Re: [PATCH 2/8] tests: always test fetch of unreachable with v0

[Junio C Hamano <gitster@pobox.com>]

Jeff King <peff@peff.net> writes:

> On Thu, Feb 14, 2019 at 11:58:25AM -0800, Jonathan Tan wrote:
>
>> > On Tue, Feb 05, 2019 at 04:21:16PM -0800, Jonathan Tan wrote:
>> > 
>> > > Some tests check that fetching an unreachable object fails, but protocol
>> > > v2 allows such fetches. Unset GIT_TEST_PROTOCOL_VERSION so that these
>> > > tests are always run using protocol v0.
>> > 
>> > I think this is reasonable, but just musing on a few things:
>> > 
>> >   1. Are we sure going forward that we want to retain this behavior? I
>> >      feel like we discussed this on the list recently with no real
>> >      conclusion, but I'm having trouble digging it up in the archive.
>> 
>> One such discussion is here:
>> https://public-inbox.org/git/20181214101232.GC13465@sigill.intra.peff.net/
>
> Thanks, that was what I was thinking of, though there doesn't seem to
> have been much discussion in response. :)
>
>> >   2. If it does change, is there any way we could automatically find
>> >      spots like this that would then need to be undone? I cannot think
>> >      of a good solution, and I don't think it's a show-stopper not to
>> >      have one, but I thought I'd put it forward as a concept.
>> 
>> We can do so now either by "blaming" one and finding the originating
>> commit, or by searching for "support fetching unadvertised objects" (I
>> used the same comment everywhere in the commit [1] so that people can do
>> this), but I don't know how to enforce this for future work. (We can add
>> a special variable, but I think it's unnecessary and we can't enforce
>> that people use that new special variable instead of
>> GIT_TEST_PROTOCOL_VERSION anyway.)
>
> Yeah, I think we can find them once we know they need fixing. I was more
> wondering whether we would remember to do so. I.e., is there a way the
> test suite could remind us when our assumptions change. I can't think of
> an easy way to do so, though.

Perhaps looking it a different way may help.  Instead of saying "v2
will not protect unreachable objects, so this test must be run with
v0", think of it like "Git ought to protect unreachable objects, so
test with different versions of protocols to make sure all satisfy
the requirement---for now, v2 is known to be broken, so write it
with test_expect_failure".  IOW, have one test for each version,
some of them may document a known breakage.