FYI this does complete for me in 8 minutes, but I have a beefy workstation, I expect it will run for hours on GH. The issue is there's just some gigantic examples generated (Github, Twitter, Docker API) which contain thousands of files which now all need to be processed before comparing them.

Excluding Docker API, Github and issue-445 would result in significant speedup.

I've excluded the biggest fixtures from being tested on old PHP Parser, the tests now run in 47s for me (as opposed to 8min). It's now useable again.

This should now work except for nikic/PHP-Parser#1009

Since we don't know when and if this gets fixed, the solution I've used is to NOT use ternaries in Client and BaseEndpoint. It results in slightly longer code, but who cares since it's generated. PHP Parser 5.1 got released fixing this issue, we're immediately requiring it.

OK, I've let the v4 version of tests run without excluding the beefy fixtures locally and the result is now

Tests: 131, Assertions: 22699, Failures: 1.

It all works on both v4 and v5, except the datetime-no-format fixture as seen in the checks here too. It means we can fix that by removing the tertiary and should be able to merge.

@Korbeil is that OK with you? /cc @vdauchy

@Korbeil ready for review.

BTW I've also ran the entire test suite on v4 locally and it all works

Time: 07:48.009, Memory: 534.00 MB

OK (131 tests, 22699 assertions)

Ah, I need to update the composer.json files in subrepos too.

I guess the fact tests-lowest takes 17min is due to CS-Fixer running ?

@Korbeil yes, it needs to fix code style for thousands of files (both fixtures and generated). This is after we've marked the six biggest fixtures as skipped.

Great job!

@l-you thanks, all heavy lifting was done by @vdauchy 😃

@Korbeil done.