Re: [PATCH] t1309: use a non-loaded branch name in the `onbranch` test cases

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

[-- Attachment #1: Type: text/plain, Size: 2386 bytes --]

Hi Ævar,

On Thu, 19 Nov 2020, Ævar Arnfjörð Bjarmason wrote:

> On Thu, Nov 19 2020, Johannes Schindelin wrote:
>
> > On Wed, 18 Nov 2020, Ævar Arnfjörð Bjarmason wrote:
> >
> >> On Wed, Nov 18 2020, Johannes Schindelin via GitGitGadget wrote:
> >>
> >> > From: Johannes Schindelin <johannes.schindelin@gmx.de>
> >> >
> >> > Therefore, the actual branch name does not matter at all. We might
> >> > just as well avoid racially-charged names here.
> >>
> >> It seems to me the actual name matters a lot, and it must whatever
> >> the default branch name is.
> >
> > Nope. Not at all. Because what we're exercising is the code paths when
> > we _don't_ have a branch name to work with.
> >
> > In the non-Git case, this is trivial to see. There is not even a
> > repository! How can there be a branch?
> >
> > In the early config case, it is too early to access the refs. I meant
> > to reference (but forgot) the commit 85fe0e800ca (config: work around
> > bug with includeif:onbranch and early config, 2019-07-31) because that
> > commit's commit message describes the catch-22 that is the reason why
> > the early config cannot see the current branch name (if any).
> >
> > I should probably have thought of referencing 22932d9169f (config:
> > stop checking whether the_repository is NULL, 2019-08-06) for the
> > second test case, too.
> >
> > So again, these two test cases do _not_ exercise the code path where
> > another config file is included. To the contrary, they try to prevent
> > a regression where `onbranch` would segfault in one case, and BUG in
> > the other (in both cases because the now-fixed code used to try to
> > look at the current branch name _anyway_).
> >
> >> I.e. what the test is doing is producing intentionally broken config,
> >> and asserting that we don't read it at an early stage.
> >>
> >> Therefore if we regressed and started doing that the test wouldn't
> >> catch it, because the default branch name is "master", or "main"
> >> if/when that refs.c change lands, neither of which is "topic".
> >
> > No, if we regressed, the code would start to throw a BUG, or a
> > segfault, respectively.
> >
> > We never expect these two test cases to look at any branch name at
> > all.
>
> Thanks. I mis(understood|read) it.

I guess the commit message could use an update.

Ciao,
Dscho