0.23.3 fails to compile on Fedora 23 i686 #597

Activity

1. pypingou commented on Feb 22, 2016

   

   pypingou

   on Feb 22, 2016

   Author

   More actions

   Note the code was added commit 57507982 and line which might be responsible may be: 99dfce9#diff-57507982bc2ad39925ca8f387e172b29R720 but I don't see anything wrong w/ it, so I wonder if the problem isn't lower down the stack, no idea.
2. carlosmn commented on Feb 22, 2016

   

   carlosmn

   on Feb 22, 2016

   Member

   More actions

   It looks like the string might not be kept alive long enough, so libgit2 ends up reading nonsense. This looks like a dupe of #592 though.
3. pypingou commented on Feb 22, 2016

   

   pypingou

   on Feb 22, 2016

   Author

   More actions

   Indeed it's a dupe, should I close this one or do we prefer to close the other one?
4. pypingou commented on Feb 23, 2016

   

   pypingou

   on Feb 23, 2016

   Author

   More actions

   Any idea on how we could help debugging this?
5. carlosmn commented on Feb 23, 2016

   

   carlosmn

   on Feb 23, 2016

   Member

   More actions

   There's looking at the code to see if we make sure we keep a reference to the unmanaged string. Then there's also a way to build python so it doesn't pool allocations and you can throw valgrind and friends at it.
6. carlosmn commented on Feb 23, 2016

   

   carlosmn

   on Feb 23, 2016

   Member

   More actions

   This should keep the pointer alive.

   diff --git a/pygit2/repository.py b/pygit2/repository.py
   index f8a6cad..d7ca58e 100644
   --- a/pygit2/repository.py
   +++ b/pygit2/repository.py
   @@ -716,8 +716,10 @@ class Repository(_Repository):
                    format_options.abbreviated_size = abbreviated_size
                if always_use_long_format is not None:
                    format_options.always_use_long_format = always_use_long_format
   +            dirty_ptr = None
                if dirty_suffix:
   -                format_options.dirty_suffix = ffi.new('char[]', to_bytes(dirty_suffix))
   +                dirty_ptr = ffi.new('char[]', to_bytes(dirty_suffix))
   +                format_options.dirty_suffix = dirty_ptr
   
                buf = ffi.new('git_buf *', (ffi.NULL, 0))
7. pypingou commented on Feb 24, 2016

   

   pypingou

   on Feb 24, 2016

   Author

   More actions

   thanks I'll test this patch :)
8. pypingou commented on Feb 24, 2016

   

   pypingou

   on Feb 24, 2016

   Author

   More actions

   Worked perfectly, thanks! http://koji.fedoraproject.org/koji/taskinfo?taskID=13116069
9. pypingou commented on Feb 24, 2016

   

   pypingou

   on Feb 24, 2016

   Author

   More actions

   So as far as I am concerned this can be closed as fixed, I could build 0.23.3 for Fedora 23, 24 and rawhide successfully on all arches (i686, x86_64 and arm).

   If you submit the patch for inclusion, I can vouch that it fixed the unit-tests for us.

   Many thanks for your help!
10. jdavid

    closed this as completedin #599on Feb 25, 2016