Re: [PATCH] Diff updates, fixing pathspec and rename/copy interaction.

quoted

quoted

quoted

quoted

"LT" == Linus Torvalds [off-list ref] writes:

LT> I would much prefer a _much_ simpler fix at least for the pathname part,
LT> which is to just always require that pathspec handling is done _first_.

LT> Why? Because that's fundamentally how git-diff-tree has to work, and it's
LT> how my mental model has always been: the path limitations are a
LT> first-order filter, and if you give a directory, the end result should
LT> always look exactly as if that directory was a project of its own.

Enlightment!

My initial thinking was to do rename/copy first, deliberately
ignoring pathspec (I was thinking about tweaking diff-tree to
break out of its built-in pathspec limit to feed diffcore
_everything_, even things outside of the specified directories
when --detect-copies-harder is used), to cast a wider net so
that it can come up with "the best matches".  But what you just
said made me realize that the definition of "the best matches" I
had in mind was not best at all.  I agree with you 100% that
pathspec should come first before everything else to limit the
world diffcore operates in, not just to limit the set of output
paths.

I still want to do what this patch does for another reason.  It
is so much simpler if I do not have to carry around _some_
unmatched pair after rename/copy.  The earlier representation
was forcing all the downstream diffcore filters to be aware of
what rename/copy did, which was simply _wrong_.  They should not
have to care.

About ordering of pickaxe and rename/copy, I think for most uses
of diffcore filters, not limited to pickaxe, would make more
sense if they come after rename/copy when rename/copy detection
is in effect.  What I really wanted to do (I made a side note
comment in the earlier discussion about this, saying "something
like streams", but have not pursued it further) is to make these
diffcore filters/transformations stackable so that the main
programs can control not just if each of them is used or not,
but the order of application of the used ones.  I cannot offhand
think of a good use case of having pickaxe come _before_
rename/copy, but there may be a case the user want to have
things in that order, and having the application order not
hardcoded in three diff-* brothers is a major problem if we
wanted to give that option.