Fix memory-leak in CLI web server

[b-viguier]

Hi all 👋 🙂 ,
my local test suite is complaining about a memory leak (see valgrind report below)

Valgrind Report

==471513== 112 bytes in 3 blocks are definitely lost in loss record 22 of 30
==471513==    at 0x483B7F3: malloc (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==471513==    by 0x8BBB7B: __zend_malloc (zend_alloc.c:3089)
==471513==    by 0xA79298: zend_string_alloc (zend_string.h:150)
==471513==    by 0xA79338: zend_string_init (zend_string.h:172)
==471513==    by 0xA7DA55: php_cli_server_client_read_request_on_header_field (php_cli_server.c:1706)
==471513==    by 0xA780CC: php_http_parser_execute (php_http_parser.c:1121)
==471513==    by 0xA7DF17: php_cli_server_client_read_request (php_cli_server.c:1843)
==471513==    by 0xA7FCAC: php_cli_server_recv_event_read_request (php_cli_server.c:2557)
==471513==    by 0xA801C3: php_cli_server_do_event_for_each_fd_callback (php_cli_server.c:2661)
==471513==    by 0xA7BAE8: php_cli_server_poller_iter_on_active (php_cli_server.c:909)
==471513==    by 0xA8025D: php_cli_server_do_event_for_each_fd (php_cli_server.c:2681)
==471513==    by 0xA802E9: php_cli_server_do_event_loop (php_cli_server.c:2692)

Based on my research, it has been introduced in Php8.2 in this PR #8605
This is related to the zend_string containing the current header name during parsing.
Then, if your HTTP request doesn't include any header, there is no leak.

I'm not yet usual with the php-src test suite, but if you have an idea how to cover this leak by a test, I can give it a try.

Thanks a lot for your amazing work 🤗

[cmb69]

I'm not yet usual with the php-src test suite, but if you have an idea how to cover this leak by a test, I can give it a try.

I'm surprised that this leak hasn't been reported by at least one of our CI pipelines. I can reproduce that with a simple request to a not found page (using Windows debug heap, not even valgrind).

Anyway, thanks for the PR! I think this patch is good.

/cc @Girgias

[Girgias]

LGTM, @cmb69 I suppose you did the request manually and don't have a test? (as otherwise it would have been flagged already?)