[0/2] rebase.useBuiltin doc & test mode

Message

Ævar Arnfjörð Bjarmason Nov. 14, 2018, 9:01 a.m. UTC

On Wed, Nov 14 2018, Stefan Beller wrote:

>> But maybe I'm being overly paranoid. What do those more familiar with
>> this think?
>
> I am not too worried,
> * as rebase is a main porcelain, that is even hard to use in a script.
>   so any failures are not deep down in some automation,
>   but when found exposed quickly (and hopefully reported).
> * 5541bd5b8f was merged to next a month ago; internally we
>    distribute the next branch to Googlers (on a weekly basis)
>    and we have not had any bug reports regarding rebase.
>    (Maybe our environment is too strict for the wide range
>     of bugs reported)

I do the same at Booking.com (although at a more ad-hoc schedule) and
got the report whose fix is now sitting in "pu" noted upthread.

I fear that these sorts of corporate environments, both Google's and
Booking's, end up testing a relatively narrow featureset. Most people
have similar enough workflows, e.g. just using "git pull --rebase",
I'd be surprised if we have more than 2-3 internal users who ever use
the --onto option for example.

> * Johannes reported that the rebase is used in GfW
>    https://public-inbox.org/git/[e-mail]/
>    https://github.com/git-for-windows/git/pull/1800
>    and from my cursory reading it is part of
>    2.19.windows, which has a large user base.
>
>> (and would re-enable rebase.useBuiltin=true in
>> master right after 2.20 is out the door).
>
> That would be fine with me as well, but I'd rather
> document rebase.useBuiltin instead of flip-flopping
> the switch around the release.
>
> Have there been any fixes that are only in
> the C version (has the shell version already bitrotted)?

That's a good question, one which I don't think we knew the answer to
before the following patches. As it turns out no, we still run the
tests without failures with GIT_TEST_REBASE_USE_BUILTIN=false.

Ævar Arnfjörð Bjarmason (2):
  rebase doc: document rebase.useBuiltin
  tests: add a special setup where rebase.useBuiltin is off

 Documentation/config/rebase.txt | 14 ++++++++++++++
 builtin/rebase.c                |  5 ++++-
 t/README                        |  3 +++
 3 files changed, 21 insertions(+), 1 deletion(-)

Comments

Johannes Schindelin Nov. 14, 2018, 2:07 p.m. UTC | #1

Hi Ævar,

On Wed, 14 Nov 2018, Ævar Arnfjörð Bjarmason wrote:

> On Wed, Nov 14 2018, Stefan Beller wrote:
> 
> >> But maybe I'm being overly paranoid. What do those more familiar with
> >> this think?
> >
> > I am not too worried,
> > * as rebase is a main porcelain, that is even hard to use in a script.
> >   so any failures are not deep down in some automation,
> >   but when found exposed quickly (and hopefully reported).
> > * 5541bd5b8f was merged to next a month ago; internally we
> >    distribute the next branch to Googlers (on a weekly basis)
> >    and we have not had any bug reports regarding rebase.
> >    (Maybe our environment is too strict for the wide range
> >     of bugs reported)
> 
> I do the same at Booking.com (although at a more ad-hoc schedule) and
> got the report whose fix is now sitting in "pu" noted upthread.
> 
> I fear that these sorts of corporate environments, both Google's and
> Booking's, end up testing a relatively narrow featureset. Most people
> have similar enough workflows, e.g. just using "git pull --rebase",
> I'd be surprised if we have more than 2-3 internal users who ever use
> the --onto option for example.
> 
> > * Johannes reported that the rebase is used in GfW
> >    https://public-inbox.org/git/[e-mail]/
> >    https://github.com/git-for-windows/git/pull/1800
> >    and from my cursory reading it is part of
> >    2.19.windows, which has a large user base.
> >
> >> (and would re-enable rebase.useBuiltin=true in
> >> master right after 2.20 is out the door).
> >
> > That would be fine with me as well, but I'd rather
> > document rebase.useBuiltin instead of flip-flopping
> > the switch around the release.
> >
> > Have there been any fixes that are only in
> > the C version (has the shell version already bitrotted)?
> 
> That's a good question, one which I don't think we knew the answer to
> before the following patches.

I pay close attention to `git rebase` bug reports and patches (obviously),
and there have not been any changes going into the built-in rebase/rebase
-i that necessitated changes also in the scripted version.

Ciao,
Dscho

> As it turns out no, we still run the tests without failures with
> GIT_TEST_REBASE_USE_BUILTIN=false.
> 
> Ævar Arnfjörð Bjarmason (2):
>   rebase doc: document rebase.useBuiltin
>   tests: add a special setup where rebase.useBuiltin is off
> 
>  Documentation/config/rebase.txt | 14 ++++++++++++++
>  builtin/rebase.c                |  5 ++++-
>  t/README                        |  3 +++
>  3 files changed, 21 insertions(+), 1 deletion(-)
> 
> -- 
> 2.19.1.1182.g4ecb1133ce
> 
>