On Wed, Mar 30, 2022 at 6:39 PM Junio C Hamano [off-list ref] wrote:

Tao Klerks [off-list ref] writes:

quoted

Sorry, I'm not completely sure whether my comment was misleading, or
whether I'm misunderstanding your feedback.

The test added here does not test "desirable" behavior from an
end-user functional perspective, but it does test behavior that is
working "as-designed" as of many years ago.

Is it "as-designed", or just "left buggy"?

I somehow had an impression that you meant the latter, and it would
be our aspirational goal to eventually fix it.  And for such case,
it would be better to write the test body to show what the command
should do, which would make the test fail with today's Git, and mark
it as test_expect_failure (which is not an ideal mechanism to prepare
for a future fix, but that is what we have now and should use, until
a better alternative being worked on is finished).

But if it is "as-designed, some users may find it suboptimal or
confusing or with any other negative adjectives, but it is too late
to change now and more importantly it is unthinkable to change it
because existing tools and users do depend on the current behaviour",
then what you did is perfectly fine.

Heh, the answer is in-between: there certainly is an aspiration that
someone figure out how to have the single untracked cache structure
safely support both -uall and -unormal modes with the same data... but
it's not clear that this is possible, and it's certainly not planned.

I don't feel comfortable saying "we should fix this", and that's not
the intent of this test - its intent is to say "this is how it is
(currently) designed to work. If you accidentally change it, be sure
that you are changing the design, not just accidentally changing
behavior and potentially/probably introducing a bug".

I'll try with this updated comment, using the term "current design" to
make the distinction:

# Bypassing the untracked cache here is not desirable from an
# end-user perspective, but is expected in the current design.
# The untracked cache data stored for a -unormal run cannot be
# correctly used in a -uall run - it would yield incorrect output.