On Sat, Sep 18, 2010 at 09:21:31AM +0200, Alan Barrett wrote: > On Sat, 18 Sep 2010, Daniel Shahaf wrote: > > "Fewest path components, shortest basename, total filename length". > > > > It's predictable, but it seems a bit arbitrary? > > It does seem a bit arbitrary, but I was able to rationalise it as > follows: "fewest path components" is good for choosing between > "subdir/file" and "../../other-branch/dir/subdir/file"; "shortest > basename" is good for choosing between "file.old" and "file", or > between "file" and "file.new"; "total filename length" is good for > choosing between "subdir/file" and "subdir.old/file". You still need a > tiebreaker of last resort after all that.
If there's a tie, we can use the "new" name. I suppose what UNIX patch does has been working well for everybody for a heck of a long time. There's no point in trying to improve on it unless we encounter real shortcomings in practice. Stefan
