Changeset unified diff doesn't work with new files

[paolo]

hi, this issue came up while try to track madwifi (madwifi.org) devel: I'd grab CS then apply in order, but sometime - seems when files are moved around - resulting CS is broken, i.e. don't match what "svn -r …" would report. Which sounds pretty wierd - isn't just *that* same command used by TRAC do generate the CS? For ref. pls see ​http://article.gmane.org/gmane.linux.drivers.madwifi.devel/3246/match=ath+hal+directory

[Christian Boos]

It's not the same command. Trac uses Python's difflib to compare the two contents.

Plus, I was not really able to relate changeset the ​http://madwifi.org/changeset/1742 to the example in the mail… can you be more precise about the issue?

Also, as 0.10 is out, it might be a good idea to make a quick experimental install, and check if the problem is still there.

[Matthew Good]

Yeah, I was very confused by the description for this as well, but after checking the unified diff I can see that the diff headers generated for newly added files are incorrect. This is what Trac generates:

--- (revision )
+++ ah_target.inc (revision 1742)
@@ -1,0 +1,36 @@

It should be:

--- ah_target.inc (revision 0)
+++ ah_target.inc (revision 1742)
@@ -0,0 +1,36 @@

Patch will take the filename from the "—-" line, so it needs a filename there. Also patch gets confused by the "-1" indicated that the patch starts at line 1 of the old file and there was no line 1. So, that needs to be line 0.

[Christian Boos]

I don't remember the details, but I had to make a fix for the Add case in the TracDiff branch quite a while ago. Probably this is needed in the Copy and Rename case too.

[Matthew Good]

AFAICT this should be working. Can you upgrade Trac on madwifi.org to 0.10.3.1 and check that this is working?

[Matthew Good]

I'll take the lack of response as an indication this is not still a problem. Please reopen with more information if it still exists.