Bugfix 72791: fix memory leak in PDO persistent connections#2066
Bugfix 72791: fix memory leak in PDO persistent connections#2066keyurdg wants to merge 1 commit intophp:PHP-7.0from
Conversation
|
I'm not sure I understand this change. Is this a root-cause fix or is it only symptomatic? This will (unless I misunderstood) still leak the dbh until the end of the request, which is still a leak. |
|
|
||
| ZEND_BEGIN_MODULE_GLOBALS(pdo) | ||
| zend_long global_value; | ||
| HashTable *pdo_pdbh_to_delete; |
There was a problem hiding this comment.
Changing globals would likely breach non core PDO drivers. 7.1 is already in feature freeze, not talking about 7.0, so were a no go. However, from what I see, pdo_pdbh_to_delete is not required to be accessed by drivers. Thus, using a global ZEND_TLS variable instead of a global would be suitable and wouldn't breach binary compat.
Thanks.
There was a problem hiding this comment.
To clarify, are you saying that ABI breaks for 7.1 are no longer acceptable?
There was a problem hiding this comment.
Feature freeze implies no ABI/ABI changes, at least without a weighty reason and consideration. Should be to agree by RMs, just in case. Not relevant here anymore. My accent was more on mentioning the alternative way without breaching globals.
Thanks.
There was a problem hiding this comment.
@weltling If that's the case, can we please have an announcement on internals? I assumed that ABI breaks (not necessarily source breaks) are fine until the final release, similar to how it has been with previous versions.
There was a problem hiding this comment.
There was an announce from Davey http://news.php.net/php.internals/94330 . But i don't think it was just ok with the previous versions as well, AFAIR (and how i handled it) the practice is always to freeze both API/ABI, like fe https://fedoraproject.org/wiki/Updates_Policy#Beta_to_Pre_Release and other projects do. But there's always room for an exception, sometimes a breach is acceptable or even unavoidable, so @dshafik and @krakjoe would be in charge for that.
Thanks.
|
See new PR #2067 for a ABI compliant implementation. |
No description provided.