Hi, It turns out that it is memory corruption. From valgrind:
==30076== Invalid write of size 1 ==30076== at 0x402B788: memcpy (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==30076== by 0x408803: update_pre_post_images (apply.c:2036) ==30076== by 0x40A6F7: check_patch (apply.c:2274) ==30076== by 0x40B6E9: apply_patch (apply.c:3173) ==30076== by 0x40E18A: cmd_apply (apply.c:3954) ==30076== by 0x4057E3: handle_internal_command (git.c:308) ==30076== by 0x404FA2: main (git.c:556) ==30076== Address 0x5827c3d is 4 bytes after a block of size 265 alloc'd ==30076== at 0x402994D: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==30076== by 0x4DFDE5: xmalloc (wrapper.c:35) ==30076== by 0x40888B: update_pre_post_images (apply.c:2008) ==30076== by 0x40A6F7: check_patch (apply.c:2274) ==30076== by 0x40B6E9: apply_patch (apply.c:3173) ==30076== by 0x40E18A: cmd_apply (apply.c:3954) ==30076== by 0x4057E3: handle_internal_command (git.c:308) ==30076== by 0x404FA2: main (git.c:556) ==30076== ==30076== Invalid write of size 8 ==30076== at 0x402B7AD: memcpy (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==30076== by 0x408803: update_pre_post_images (apply.c:2036) ==30076== by 0x40A6F7: check_patch (apply.c:2274) ==30076== by 0x40B6E9: apply_patch (apply.c:3173) ==30076== by 0x40E18A: cmd_apply (apply.c:3954) ==30076== by 0x4057E3: handle_internal_command (git.c:308) ==30076== by 0x404FA2: main (git.c:556) ==30076== Address 0x5827c48 is 15 bytes after a block of size 265 alloc'd ==30076== at 0x402994D: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==30076== by 0x4DFDE5: xmalloc (wrapper.c:35) ==30076== by 0x40888B: update_pre_post_images (apply.c:2008) ==30076== by 0x40A6F7: check_patch (apply.c:2274) ==30076== by 0x40B6E9: apply_patch (apply.c:3173) ==30076== by 0x40E18A: cmd_apply (apply.c:3954) ==30076== by 0x4057E3: handle_internal_command (git.c:308) ==30076== by 0x404FA2: main (git.c:556) ==30076== The call trace is not quite right. The problem comes from the call at the end of match_fragement() (apply.c:2274) which says it's fine to do the update in-place. There, the memory gets overwritten. Making the newlines buffer in apply_one_fragment() double the size makes this go away. cmn -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
