Add tests for uncastable streams and dataloss streams

[Girgias]

Moreover, stop using php_stream_can_cast() just before a php_stream_cast()

This is a follow-up from #10093.
I mimicked the online test by using the same set-up as testing the HTTP wrapper locally, as I couldn't seem to trigger a data loss in any other available by default streams wrappers.

However, I think GD is broken somehow, as even prior to the change it was hanging on my non-castable user stream wrapper. Maybe it's because of how I'm trying to pass it as a wrapper?

[codecov-commenter]

Codecov Report

Merging #10173 (5f42a46) into master (c90c944) will decrease coverage by 0.12%.
The diff coverage is 87.08%.

❗ Current head 5f42a46 differs from pull request most recent head f5fe9f3. Consider uploading reports for the commit f5fe9f3 to get more accurate results

@@            Coverage Diff             @@
##           master   #10173      +/-   ##
==========================================
- Coverage   68.34%   68.21%   -0.13%     
==========================================
  Files         852      852              
  Lines      313348   313345       -3     
==========================================
- Hits       214156   213754     -402     
- Misses      99192    99591     +399     

Impacted Files	Coverage Δ
Zend/Optimizer/zend_inference.h	72.72% <ø> (ø)
Zend/zend_alloc.c	74.24% <ø> (ø)
Zend/zend_builtin_functions.c	77.44% <0.00%> (+0.10%)	⬆️
Zend/zend_execute.h	93.75% <ø> (ø)
Zend/zend_hash.c	80.09% <ø> (ø)
Zend/zend_hash.h	90.71% <ø> (ø)
Zend/zend_ini_scanner.l	90.47% <ø> (ø)
Zend/zend_smart_str.h	96.29% <ø> (ø)
Zend/zend_sort.c	100.00% <ø> (ø)
Zend/zend_string.h	85.07% <ø> (ø)
... and 59 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[Girgias]

So the GD stream casting test segfaults on FreeBSD meaning there is likely a bug in it. But I can't reproduce it on Linux :/

[Girgias]

Okay reverted the can_cast drop, added a test where a stream cast actually loses data based on @arnaud-lb's script and suppress the warnings on the casts which don't actually lose data.

The only thing which I'm still confused by are the new GD tests that are inconsistent between platforms and also segfaults on FreeBSD, but maybe it's best to split those into a new PR so that we can merge all these new tests.

[cmb69]

The only thing which I'm still confused by are the new GD tests that are inconsistent between platforms and also segfaults on FreeBSD, but maybe it's best to split those into a new PR so that we can merge all these new tests.

Yeah, this might be the best solution for now.

[Girgias]

Everything should be green now, but will let CI run and squash/merge afterwards and open the GD changes in it's own PR again.

[cmb69]

Got some "reds" already. :(

[Girgias]

The failed test is:

 Test posix_fpathconf [ext/posix/tests/posix_fpathconf.phpt]

So I'll have a look at this.

[Girgias]

The failed test is:

 Test posix_fpathconf [ext/posix/tests/posix_fpathconf.phpt]

So I'll have a look at this.

The issue seems to be that the $fd = fopen(sys_get_temp_dir(), "r"); line is flacky, so I've used __DIR__ instead.

[Girgias]

@devnexen do you know why the Test posix_fpathconf [ext/posix/tests/posix_fpathconf.phpt] test is being flaky?

[cmb69]

do you know why the Test posix_fpathconf [ext/posix/tests/posix_fpathconf.phpt] test is being flaky?

Addding a posix_get_last_error() might clarify.

[devnexen]

@devnexen do you know why the Test posix_fpathconf [ext/posix/tests/posix_fpathconf.phpt] test is being flaky?

not at the moment but will have a look when I get the chance.

[bukka]

@Girgias I had another look on this one and all makes sense to me (the code looks good to me). It just needs rebasing / resolving conflicts and making sure that win jobs also pass.

[Girgias]

I don't understand why the Windows job now hangs compared to 6 months ago.

[iluuu1994]

@Girgias It would be easiest to just remove the --no-progress flag temporarily.

[bukka]

Well if GH actions Win jobs don't work, we should maybe put back AppVeyor or something that runs win. I don't really feel comfortable to merge this without any Win tests.

[bukka]

Specifically I'm talking about sapi_windows_vt100_support test...

[Girgias]

Well if GH actions Win jobs don't work, we should maybe put back AppVeyor or something that runs win. I don't really feel comfortable to merge this without any Win tests.

Appveyor also doesn't work.

I asked @arnaud-lb as he helped design the test and he told me those are not meant to actually work on Windows, so I'll skip them.
Indeed popen('echo 1; echo 2; rm $file') would be launched in cmd.exe, thus the file never gets deleted and we have an infinite loop.

[iluuu1994]

These look related:
https://github.com/php/php-src/actions/runs/5514770188/jobs/10054432846

[Girgias]

They do, thanks!