diff attribute ignored by show and log -p

% cat .git/info/attributes *.xib diff=xibdiff % cat $(git config diff.xibdiff.command) #!/bin/sh trap "rm -f \"$2.tmp\" \"$5.tmp\"" 0 1 2 3 15 ibtool --all "$2" > "$2".tmp ibtool --all "$5" > "$5".tmp colordiff -u "$2.tmp" "$5.tmp"

Works great for things like:

% git diff <commit1> <commit2> -- /path/to/*.xib

But is apparently ignored by "git log -p" and "git show" which just use internal diff. Is this behavior intentional?

j.

On Wed, Dec 16, 2009 at 11:46 PM, Jay Soffian <jaysoffian@gmail.com> wrote:

Ah, --ext-diff, and the reasoning behind requiring it for log/show is explained in 72909be.

j.

Quoting Jay Soffian <jaysoffian@gmail.com>

The need to give --ext-diff is mentioned in 72909be (Add diff-option --ext-diff, 2007-06-30) but its log message doesn't 'explain' why external diff isn't used by default and you need to pass that extra option.

Probably --ext-diff should be the default?

On Thu, Dec 17, 2009 at 8:17 AM, Nanako Shiraishi <nanako3@lavabit.com> wrote:

"To prevent funky games with external diff engines, git-log and friends prevent external diff engines from being called."

Seemed reason enough to me.

Or available via a config option anyway.

j.

On Thu, Dec 17, 2009 at 11:44:16AM -0500, Jay Soffian wrote:

I don't remember discussing it at that time, but much later when touching the external diff code and adding textconv, Junio and I came to the conclusion that textconv was reasonable to do as part of "git log -p", as the result is just an internal conversion that still results in a text diff. Whereas an external diff might be terribly confusing, as it could be spawning graphical viewers or producing output which does not match well with the log output.

If you were going to do such a config option, it might make sense to make it an attribute of the diff driver rather than a global "use external diff". Then each diff driver could say "Yes, I am reasonable to be run as part of log -p".

-Peff