On 3/28/23 4:24 PM, Junio C Hamano wrote:

Derrick Stolee [off-list ref] writes:

quoted

quoted

quoted

 			setup_work_tree();
+			prepare_repo_settings(the_repository);
+			the_repository->settings.command_requires_full_index = 0;

Offhand, the only case I know that "describe" even _needs_ a working
tree or the index is when asked to do the "--dirty" thing.  To
figure out if the working tree files are modified, the code calls
into run_diff_index(), but has that codepath been made sparse-index
aware already?

It seems that this is a case where we can rely on the existing
changes around run_diff_index(), which is nice. We get a very
easy win for a narrow case.

Hmph, so "diff-index" is already sparse-index aware?  I looked for
any old commit that flips .command_requires_full_index to false for
the command and didn't find it, and if I recall correctly (I only
checked when I wrote the comment you are responding to),
cmd_diff_index() does not do that, either, so I assumed that nobody
has converted run_diff_index() yet.  And that was where my question
came from.

Since 'git diff --cached' is covered as of 51ba65b5c35 (diff:
enable and test the sparse index, 2021-12-06), that means that
the implementation run_diff_index() is covered.

In fact, it's likely that the reason it's covered is because
run_diff_index() calls diff_cache() which calls unpack_trees() and
unpack_trees() needs to work for _anything_ to work, especially
'git status' and 'git checkout'.

The only thing keeping us from enabling the sparse index on many
of these "diff" builtins is the necessary testing in t1092. I'm
sure that diff-index is very easy to convert. Hopefully, we will
have a GSoC student this summer who can tackle most of these easy
ones, as well as a tricky one or two.

Thanks,
-Stolee