On Mon, Nov 14 2022, Glen Choo wrote: > Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes: > >> It's also proposing to replace Glen's one-patch[6], which is working >> around the problem shown in the test added in 1/10 here. Per >> downthread of [7] I think Glen was aiming for getting a more narrow >> fix in case we split off 9/10 here into some later fix. >> >> As we're fixing an edge case in something that's always been broken >> (and thus wouldn't backport) I think it's better to just fix the >> problem directly, rather than introducing new "--super-prefix" use, >> just to take it away later. > > I still prefer that we take the one-patch to unbreak new releases, > because partial clone + submodules is absolutely broken (e.g. it's > already causing quite a lot of headaches at $DAYJOB) and the patch is > obviously harmless. > > And more importantly, it lets us take our time with this series and get > it right without time pressure. It's not as pressing as, e.g. a > regression fix, but it does render certain Git setups unusable. > > With regards to urgency and when to choose "small and harmless fixes vs > bigger and better fixes", I think Junio has generally made those calls > in the past. @Taylor if you have an opinion, I'd love to hear it. I feel like I'm missing something here. What's the regression? The test you're adding here didn't work at all until 0f5e8851737 (Merge branch 'rc/fetch-refetch', 2022-04-04), as the command didn't exist yet. That commit went out with v2.36.0. If it never worked there's no regression, and we wouldn't be merging down a fix for older point-releases. Or is there some case I've missed here which did work before, doesn't now, but just isn't captured in this test? If so what case is that, and when did it break?
