[lazerg]

The plain wrapper already clears the stat cache after reads, writes and flushes, but not after truncating, so filesize() kept returning the old size after ftruncate() or SplFileObject::ftruncate(). It now clears it on a successful truncate too.

Fixes #25232

[prateekbhujel]

Clearing the global stat cache here fixes filesize() for external lookups, but leaves the open stream's internal buffer (stream->readbuf) untouched when truncation occurs below the buffer horizon.

Unlike _php_stream_seek() which invalidates stream read/write positions (stream->readpos = stream->writepos = 0;), calling ftruncate() on a stream that has already performed buffered reads leaves stale bytes in memory, resulting in ghost reads past EOF:

$f = tempnam(sys_get_temp_dir(), 'trunc_');
file_put_contents($f, str_repeat('A', 100));

$h = fopen($f, 'r+');
fread($h, 10);      // stream pre-buffers 8KB into stream->readbuf
ftruncate($h, 10);  // truncated to 10 bytes on disk

var_dump(filesize($f));  // int(10)
var_dump(fread($h, 50)); // returns 50 bytes of stale buffer past EOF

fclose($h);
unlink($f);

Should stream buffer invalidation be handled in _php_stream_truncate_set_size() in main/streams/streams.c alongside this?

[lazerg]

I want to keep this PR to the stat cache. The stale read buffer is real, but stock 8.4.27 shows the same output, so it is a separate bug and should get its own issue and fix.

[prateekbhujel]

Writing directly to __DIR__ breaks out-of-tree builds where the source tree is mounted read-only (e.g. CI packaging setups or --srcdir testing). If the test crashes or aborts before reaching --CLEAN--, it also leaves untracked temporary files in the worktree.

Using tempnam(sys_get_temp_dir(), ...) (as in the original GH-25232 issue report) would be safer and keep the source tree clean.

[lazerg]

I kept __DIR__ because most tests in ext/standard/tests/file write there and clean up in --CLEAN--. For a read-only source tree, run-tests has --temp-source and --temp-target.