git segfault when concurrently writing conflicting strings with leading newline about acidfs HOT 5 OPEN

It's a bug in git. I was able to reproduce with a shell script independently of AcidFS:

https://gist.github.com/chrisrossi/f09c8bed70b364f9f12e

Maybe you can show that to the git folks.

HTH.

from acidfs.

FWIW, even with the malloc error, the output of merge-tree is still correct and AcidFS still works as expected. You just get a nasty-gram to stderr.

from acidfs.

Thank you! I wasn't aware that the git test case would be so simple. Sorry for the extra work.. I guess the fact that that the code starting at line 854 in __init__.py is not visited made me think that it had to do with some specific merge attempt...

I sent a bug report to [email protected]. Not visible on gmane yet, I'll update this post.

EDIT: http://comments.gmane.org/gmane.comp.version-control.git/286252

from acidfs.

Bad news. It looks like merge-tree is something like a proof-of-concept, follows a different merge-logic than git merge and probably has more design/implementation flaws yet to be discovered...

Jeff King suggests an alternative for a tree-level merge:

If you just care about doing the tree-level merge
without content-level merging inside blobs, you can do it in the index
like:

export GIT_INDEX_FILE=temp.index
base=$(git merge-base $ours $theirs)
git read-tree -i -m --aggressive $base $ours $theirs

If you want to do content-level resolving on top of that, you can do it
with:

git merge-index git-merge-one-file -a

though it will need a temp directory to write out conflicts and
resolved content.

I can't tell whether this is useful for acidfs, but maybe the merge logic code should be changed to rely on more robust git tools.

Anyway, there now is a patch for the segfault issue.

from acidfs.

Interesting. I'll take a look when I get a chance. Thanks! If the output
format is the same or very similar, it shouldn't be too hard to switch.

Chris

On Fri, Feb 19, 2016 at 7:59 AM, epistemery [email protected]
wrote:

Bad news. It looks like merge-tree is something like a proof-of-concept,
follows a different merge-logic than git merge and probably has more
design/implementation flaws yet to be discovered...

Jeff King suggests
http://permalink.gmane.org/gmane.comp.version-control.git/286429 an
alternative for a tree-level merge:

If you just care about doing the tree-level merge
without content-level merging inside blobs, you can do it in the index
like:

export GIT_INDEX_FILE=temp.index
base=$(git merge-base $ours $theirs)
git read-tree -i -m --aggressive $base $ours $theirs

If you want to do content-level resolving on top of that, you can do it
with:

git merge-index git-merge-one-file -a

though it will need a temp directory to write out conflicts and
resolved content.

I can't tell whether this is useful for acidfs, but maybe the merge logic
code should be changed to rely on more robust git tools.

Anyway, there now is a patch
http://permalink.gmane.org/gmane.comp.version-control.git/286298 for
the segfault issue.

—
Reply to this email directly or view it on GitHub
#3 (comment).

from acidfs.