From 8c66da40fb74695bb7aab8d092e1d5d16258ebfe Mon Sep 17 00:00:00 2001
From: Sylvain Beucler <beuc@beuc.net>
Date: Sun, 29 Jun 2025 10:44:31 +0200
Subject: [PATCH] filter-repo: reduce export dump differences

While the order of files passed to `git fast-import` do not matter, we
can help users inspect --dry-run results if we can match the order that
fast-export uses.  fast-export uses a depth-first sorting order to
ensure that entries underneath a directory would appear before the
directory or some kind of file or symlink that has replaced a directory.
However, its depth-first ordering is implemented in a way that affects
things besides directories becoming symlinks or files -- in contrast to
normal lexicographic ordering where a substring would sort before a
longer string, fast-export reverses that.  This means that when we try
to inspect differences after a --dry-run, we may see a spurious
difference such as:

```
$ ./git-filter-repo --dry-run --proceed
$ diff -u .git/filter-repo/fast-export.original .git/filter-repo/fast-export.filtered
...
@@ -1451,25 +1329,23 @@
 D testcases/expected/case1-twenty
 D testcases/inputs/case1
 M 100755 0a13abf13c87a13ae56c1fc664baa8b44bdac1de testcases/t9390-repo-filter.sh
+M 100644 de3799fa8cd84a9ac56fad023ddf3da44eb444bb testcases/t9390/case1
 M 100644 e0c88454095bd887852791401a3b2336ad47d012 testcases/t9390/case1-filename
 M 100644 a1aa78ff11f3a40cafdcd9b531e8c1342e8a75e5 testcases/t9390/case1-ten
 M 100644 488cbd9d101b107809611940d1cd00a8f72e1b17 testcases/t9390/case1-twenty
-M 100644 de3799fa8cd84a9ac56fad023ddf3da44eb444bb testcases/t9390/case1
```

Fix this by modifying filter-repo to use the same sorting of paths as
fast-export when --dry-run is specified.

Signed-off-by: Sylvain Beucler <beuc@beuc.net>
---
 git-filter-repo | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/git-filter-repo b/git-filter-repo
index 39c8680a..149f6d8c 100755
--- a/git-filter-repo
+++ b/git-filter-repo
@@ -33,6 +33,7 @@ operations; however:
 import argparse
 import collections
 import fnmatch
+import functools
 import gettext
 import io
 import os
@@ -3942,7 +3943,29 @@ class RepoFilter(object):
         continue
       # Otherwise, record the change
       new_file_changes[change.filename] = change
-    commit.file_changes = [v for k,v in sorted(new_file_changes.items())]
+
+    # Use the git fast-export sorting algorithm for filenames
+    # https://github.com/git/git/blob/14de3eb34435db79c6e7edc8082c302a26a8330a/builtin/fast-export.c#L444-L448
+    def depth_first(a, b):
+      fn_a = a[0]
+      fn_b = b[0]
+
+      # Sort 'd/e' before 'd'
+      # first compare common length, then if equal give priority to longer one
+      min_len = min(len(fn_a), len(fn_b))
+      # memcmp equivalent https://docs.python.org/3.0/whatsnew/3.0.html#ordering-comparisons
+      cmp = (fn_a[:min_len] > fn_b[:min_len]) - (fn_a[:min_len] < fn_b[:min_len])
+      if cmp != 0:  # different content
+        return cmp  # return normal comparison
+      cmp = len(fn_b) - len(fn_a)
+      if cmp != 0:  # different size
+        return cmp  # longer one first
+
+      # 'R' (rename) entries last
+      cmp = (a[1].type == 'R') - (b[1].type == 'R')
+      return cmp
+    commit.file_changes = [v for k,v in sorted(new_file_changes.items(),
+        key=functools.cmp_to_key(depth_first))]
 
   def _tweak_commit(self, commit, aux_info):
     if self._args.replace_message: