filter-repo: allow rename collision if files are identical

We previously would abort if we had been requested to rename files and
that caused two different files to go to the same path.  However, if
the files have identical contents and mode, then we can treat the
request as a desire from the user to just coalesce the extra copies.

Signed-off-by: Elijah Newren <newren@gmail.com>

git-filter-repo

@@ -2843,16 +2843,30 @@ class RepoFilter(object):
         continue # Filtering criteria excluded this file; move on to next one
       if change.filename in new_file_changes:
         # Getting here means that path renaming is in effect, and caused one
-        # path to collide with another.  That's usually bad, but sometimes
-        # people have a file named OLDFILE in old revisions of history, and they
-        # rename to NEWFILE, and would like to rewrite history so that all
-        # revisions refer to it as NEWFILE.  As such, we can allow a collision
-        # when (at least) one of the two paths is a deletion.  Note that if
-        # OLDFILE and NEWFILE are unrelated this also allows the rewrite to
-        # continue, which makes sense since OLDFILE is no longer in the way.
+        # path to collide with another.  That's usually bad, but can be okay
+        # under two circumstances:
+        #   1) Sometimes people have a file named OLDFILE in old revisions of
+        #      history, and they rename to NEWFILE, and would like to rewrite
+        #      history so that all revisions refer to it as NEWFILE.  As such,
+        #      we can allow a collision when (at least) one of the two paths
+        #      is a deletion.  Note that if OLDFILE and NEWFILE are unrelated
+        #      this also allows the rewrite to continue, which makes sense
+        #      since OLDFILE is no longer in the way.
+        #   2) If OLDFILE and NEWFILE are exactly equal, then writing them
+        #      both to the same location poses no problem; we only need one
+        #      file.  (This could come up if someone copied a file in some
+        #      commit, then later either deleted the file or kept it exactly
+        #      in sync with the original with any changes, and then decides
+        #      they want to rewrite history to only have one of the two files)
+        colliding_change = new_file_changes[change.filename]
         if change.type == 'D':
           # We can just throw this one away and keep the other
           continue
+        elif change.type == 'M' and (
+            change.mode == colliding_change.mode and
+            change.blob_id == colliding_change.blob_id):
+          # The two are identical, so we can throw this one away and keep other
+          continue
         elif new_file_changes[change.filename].type != 'D':
           raise SystemExit("File renaming caused colliding pathnames!\n" +
                            "  Commit: {}\n".format(commit.original_id) +