Bugfix 72791: fix memory leak in PDO persistent connections

[keyurdg]

Attempt 2. This supersedes PR #2066

[keyurdg]

@nikic This is a real leak, see valgrind output here: https://bugs.php.net/bug.php?id=72791

[nikic]

@keyurdg I realize that it's a real leak, the question is more whether this is a "real" fix, or whether it effectively only suppresses the valgrind warning. I'm asking because I don't fully understand the reason why it's necessary to delay this free to the end of the request. Doesn't this indicate that refcounting is going wrong somewhere else?

[nikic]

Ah wait, I didn't see that this PR no longer delays the destruction, so that comment is no longer relevant.

[keyurdg]

@nikic just to clarify, the old PR actually free'd the memory. Without a list to save pointer references in, all references to the older pecalloc'd pdbh were lost when it was updated in the EG(persistent_list). zend_hash_str_update_mem() free's the enclosing zend_resource object (php_pdo_list_entry), but doesn't actually free the pointer saved inside it.

I do take the point that within a request the memory was not free'd as soon as feasible.

In any case, no longer relevant :)

[rlerdorf]

This looks correct to me