csound memory issue on OSX

[benhackbarth]

Hi all,

I’m trying to chase down a memory issue with the latest csound (6.05) on OSX built from git. I frequently get malloc errors when running this csd file: http://pastebin.com/Wndeh5Qf

Note that sometimes compilation completes successfully, other times it exits with an error, at other times the process hangs. Here is the shell output of this csd with the malloc error message: http://pastebin.com/n05Ycywj

And here is the output of valgrind using this csd on OSX 10.9.5: http://pastebin.com/fMBjUrXP

I cannot seem to reproduce the malloc fatal error when using valgrind, but it does consistently report some issues, perhaps especially pertinent around line 153.

Sorry that I cannot provide a csd that works independently of my soundfile database — removing/changing path names seems to affect whether or not this error occurs.

Best, Ben

[jpffitch]

Hard to see what to do as the input is so specific. e valgrind issue suggests it is in a string assignment but the code looks sufficiently defensive. You coud try running under gdb and look at the function copy_string_value, in particular the size od dst ad src

[benhackbarth]

copy_string_value seemed to be working as expected.

I've discovered the issue: line 117 uses '=' to assign a string variable (
"SspatAlgorithm = p75"). Changing this to strget seems to stop the periodic seg faults. If '=' isn't supported for strings (?), I would suggest throwing an error.

[megastep]

I'm also seeing some apparently related crashes on iOS with the code from the current devel branch. The address sanitizer in Xcode 7 picked up some invalid uses of strcpy() with overlapping addresses (basically copying a string to itself) in some of the code setting the configuration variables. I'm trying to investigate this further...

[benhackbarth]

Still having issues with this myself. My attempts to make a demonstrative csd which is untangled from local file dependancies has proven difficult. Will keep trying..

[megastep]

This patch helped me with some of the crashes I have been experiencing, some unsafe dealings with memory in the MIDI code, I suspect this might be the case in other areas. You may want to check it out.

https://github.com/audiokit/csound/commit/20f2547dc2ee74cbd175268d268a09521f13aba7

[vlazzarini]

@megastep can you submit your patch as a pull request ? It will be more easily dealt with

[megastep]

@vlazzarini I suppose, this branch has a lot of other commits I don't really want to bundle with this, but I'll do it from another branch.

[vlazzarini]

Thanks, that's helpful

[benhackbarth]

I’ve removed the file dependancies and, with this csd, I am getting the same valgrind errors.

http://pastebin.com/tHMe15Li

It seems to be a strange confluence of factors causing this, dependant on both the Udo and the strings. Perhaps it has to do with string handling in useropcd2? Really a shot in the dark though.

==23949== Invalid write of size 8
==23949== at 0x66AB91: _platform_memmove$VARIANT$Unknown (in /usr/lib/system/libsystem_platform.dylib)
==23949== by 0x324AD: string_copy_value (in /Users/ben/Library/Frameworks/CsoundLib64.framework/Versions/6.0/CsoundLib64)
==23949== by 0x1007933EF: ???
==23949== Address 0x101b619be is 62 bytes inside a block of size 66 alloc'd
==23949== at 0x47E1: malloc (vg_replace_malloc.c:300)
==23949== by 0x23F53: mmalloc (in /Users/ben/Library/Frameworks/CsoundLib64.framework/Versions/6.0/CsoundLib64)
==23949==
==23949== Invalid write of size 8
==23949== at 0x66AB91: _platform_memmove$VARIANT$Unknown (in /usr/lib/system/libsystem_platform.dylib)
==23949== by 0x324AD: string_copy_value (in /Users/ben/Library/Frameworks/CsoundLib64.framework/Versions/6.0/CsoundLib64)
==23949== by 0x10035AC1F: ???
==23949== by 0x2079E7: ??? (in /Users/ben/Library/Frameworks/CsoundLib64.framework/Versions/6.0/CsoundLib64)
==23949== by 0x1007933EF: ???
==23949== by 0x21C65: useropcd2 (in /Users/ben/Library/Frameworks/CsoundLib64.framework/Versions/6.0/CsoundLib64)
==23949== by 0x1000E6C1F: ???
==23949== by 0x1005D114F: ???
==23949== by 0x1007B2A8F: ???
==23949== by 0x1000DF44F: ???
==23949== by 0x10038F78F: ???
==23949== by 0x1013FFDBF: ???
==23949== Address 0x101b619be is 62 bytes inside a block of size 66 alloc'd
==23949== at 0x47E1: malloc (vg_replace_malloc.c:300)
==23949== by 0x23F53: mmalloc (in /Users/ben/Library/Frameworks/CsoundLib64.framework/Versions/6.0/CsoundLib64)
==23949==

[megastep]

This may or may not help with this bug, but I just made a bunch of weird crashes go away simply by reverting this commit: 24b7e94

[benhackbarth]

Thanks Stéphane. I should have mentioned: I’ve cloned into origin/release/6.05 to avoid any issues in the dev branch. So my errors in valgrind are coming from the 6.05 release version.

Best,
– ben

On 6 Aug 2015, at 09:52, Stéphane Peter notifications@github.com wrote:

This may or may not help with this bug, but I just made a bunch of weird crashes go away simply by reverting this commit: 24b7e94

—
Reply to this email directly or view it on GitHub.

[vlazzarini]

Hi Ben, could you check this now with the latest sources to see if the string fixes that were committed yesterday help your problem in any way?

[benhackbarth]

Seems like this has taken care of my problems, thanks Victor & John. Will report back if I discover any additional issues.

[vlazzarini]

OK, I'll close this issue