Submit by mail

[Matthieu Moy]

Michael Ellerman <michael at ellerman.id.au> writes:

>> One of the problems is that the line separator is not always constant--
>> some files mix LF and CRLF.  Another is that whitespace at the end of
>> lines is often eaten, or used to indicate a soft break in format=flowed
>> text.
>
> Yeah, I feel your pain. Unfortunately I think it's human to consider
> whitespace as insignificant in any format that _looks like text_. The
> easy option might be just to ignore the textual diff, and use a base64
> encoded one. Although then some idiot user will edit the diff and expect
> that to hold :)

You can also keep the textual diff, and add some somewhat encoded
information after it, that would be ignored by patch, but used by bzr
merge.

Like

--- foo 2006-06-14 09:24:02.824978750 +0530
+++ bar 2006-06-14 09:24:15.609777750 +0530
@@ -4,7 +4,7 @@
 a
 a
 a
-a
+b
 a
 a
 a
# Line-endings: uuuuwuuu

(my encoding here meaning "unix for the 4 first lines of the patch,
windows for the 5th one, and unix for the 3 last ones", but one could
imagine anything else here)

I'd find it quite ugly, but at least, you'd keep the patch, and the
reproducible result.

-- 
Matthieu