Fix GH-16955: Use ephemeral ports for OpenSSL server client tests

[bukka]

This introduces usage of ephemeral port for majority of OpenSSL test.

The only tests that need more work are following:

• stream_server_reneg_limit.phpt - quite manual reneg test - might need some update anyway
• bug77390 - proxy scenario - it needs more work in empheral port support
• ext/standard/tests/streams/stream_context_tcp_nodelay_server.phpt - this is a bit non standard test - need to figure out how to best address it there

[cmb69]

The only tests that need more work are following:

If there are only three left, I'm fine with manually choosing hard-coded port numbers, or adding a conflict marker.

[bukka]

I rewrote that stream_context_tcp_nodelay_server.phpt which was just wrong and then added support for multiple workers which fixes bug77390.phpt . It's a bit more complex now but it works. The only one left is stream_server_reneg_limit.phpt but that's skipped on Windows. It might need some rewrite or we will just eventually drop it (reneg is not really used and not even supported in TLS 1.3). I will leave it for now and if there are any issues with it, I will look into it.

[bukka]

hmm that stream one no longer works on Windows. Will check it out next week. Also will need to check if there are more tests in later versions.

[cmb69]

The only one left is stream_server_reneg_limit.phpt but that's skipped on Windows.

See #17193.

that stream one no longer works on Windows.

I'll have a look as soon as possible.

[cmb69]

I have not been able to reproduce failing of stream_context_tcp_nodelay_server.phpt locally, so I've triggered the Windows job again to check whether this is flaky.

Also will need to check if there are more tests in later versions.

Indeed; in master there are also gh11418.phpt and bug51056.phpt (and maybe more).

[cmb69]

stream_context_tcp_nodelay_server.phpt failed again for the same reason (no output). Might be a crash in the server code, or due to writing to STDERR (could use fwrite() instead of fprintf() although unlikely to cause issues), or whatever. I still haven't been able to reproduce locally.

[bukka]

@cmb69 So I did a bit of debugging and in this build https://github.com/php/php-src/actions/runs/12559675194/job/35015813633 with extra debug output, it gave me:

     client start: 127.0.0.1:%d
     server-accepted
003+ 
004+ Fatal error: Uncaught Error: Call to undefined function socket_import_stream() in D:\a\php-src\php-src\ext\openssl\tests\ServerClientTestCase.inc(124) : eval()'d code:15
005+ Stack trace:
006+ #0 D:\a\php-src\php-src\ext\openssl\tests\ServerClientTestCase.inc(124): eval()
003- server-imported
004- si:delay
005- conn-imported
006- server-delay:conn-nodelay

I decided to skip it for now on Windows so I can get it merged as I need those changes for another thing (well I don't want to port that thing later) that I'm looking to and created #17308 which debug output. Not sure at the moment why socket_import_stream is not present but might be related to how the worker is created on Win:

php-src/ext/openssl/tests/ServerClientTestCase.inc

Lines 70 to 77 in b873176

	if (defined("PHP_WINDOWS_VERSION_MAJOR")) {
	$ini = php_ini_loaded_file();
	$cmd = sprintf(
	'%s %s "%s" %s',
	PHP_BINARY, $ini ? "-n -c $ini" : "",
	__FILE__,
	WORKER_ARGV_VALUE
	);

. Please comment to that new PR if you have any idea - feel free to also push the changes into it if you want.