conflict markers on recursive strategy with diff3 enabled

Hi!

I'm sitting down hunting for conflict examples and while looking through past conflicts in git I found one that has a format that I hadn't seen before. When merging the parents of 8b79343fc0 we get this:

<<<<<<< HEAD

/*
* Unlink the .pack and associated extension files.
* Does not unlink if 'force_delete' is false and the pack-file is
* marked as ".keep".
*/
extern void unlink_pack_path(const char *pack_name, int force_delete);
||||||| merged common ancestors

=======

That is with merge.conflictStyle set to diff3. What I would like to know is if the details about how the additional information that is not normally seen on a conflict has to be interpreted (to be read as "the part about the temporary branches"). I see some explanation about it in [1] but when checking inside "git help merge" all I see is that when you are using diff3 you will get the content of the parent but there's more stuff than just that. Is it documented somewhere?

Thanks!

[1]https://stackoverflow.com/questions/31392513/git-diff3-conflict-style-temporary-merge-branch

Hi Edmundo,

On Fri, Oct 25, 2019 at 12:19 PM Edmundo Carmona Antoranz <eantoranz@gmail.com> wrote:

The "||||||| merged common ancestors" section shows the content of the merge base, which itself contained conflict markers.

The way the merge base can itself have conflict markers is as follows: when you go to merge your two branches, git finds that there isn't a unique merge base for doing the three-way merge. You, in fact, have more than one merge base. And so, it first merges those (and if those don't have a unique merge base, then their merge bases must first be merged...and so on -- thus the name of "recursive" for the merge algorithm). When the algorithm runs into conflicts while merging the merge bases, it simply accepts the version of the file with the conflict markers as the correct resolution (because it has to pick something, and that is unlikely to erroneously match the outer contents on either side).

Thus when using diff3 conflict style, you can sometimes see nested conflict markers. And technically you can construct weird histories where you have arbitrarily deeply nested conflict markers as well, but it certainly isn't very common.

Hope that helps, Elijah

Hi Edmundo ,

On 25/10/2019 01:13, Edmundo Carmona Antoranz wrote:

- "Temporary merge branch"

I believe this happens when no merge-base is found between the two lines of development that are being merged, so a fake merge-base is created to allow the regular 3-way merge to occur.

It may be described in the technical section of the Documentation.

Philip

Philip Oakley <philipoakley@iee.email> writes:

s/no merge-base is/more than one merge-base are/; then a synthetic merge base is computed in order to perform the merge the end-user requested first, and the computation of that synthetic merge base may involve merge conflicts. After that, the merge logic pretends that the two versions being merged both derived from such a common ancestor version _with_ conflict markers in it, and may conflict with each other around that existing (inner) conflict markers to leave a rather messy result. It is particularly bad when both branches did the same thing only on one side the inner merge conflicted but not on the other side---one side of inner conflict marker (say "<<<<<<<<") may disappear when the merged branches made the same change (so "both side changed identically, resolve to their same result" 3-way merge rule would apply) and around that area the result will not have the conflict marker from the inner merge, and the other side will say "one side made into this shape, the other side made into that shape, from this common ancestor version" in the diff3 format, which would mean that we would see conflict marker lines from the inner merge between "|||||" and "=====" lines in the outer merge (i.e. the common ancestor version). The conflict markers of the inner merge are longer than those of the outermost merge, so you could sift them by careful eyeballing, but it is rather hard to read.