svn-create-patch gets confused at strings files because diff isn't treating the files as text (it sees the BOM in the UTF-16 file) Patch to be attached forces all files that svn-create-patch thinks are text to be treated as text.
Created attachment 13534 [details] Patch Trivial patch
If you create the patch with this change to svn-create-patch, run svn-unapply, then svn-apply, do you end up with the same changes once you're done? I'm wondering if there would be any text conversion funny-business where one two-byte UTF-16 character would get split into two one-byte UTF-8 characters in the process.
In my testing it appears to work correctly in the situation David described. I just tried adding a line of Japanese text to a file, checking md5 of the file, creating a patch, svn-unapply, svn-apply, checking the md5 of the file. At both points the md5 was identical.
Comment on attachment 13534 [details] Patch Can svn-apply and svn-unapply handle these kind of diffs? If not, we need to fix those first.
Comment on attachment 13534 [details] Patch r=me
Committed revision 20095.