fix memory leak in pdo_pgsql closeCursor (bug 69752) #1319
Conversation
|
currently investigating the test failures. Please stay tuned |
|
I have fixed the fix to deal with the failing test cases. Please note that even without this patch, |
|
@pilif what is the diff you've got for that test? |
|
@weltling the issue was on my end, running the tests with a clean php built with just |
|
@pilif, you mean the issue you originally reported isn't reproducable anymore? Then probably no need to merge the patch. Thanks. |
|
@weltling no. The original issue is reproducible and the memory leak exists. But the non-related test that was failing before doesn't fail any more. So on my machine now: Current PHP-5.6 HEAD:
With this PR applied:
|
|
@pilif ah, so i misunderstood then. Ok, gonna test it then. Btw did you check the same exists in 5.5 by chance? Thanks. |
|
@weltling - I haven't actually tested it, but the So I would say everything after 2d4b8e1 leaks and everything before crashes. |
|
@pilif tested your patch - the functionality is ok. I would right apply it, but it seems to have a small charset issue. You seem to have saved some weird chars in there which look like space. Please check what i see when opening the downloaded patch with a non multibyte charset
something is definitely looks like a unicode char, Visual Studio at least fails to compile it. Please put a plain space there :) Thanks. |
|
oh crap. I'm terribly sorry about this. This is a non-breaking space which happens to me all the time with my swiss german keyboard as the sign is Alt-Space and to make the | sign, I have to type Alt-7. So if I leave the finger on Alt for just a small moment too long, this sneaky sucker gets added which looks just like a normal space. I'll rebase a fixed patch and notify you again. Terribly sorry about this - neither gcc nor clang seemed to mind :/ |
the parent PDO closeCursor method resets the pdo_stmt_t's executed flag which is used by the postgres driver as a flag to check whether to allocate memory for the column data or not. This means that after the parent closeCursor() has been called, the pdo_pgsql driver will allocate a new buffer for the columns, so the existing buffer should be freed when the cursor is being closed.
|
Fine now, gonna merge. Thanks ) |
|
there we go. Squeaky clean without U+00a0 Sorry about wasting your time. |
|
Comment on behalf of ab at php.net: Merged into 5.5+. Thanks your work :) |
the parent PDO closeCursor method resets the pdo_stmt_t's executed flag
which is used by the postgres driver as a flag to check whether to
allocate memory for the column data or not.
This means that after the parent closeCursor() has been called, the
pdo_pgsql driver will allocate a new buffer for the columns, so the
existing buffer should be freed when the cursor is being closed.