ignoring expectations sounds like a bad idea.. maybe we should turn such a test "risky" and/or can introduce a warning telling the author of the test, that not everything worked according to plan.

siltent ignore sounds like a bad idea.

I guess @sebastianbergmann has an opinion on that one

Good point.

"incomplete" should be the right status. If error_log file could not be created, test would still complete but will be marked as incomplete.

Should the "throw incomplete exception" be moved then into TestCase::runBare to always run https://github.com/sebastianbergmann/phpunit/blob/12.1.5/src/Framework/TestCase.php#L500-L502?

Another option is to revert/remove the newly introduced TestCase::expectErrorLog() until php-src allows to check it without temporary file. Currently, it suffers not only the issue here, but it is also not the fastest to having to create temporary file before each TestCase run.

but it is also not the fastest to having to create temporary file before each TestCase run.

please share a benchmark which proves your claim - noone else reported something like that since the change was released. how big is the impact?

<?php

$t1 = 0;
$t2 = 0;

for ($i = 0; $i < 1000; ++$i) {
    $ts = microtime(true);
    $capture = tmpfile();
    $ts2 = microtime(true);
    $t1 += $ts2 - $ts;
    $path = stream_get_meta_data($capture)['uri'];
    $t2 += microtime(true) - $ts2;
}

var_dump($t1 / $i);
var_dump($t2 / $i);

T2 does not call any FS, T1 is about 0.4 ms / per iteration (TestCase run).

I mean a real world impact, not microbenchmarks.
I am aware that not running stuff is faster then running stuff, but thats not helpful.

In another words about 1 second extra time per 2500 tests :)

This is the real world overhead of requiring FS.

Let @sebastianbergmann decide this thread incl. what to throw.