Some people tend to use lines full of '=' as a fancy way to format headings
in their commit messages in a rst-like style. However, the current parser
treats such lines as a beginning of a diff.
The only currently used tool that produces diffs with '=' lines is quilt in
the default configuration. However, even with quilt, the diff looks this
way:
Index: dir/file
===================================================================
--- dir.orig/file
+++ dir/file
@@ ...etc...
It's enough to match on the "Index:" line. The state of the state machine is
kept at 1 when it encounters the '=' line, thus it's safe to remove the
match on '=' completely.
[This prevents us from properly parsing metadata out of the changelog. -dcz ]
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Reviewed-by: Stephen Finucane <stephen@that.guru>
(cherry picked from commit
67faf96ab96d93252c89967ef766bcbe8214c0fc)
# state specified the line we just saw, and what to expect next
state = 0
# 0: text
- # 1: suspected patch header (diff, ====, Index:)
+ # 1: suspected patch header (diff, Index:)
# 2: patch header line 1 (---)
# 3: patch header line 2 (+++)
# 4: patch hunk header line (@@ line)
# 6: patch meta header (rename from/rename to)
#
# valid transitions:
- # 0 -> 1 (diff, ===, Index:)
+ # 0 -> 1 (diff, Index:)
# 0 -> 2 (---)
# 1 -> 2 (---)
# 2 -> 3 (+++)
line += '\n'
if state == 0:
- if line.startswith('diff ') or line.startswith('===') \
+ if line.startswith('diff ') \
or line.startswith('Index: '):
state = 1
buf += line