Re: [PATCH v5 5/6] p5310-pack-bitmaps.sh: enable `pack.writeReverseIndex`

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Jul 26, 2022, 13:54 UTC
Message-ID: <220726.8635eonfjp.gmgdl@evledraar.gmail.com>
In-Reply-To: <733f2432-00ab-f0c2-269f-90af02b2105c@github.com>

On Tue, Jul 26 2022, Derrick Stolee wrote:

Show 10 quoted lines

> On 7/26/2022 3:15 AM, Ævar Arnfjörð Bjarmason wrote:
>> Rather than comparing HEAD to your current tree it's generally better
>> to do something like:
>> 
>> 	GIT_PERF_MAKE_OPTS='-j3' ./run HEAD~ HEAD [...]
>
> Using the 'run' script fixes the perf test in the worktree and tests
> different versions of the 'git' executable.
>
> That doesn't work when the change is in the performance test itself.

Thanks, I'm clearly wrong about that. I didn't look enough at the context.

But then we're losing the perf test coverage for the case where we don't have the *.rev files. Isn't it better to run both with & without *.rev, perhaps by splitting up the test file? We could make it a function in perf/lib-bitmap.sh that we call both with & without the wanted *.rev repack config.

I suspect that's also subtly broken, in that t/perf assumes that it can re-use the repo for a given <rev>, but this is modifying that repo, so if you run e.g. test Y after this Y, that Y will unexpectedly get a repack'd repo ...

But we could just start the test with a git clone . "$TEST_NAME" or whatever, then repack that with whatever options we want...