copies: choose target directory based on longest match

copies: choose target directory based on longest match

If one side of a merge renames `dir1/` to `dir2/` and the subdirectory
`dir1/subdir1/` to `dir2/subdir2/`, and the other side of the merge
adds a file in `dir1/subdir1/`, we should clearly move that into
`dir2/subdir2/`. We already detect the directories correctly before
this patch, but we iterate over them in arbitrary order. That results
in the new file sometimes ending up in `dir2/subdir1/` instead. This
patch fixes it by iterating over the source directories by visiting
subdirectories first. That's achieved by simply iterating over them in
reverse lexicographical order.

Without the fix, the test case still passes on Python 2 but fails on
Python 3. It depends on the iteration order of the dict. I did not
look into how it's built up and why it behaved differently before the
fix. I could probably have gotten it to fail on Python 2 as well by
choosing different directory names.

Differential Revision: https://phab.mercurial-scm.org/D10115

--- a/mercurial/copies.py	Thu Mar 04 21:58:55 2021 +0100
+++ b/mercurial/copies.py	Thu Mar 04 16:06:55 2021 -0800
@@ -1096,11 +1096,17 @@
             b"   discovered dir src: '%s' -> dst: '%s'\n" % (d, dirmove[d])
         )
 
+    # Sort the directories in reverse order, so we find children first
+    # For example, if dir1/ was renamed to dir2/, and dir1/subdir1/
+    # was renamed to dir2/subdir2/, we want to move dir1/subdir1/file
+    # to dir2/subdir2/file (not dir2/subdir1/file)
+    dirmove_children_first = sorted(dirmove, reverse=True)
+
     movewithdir = {}
     # check unaccounted nonoverlapping files against directory moves
     for f in addedfilesfn():
         if f not in fullcopy:
-            for d in dirmove:
+            for d in dirmove_children_first:
                 if f.startswith(d):
                     # new file added in a directory that was moved, move it
                     df = dirmove[d] + f[len(d) :]

--- a/tests/test-rename-dir-merge.t	Thu Mar 04 21:58:55 2021 +0100
+++ b/tests/test-rename-dir-merge.t	Thu Mar 04 16:06:55 2021 -0800
@@ -294,3 +294,45 @@
   M t/t
   R a/s
   R a/t
+
+  $ cd ..
+
+
+Test that files are moved to a new directory based on the path prefix that
+matches the most. dir1/ below gets renamed to dir2/, and dir1/subdir1/ gets
+renamed to dir2/subdir2/. We want dir1/subdir1/newfile to move to
+dir2/subdir2/ (not to dir2/subdir1/ as we would infer based on just the rename
+of dir1/ to dir2/).
+
+  $ hg init nested-renames
+  $ cd nested-renames
+  $ mkdir dir1
+  $ echo a > dir1/file1
+  $ echo b > dir1/file2
+  $ mkdir dir1/subdir1
+  $ echo c > dir1/subdir1/file3
+  $ echo d > dir1/subdir1/file4
+  $ hg ci -Aqm initial
+  $ hg mv dir1 dir2
+  moving dir1/file1 to dir2/file1
+  moving dir1/file2 to dir2/file2
+  moving dir1/subdir1/file3 to dir2/subdir1/file3
+  moving dir1/subdir1/file4 to dir2/subdir1/file4
+  $ hg mv dir2/subdir1 dir2/subdir2
+  moving dir2/subdir1/file3 to dir2/subdir2/file3
+  moving dir2/subdir1/file4 to dir2/subdir2/file4
+  $ hg ci -m 'move dir1/ to dir2/ and dir1/subdir1/ to dir2/subdir2/'
+  $ hg co 0
+  4 files updated, 0 files merged, 4 files removed, 0 files unresolved
+  $ echo e > dir1/subdir1/file5
+  $ hg ci -Aqm 'add file in dir1/subdir1/'
+  $ hg merge 1
+  5 files updated, 0 files merged, 4 files removed, 0 files unresolved
+  (branch merge, don't forget to commit)
+  $ hg files
+  dir2/file1
+  dir2/file2
+  dir2/subdir2/file3
+  dir2/subdir2/file4
+  dir2/subdir2/file5
+  $ cd ..