Re: What's cooking in git.git (Nov 2008, #06; Wed, 26)

On Thu, 11 Dec 2008, Junio C Hamano wrote:

Daniel Barkalow [off-list ref] writes:

quoted

quoted

That's the point. CE_VALID does not define checkout area while
CE_NO_CHECKOUT does.  If an entry is CE_VALID, it is still in checkout
area. But if it is CE_NO_CHECKOUT, "git grep" should ignore that path.
core.defaultsparse has nothing to do here.

My point is that the index cannot tell git grep whether it should search a 
path if the path isn't in the index.

Let's step back a bit.  I think "git grep" that stays silent outside of
the checkout area when used to grep in the work tree or in the index is a
mistake.

The problem "sparse checkout" attempts to address is not this:

    I ran "git init && git add ." in /usr/src by mistake.  There is no
    reason for coreutils that is in /usr/src/coreutils and gnucash that is
    in /usr/src/gnucash to share the same development history nor their
    should be any ordering between commits in these two independent
    projects.  I should have done N separate "init & add" independently at
    one level deeper in the directory hierarchy, but I am too lazy to
    filter branch the resulting mess now.

At least, it should not be that, at least to me.

"Sparse" is "I am not going to modify the files in these areas, and I know
they do not need to be present for my purposes (e.g. build), so I do not
need copies in the work tree."  It still works on the whole tree structure
recorded in the commit, but gives you a way to work inside a sparsely
populated work tree, iow, without checking everything out.

There's the meta question of: "Do people who have declared that they 
aren't going to modify or build with some files want their searches to 
tell them about those files?"

Say I'm the "tr" guy, and I care about the build system, library code, and 
"tr.c", and I run "make tr"; my sparse checkout doesn't include "head.c", 
and I totally ignore all the other stuff that's in coreutils. Maybe I want 
"git grep" to exclude the other stuff.

I don't really have a firm position on whether "git grep" should ignore 
"head.c" or not, but I think it should be consistent between "git grep" 
and "git grep origin/next", and I think that, if origin/next has a new 
"foot.c" that isn't in the current branch to by marked as NO_CHECKOUT, it 
should be skipped if "tail.c" (which is in my current branch) is skipped.

So "git grep -e frotz Documentation/", whether you only check out
Documentation or the whole tree, should grep only in Documentation area,
and "git grep -e frotz" should grep in the whole tree, even if you happen
to have a sparse checkout.  By definition, a sparse checkout has no
modifications outside the checkout area, so whenever grep wants to look
for strings outside the checkout area it should pretend as if the same
content as what the index records is in the work tree.  This is consistent
with the way how "git diff" in a sparsely checked out work tree should
behave.

"git diff" is an ambiguous model for "git grep". It equally describes 
the behavior of "git diff" to say that it treats files outside the 
checkout area as matching the index or to say that it never lists files 
outside the checkout area. On the other hand, there is the question of 
whether "git diff branch1 branch2" shows differences that are outside the 
checkout area, and whether "git log" shows commits that only change things 
outside the checkout area, and "git grep" should match the behavior of 
these.

	-Daniel
*This .sig left intentionally blank*