Re: gitweb using "--cc"?

Junio C Hamano [off-list ref] writes:

Junio C Hamano [off-list ref] writes:

quoted

On top of your patch, it was quite easy ;-)

After the "Evil merge" in your test script, I added these:

...

        git diff-tree -M --cc HEAD

Sorry for the noise.  The test was broken.

Second try.  Sorry again for the previous noise.

This time I made sure I am getting double-colon output;
here is what I added after your test script:

        for i in a b c d e f g h i j k l m n; do echo $i; done >d
        git-update-index --add d
        git commit -m 'Add d'

        git checkout other
        git merge fast HEAD master

        echo o >>d
        git-update-index d
        git commit -m 'Edit d'

        git checkout master
        echo 'Another' >>a
        git-update-index a
        git commit -m 'Modify a'
        git merge --no-commit 'Merge' HEAD other
        echo Extra >>a
        mv d e
        echo extra >>e
        git update-index --add --remove a d e
        git commit -m 'Evil again'

        git diff-tree -M -c HEAD

But you are right.  Rename detection with combined diff has a
funny semantics:

    diff-tree 9df5f2d... (from parents)
    Merge: 1da47fa... 09eee61...
    Author: Junio C Hamano [off-list ref]
    Date:   Thu Feb 9 14:11:13 2006 -0800

        Evil again

    ::100644 100644 100644 8c3beaf... aad9366... c54c990... M	a
    ::100644 100644 100644 4f7cbe7... f8c295c... 19d5d80... M	e

This is showing that what was "d" was somehow magically called
"e" in the merge result with its own changes.  --cc output is
more interesting but the point is there is no sign of "d" in its
output, which does not feel right.

If we really care, we could show a status letter for each parent
(both are renames in this case but it is plausible one parent is
rename-edit and another is modify) and the original path in each
parent.

Does it matter?  I presume that a Porcelain that cares would
rather use the traditional "diff-tree -m -r" to look at diff
with each parent.  I dunno.

---

diff --git a/combine-diff.c b/combine-diff.c
index 15f369e..6d78305 100644
--- a/combine-diff.c
+++ b/combine-diff.c

@@ -829,7 +829,7 @@ const char *diff_tree_combined_merge(con
 	struct combine_diff_path *p, *paths = NULL;
 	int num_parent, i, num_paths;
 
-	diff_setup(&diffopts);
+	diffopts = *opt;
 	diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
 	diffopts.recursive = 1;
 

@@ -846,6 +846,7 @@ const char *diff_tree_combined_merge(con
 		struct commit *parent = parents->item;
 		diff_tree_sha1(parent->object.sha1, commit->object.sha1, "",
 			       &diffopts);
+		diffcore_std(&diffopts);
 		paths = intersect_paths(paths, i, num_parent);
 		diff_flush(&diffopts);
 	}

On 2/9/06, Junio C Hamano [off-list ref] wrote:

Does it matter?  I presume that a Porcelain that cares would
rather use the traditional "diff-tree -m -r" to look at diff
with each parent.  I dunno.

Yes, please preserve this behaviour.

I pulled today the diff-tree -c semantic change and now I see, in git archive:

$ git-diff-tree -r ca182053c7710a286d72102f4576cf32e0dafcfb
ca182053c7710a286d72102f4576cf32e0dafcfb
::100644 100644 100644 538d21d808b7ccc287e7bdd947f1583eadcda28b
30479b4a19805132a16facf6342b1438427486b7
59042d1bc9ee65063455b50a0968efb0b8182577 MM        Makefile

$ git-diff-tree -r -m ca182053c7710a286d72102f4576cf32e0dafcfb
ca182053c7710a286d72102f4576cf32e0dafcfb
:100644 100644 538d21d808b7ccc287e7bdd947f1583eadcda28b
59042d1bc9ee65063455b50a0968efb0b8182577 M      Makefile
:100644 100644 410b758aab7efc6d777f0344500f97b1cbc52946
6c47c3a3e1acb8badaadad42dfe3d0bd7a06cac3 M      entry.c
ca182053c7710a286d72102f4576cf32e0dafcfb
:100644 100644 30479b4a19805132a16facf6342b1438427486b7
59042d1bc9ee65063455b50a0968efb0b8182577 M      Makefile

Please _do not_ change this behaviour to make -m a no-op as stated in
"diff-tree -c raw output" patch message
(ee63802422af14e43eccce3c6dc4150a27ceb1a3).

qgit has the possibility to switch from "see all merge files" to "see
interesting only", so
we really need that difference between 'git-diff-tree -r' and
'git-diff-tree -r -m'

Anyhow I am very happy with this change because it broke qgit ;-) but when fixed
it will have a lot of code removed and will be faster too.

Thanks
Marco

Marco Costalba [off-list ref] writes:

On 2/9/06, Junio C Hamano [off-list ref] wrote:

quoted

Does it matter?  I presume that a Porcelain that cares would
rather use the traditional "diff-tree -m -r" to look at diff
with each parent.  I dunno.

Yes, please preserve this behaviour.
...
Please _do not_ change this behaviour to make -m a no-op as stated in
"diff-tree -c raw output" patch message
(ee63802422af14e43eccce3c6dc4150a27ceb1a3).

The one you pulled already contains another one to fix that
ee6380 change done by gittus.  What "diff-tree -r -m ca1820"
shows should be the same as traditional "diff-tree -r -m ca1820"
output.

What is different is "diff-tree ca1820".  It used to show
*nothing* only because it is a merge.  It now defaults to show
"diff-tree -c ca1820".

For the sake of backward compatibility we could change it to not
output anything, but I sort of feel that is backwards.  If a
Porcelain wants raw-diff for 1 (or more) parents, "diff-tree -r
-m" has been the way to do so before the ee6380 change, and that
output has not changed (well ee6380 might have changed it but
now it is fixed).