[lazerg]

php_mysqlnd_rset_header_read() stored the zend_ulong from php_mysqlnd_net_field_length() in a size_t. On x32 that is 32 bits, so MYSQLND_INVALID_NET_FIELD_LENGTH was truncated and a minimal OK packet was rejected as an over-read. Making the variable a zend_ulong fixes it.

Fixes GH-24609

[kamil-tekiela]

Can this not be tested with a real server?

[lazerg]

A real server does hit this path (any plain OK reply, like an INSERT), so existing mysqli tests already cover it and would fail on an x32 build. But on 64-bit they pass with or without the fix, so I can't make a real-server test that fails before the fix. The fake server just lets me pin the exact packet.

[kamil-tekiela]

So we don't need any new test. The problem wasn't a missing test, but not running it under that specific setup.