read through the whole patch yet 😁

[Ocramius]

@kukulich I don't see any explicit BC breaks, or am I missing something?

[Ocramius]

Suspicious of this change: why is the AutoloadSourceLocator making decisions about changing the AST?

I'm not questioning the fact that this fixes something, but rather the fact that this call is specifically part of a source locator 🤔

[Ocramius]

@kukulich this is the last concern in this patch: what's going on with us manipulating nodes during *Locator activities?

[kukulich]

It looks I've copied too much yesterday. The "copy" hack is not neccessary in AutoloadSourceLocator. Removed now.

[Ocramius]

Did a local test of performance improvements:

betterreflection on  6.27.x [$?] via 🐘 v8.2.16 via ❄️  impure (nix-shell) 
❯ ./tools/vendor/bin/phpbench run --warmup=3 --revs=10 --iterations=20 --retry-threshold=4 --tag=6.27.x
betterreflection on  HEAD (8f94dee) [$?] via 🐘 v8.2.16 via ❄️  impure (nix-shell) 
❯ ./vendor/bin/phpbench report --ref=6.27.x --report=aggregate
+----------------------+------------------------------+-----+------+-----+----------+---------+--------+
| benchmark            | subject                      | set | revs | its | mem_peak | mode    | rstdev |
+----------------------+------------------------------+-----+------+-----+----------+---------+--------+
| PhpUnitTestCaseBench | benchReflectClass            |     | 10   | 20  | 17.154mb | 2.300μs | ±1.90% |
| PhpUnitTestCaseBench | benchReflectMethodParameters |     | 10   | 20  | 17.154mb | 0.900μs | ±0.00% |
+----------------------+------------------------------+-----+------+-----+----------+---------+--------+

betterreflection on  HEAD (8f94dee) [$?] via 🐘 v8.2.16 via ❄️  impure (nix-shell) 
❯ ./vendor/bin/phpbench run --warmup=3 --revs=10 --iterations=20 --retry-threshold=4 --tag=1404
betterreflection on  HEAD (8f94dee) [$?] via 🐘 v8.2.16 via ❄️  impure (nix-shell) took 8s 
❯ ./vendor/bin/phpbench report --ref=1404 --report=aggregate
+----------------------+------------------------------+-----+------+-----+----------+---------+--------+
| benchmark            | subject                      | set | revs | its | mem_peak | mode    | rstdev |
+----------------------+------------------------------+-----+------+-----+----------+---------+--------+
| PhpUnitTestCaseBench | benchReflectClass            |     | 10   | 20  | 11.882mb | 2.300μs | ±2.21% |
| PhpUnitTestCaseBench | benchReflectMethodParameters |     | 10   | 20  | 11.882mb | 0.900μs | ±0.00% |
+----------------------+------------------------------+-----+------+-----+----------+---------+--------+

Time-wise, we're running the same stuff, but memory usage dropped sharply, probably due to PHP-Parser relying less on array structures internally 💪

[Ocramius]

Most excellent, thanks @kukulich!

[Ocramius]

This is going in a minor release :)

People should pin their own nikic/php-parser, if they rely on it directly ;-)