Re: [PATCH v6 4/4] reset: use "unpack_trees()" directly instead of "git read-tree"

Hi and happy new year!

On vendredi 01 janvier 2010, Junio C Hamano wrote:

Junio C Hamano [off-list ref] writes:

quoted

At least disallowing means that the user _is notified_ and has to
manually deal with the situation.  Pretending it succeeded by resetting
only the index while still leaving the conflicted state in the work
tree intact is a bit worse in that sense.

quoted

diff --git a/Documentation/git-reset.txt b/Documentation/git-reset.txt
index c9044c9..b40999f 100644
--- a/Documentation/git-reset.txt
+++ b/Documentation/git-reset.txt

@@ -122,7 +122,7 @@ entries:
        X       U     A    B     --soft  (disallowed)
                                 --mixed  X       B     B
                                 --hard   B       B     B
-                                --merge (disallowed)
+                                --merge  X       B     B

IOW, I think the result should be "B B B" instead of "X B B" in this
case.

A squashable fix-up on top of your patch to match the wish in the part
you quoted from 9e8ecea (Add 'merge' mode to 'git reset', 2008-12-01)
would look like this, I think.

It does three things:

 - Updates the documentation to match the wish of original "reset
--merge" better, namely, "An unmerged entry is a sign that the path
didn't have any local modification and can be safely resetted to whatever
the new HEAD records";

 - Updates read_index_unmerged(), which reads the index file into the
   cache while dropping any higher-stage entries down to stage #0, not to
   copy the object name recorded in the cache entry.  The code used to
   take the object name from the highest stage entry ("theirs" if you
   happened to have stage #3, or "ours" if they removed while you kept),
   which essentially meant that you are getting random results and didn't
   make sense.

   The _only_ reason we want to keep a previously unmerged entry in the
   index at stage #0 is so that we don't forget the fact that we have
   corresponding file in the work tree in order to be able to remove it
   when the tree we are resetting to does not have the path.  In order to
   differentiate such an entry from ordinary cache entry, the cache entry
   added by read_index_unmerged() records null sha1.

 - Updates merged_entry() and deleted_entry() so that they pay attention
   to cache entries with null sha1 (note that we _might_ want to use a
   new in-core ce->ce_flags instead of using the null-sha1 hack). They
   are previously unmerged entries, and the files in the work tree that
   correspond to them are resetted away by oneway_merge() to the version
   from the tree we are resetting to.

Please take this with a grain of salt as I am under slight influence of
CH3-CH2-OH while writing it, and I usually almost never drink.

I like this patch. It seems to improve the behavior of the --keep option for 
unmerged entries too.

The previous behavior was:

    working index HEAD target         working index HEAD
    ----------------------------------------------------
     X       U     A    B     --keep   X       B     B
     X       U     A    A     --keep   X       A     A

and now it is:

    working index HEAD target         working index HEAD
    ----------------------------------------------------
     X       U     A    B     --keep  (disallowed)
     X       U     A    A     --keep   X       A     A

And I think it is better, as it is more consistent with the behavior when 
there are no unmerged entries.

The only problem is that when it fails the error message is something like:

    error: Entry 'file1' would be overwritten by merge. Cannot merge.
    fatal: Could not reset index file to revision 'HEAD^'.

which is not very nice.

I will send a RFC patch series so people interested can test this.

Thanks,
Christian.