AppVeyor: only load minimal set of exts; don't run unsupported test s…

[cmb69]

…uites

[nikic]

Is that part of the old tmp-php.ini? So our opcache test on Windows is actually testing file cache and not SHM cache?

[cmb69]

opcache.file_cache does not disable SHM caching on Windows; that would require opcache.file_cache_only to be enabled (but defaults to off). The file cache is relevant for opcache.file_cache_fallback which is enabled by default.

[nikic]

Right, that makes sense.

[nikic]

Part of the test failures are due to duplicate exts in EXTENSIONS (should possibly check that and BORK). Most of them are because we don't pass through loaded extensions to proc_open based server tests...

[cmb69]

Ugh! I'll have a look at this, but it may take some days.

[nikic]

I think I've fixed all the incorrect/missing EXTENSIONS sections, so if you rebase this should have just the proc_open issues now. Not entirely sure what to do about those. We should really forward -c/-d from the run-tests invocation of the main test file (for this and other reasons), but I'm not sure how exactly that should work...

[nikic]

Though to make some quick progress here, it may make sense to keep an unconditional load of openssl in particular for now. Seems to be the only one affected from a quick glance.

[cmb69]

I did that now. Let's wait for the test results.

[nikic]

Down to two failures:

Bug #77578 (Crash when php unload) [C:\projects\php-src\ext\com_dotnet\tests\bug77578.phpt]
Bug #68547 (Exif Header component value check error) [C:\projects\php-src\ext\exif\tests\bug68547.phpt]

The first one is an exec() test so has the forwarding problem.

I'm a bit stumped on the second one though. Not seeing anything that should be affected by extension loading there. It failed on both jobs though, so it doesn't look spurious either.

[cmb69]

I'm a bit stumped on the second one though. Not seeing anything that should be affected by extension loading there. It failed on both jobs though, so it doesn't look spurious either.

I guess we need to load mbstring as well.

[nikic]

@cmb69 Oh, nice catch! I wasn't aware exif has an optional dependency on mbstring.

Looks like only the com_dotnet test fails now, so I'd suggest to also unconditionally load that ext for now, and then this should be good to go.

[cmb69]

I've fixed the COM test by loading com_dotnet via command line argument. I've just pushed a final commit to get rid of the auto-generated php.ini altogether.

It seems to me that the test suite runs considerably faster now. :)

[nikic]

LGTM assuming CI passes. Thanks!

[cmb69]

assuming CI passes.

;) I'll need to have a closer look; also, there are more tests skipped now.

[nikic]

Presumably we're missing this part:

php-src/win32/build/confutils.js

Lines 2039 to 2044 in 9871a62

	FSO.CreateFolder(dir);
	/* Fallback is implied, if filecache is enabled. */
	INI.WriteLine("opcache.file_cache=" + dir);
	INI.WriteLine("opcache.enable=1");
	INI.WriteLine("opcache.enable_cli=1");

[cmb69]

Ah, right!

[cmb69]

CI is green, and results look okay, so I think this is good to be merged.