posix_getgrnam_basic.phpt occassionally segfaults on MACOS_X64_DEBUG_NTS

[cmb69]

Description

Just happened again, and if I remeber correctly, happens not that rarely. Does anyone happen to know the cause? Can this be fixed? Would a dynamic xfail for that environment be a good idea to keep the tests green?

PHP Version

any

Operating System

any

[cmb69]

And just again: https://github.com/php/php-src/actions/runs/10705006077/job/29679340595

[devnexen]

note that macos arm64 and x86 64 does not use the same releases and the x86 64 arch won t get any image update for CI anymore. The bug(s) might be more related to that than architecture differences.

[cmb69]

note that macos arm64 and x86 64 does not use the same releases and the x86 64 arch won t get any image update for CI anymore. The bug(s) might be more related to that than architecture differences.

Oh, didn't know this, but still my premise is unchanged: whenever CI fails, something needs to be fixed; otherwise the broken windows effect may occur. And there is even an accepted RFC about keeping CI green. So what about marking this test xfail on macOS x64?

[devnexen]

Personally, I would say yes but maybe let's hear other opinions.

[nielsdos]

I tried to reproduce this a couple of times in the past on my macOS install, couldn't reproduce it. Maybe it's some sort of race, who knows.

[cmb69]

Just for the record: ext/phar/tests/zip/033a.phpt just segfaulted on MACOS_X64_DEBUG_NTS.

[iluuu1994]

Right. I have tried to identify the issue on multiple occasions. Even with asan, I get no stack trace or additional information. This might be a CI or macOS issue, I'm not sure.

[cmb69]

Since I don't have access to a macOS machine, I wonder whether the following would do:

if (PHP_OS_FAMILY === "Darwin" && php_unmame("m") === "x86_64") {
    die("xfail Test is flaky for unknown reasons");
}

[iluuu1994]

Note though that there are multiple phar tests that randomly segfault.

[cmb69]

If this is only about a couple of tests, I think we can mark them all as xfail. If there are too many (whatever that means exactly), it doesn't make sense to xfail them all. In that case we likely have to live with the occassionally red CI.