The snmp extension has many memory leaks. The bug is reported uptream at http://bugs.php.net/45405 I've fixed it and there is a patch included.
Created attachment 160139 [details] Fix snmp memory leak Applies against latest php
In general it's fine to include such patches in our patchsets, but I'd prefer to see a comment from upstream about the proposed patch first.
It's part of php-5.2.6-r7 which you can find in the php-testing overlay available via layman. Will probably go into the official tree when the next major php-patchset update happens (probably security-related). Just for reference, I took the latest patch from the upstream bug (after lots of whitespace cleanup....). Please test and report back.
(In reply to comment #3) > It's part of php-5.2.6-r7 which you can find in the php-testing overlay > available via layman. Will probably go into the official tree when the next > major php-patchset update happens (probably security-related). Thanks a lot! > > Just for reference, I took the latest patch from the upstream bug (after lots > of whitespace cleanup....). Yeah, i didn't find a way to upload a file in that bugzilla and pasting the patch in the comment generates lots of whitespaces. Sorry I didn't upload the patch here. > > Please test and report back. As I say in the upstream bug report, we are using php with that patch applied since 15 July and we didn't find any problems yet =) Thanks, Rodrigo
OK, I know that this is slated for the next patch update, but this is actually a *major* issue for anybody using packages like net-analyzer/cacti. Is there any way that we can get this patch inclusion accelerated into the main tree? Having "stable" packages which aren't quite so stable/usable isn't the best situation to remain in for an extended period of time.
(In reply to comment #2) > In general it's fine to include such patches in our patchsets, but I'd prefer > to see a comment from upstream about the proposed patch first. > I know its already applied to your svn tree. But I wanted just to let you know that it has been applied (as is) upstream (to php 6.0, 5.3 and 5.2) as you can see here: http://marc.info/?t=122064976800003&r=1&w=2 Thanks, Rodrigo
(In reply to comment #5) > OK, I know that this is slated for the next patch update, but this is actually > a *major* issue for anybody using packages like net-analyzer/cacti. Is there > any way that we can get this patch inclusion accelerated into the main tree? I agree, but I did not want to include a user-supplied (no offense intended!) patch, which evidently had severe problems (double frees) in its first version, to the tree without extensive testing and/or upstream approval. > Having "stable" packages which aren't quite so stable/usable isn't the best > situation to remain in for an extended period of time. Just curious, but this sounds like this was a regression -- was it? (In reply to comment #6) > I know its already applied to your svn tree. But I wanted just to let you know > that it has been applied (as is) upstream (to php 6.0, 5.3 and 5.2) as you can > see here: > > http://marc.info/?t=122064976800003&r=1&w=2 Thanks for the pointer (I'm closely following both php-internals and the cvs commit lists, btw), I just didn't get to it earlier as I was / will be on vacation again. Anyway, I moved php-5.2.6-r7 with the fix from the overlay to the main tree, so that we can stable it soon. I don't think we need to wait the full 30 days here, especially considering that the problem seems rather severe, that the only change affects only ext/snmp and that no other part of php and that this version already received some testing in the overlay. So, thanks again to Rodrigo Campos, for cleaning up the patch and pushing this upstream!
(In reply to comment #7) > > Having "stable" packages which aren't quite so stable/usable isn't the best > > situation to remain in for an extended period of time. > Just curious, but this sounds like this was a regression -- was it? It seems the leak was there from the beginning (it's present at least since 1999 as i could see from php-src CVS). I also found a report about the leak from about 2 years ago. We hit the leak as we were developing an application that uses the extension intensively and then wrote the patch.
(In reply to comment #8) > It seems the leak was there from the beginning (it's present at least since > 1999 as i could see from php-src CVS). I also found a report about the leak > from about 2 years ago. Ok, thanks. I think it has been in the tree long enough now. Arches, please test and stabilize: =dev-lang/php-5.2.6-r7 Target keywords: alpha amd64 arm hppa ia64 ppc ppc64 s390 sh sparc x86 ~x86-fbsd Already done: amd64 We are quick-stabling because of the seriousness of the bug. In the worst case this should only break things for USE=snmp users, in the best case it'll make USE=snmp usable again by fixing the mentioned memory leaks.
ppc and ppc64 done
x86 stable
alpha/ia64/sparc stable
Stable for HPPA.