(no subject)

[Han-Wen Nienhuys]

Hey folks,

I spent the last few weeks cobbling together an implementation of the
reftable format in C and in Go. I thought this would be cool to add to
git-core, but I doubt whether I will have enough time to see such an
effort through. Maybe some of you would want to try integrating it
into the Git-core code base?  Example code is here:

  https://github.com/google/reftable/blob/master/c/api.h#L153

cheers!
--
Han-Wen Nienhuys - Google Munich
I work 80%. Don't expect answers from me on Fridays.
--
Google Germany GmbH, Erika-Mann-Strasse 33, 80636 Munich
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg
Geschäftsführer: Paul Manicle, Halimah DeLaine Prado

Re: none

[Junio C Hamano]

Han-Wen Nienhuys <hanwen@google.com> writes:

> I spent the last few weeks cobbling together an implementation of the
> reftable format in C and in Go. I thought this would be cool to add to
> git-core, but I doubt whether I will have enough time to see such an
> effort through. Maybe some of you would want to try integrating it
> into the Git-core code base?  Example code is here:
>
>   https://github.com/google/reftable/blob/master/c/api.h#L153
>
> cheers!

My initial impression was that the API overuses typedef.  We tend to
avoid doing

	struct _foo { ... };
	typedef struct _foo foo;

and instead write "struct foo" explicitly to make us well aware of
what we are talking about.  That lets us see that we are passing or
returning a structure by value (which we would like authors to think
thrice before doing in C) like so quite easily:

	foo some_function(foo arg1, ...) { ... }

because it would be clear if it were written like so

	struct foo some_function(struct foo arg1, ...) { ... }

without hiding the structure behind a typedef (it also lets us avoid
names with leading underscore, which is frowned upon by some people
for different reasons).

But the set of operations defined in the header file seemed at the
right granularity in order to interface with the refs.h & refs/* API
we have.  It however was unclear to me how transactional ref updates
would work with it.

Thanks.

Re: none

[Han-Wen Nienhuys]

On Tue, Nov 19, 2019 at 8:53 PM Junio C Hamano <gitster@pobox.com> wrote:
> My initial impression was that the API overuses typedef.  We tend to
> avoid doing
>
>         struct _foo { ... };
>         typedef struct _foo foo;
>
> and instead write "struct foo" explicitly to make us well aware of
> what we are talking about.

Thanks, i'll have a look at changing it. I use typedef mainly for
ergonomics, but now that the code is written, I can introduce more
verbosity.

> But the set of operations defined in the header file seemed at the
> right granularity in order to interface with the refs.h & refs/* API
> we have.  It however was unclear to me how transactional ref updates
> would work with it.

Transactions have to interface with the file system. I imagine that
different libraries (libgit vs. cgit) would have different primitives
for dealing with the file system, hence I haven't implemented that
part. Do you have an idea of how I could implement it in a way that is
agnostic of libgit2 vs. cgit?

-- 
Han-Wen Nienhuys - Google Munich
I work 80%. Don't expect answers from me on Fridays.
--

Google Germany GmbH, Erika-Mann-Strasse 33, 80636 Munich

Registergericht und -nummer: Hamburg, HRB 86891

Sitz der Gesellschaft: Hamburg

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado

Re: your mail

[Taylor Blau]

Hi Han-Wen,


On Tue, Nov 19, 2019 at 07:49:17PM -0800, Han-Wen Nienhuys wrote:
> Hey folks,
>
> I spent the last few weeks cobbling together an implementation of the
> reftable format in C and in Go. I thought this would be cool to add to
> git-core, but I doubt whether I will have enough time to see such an
> effort through. Maybe some of you would want to try integrating it
> into the Git-core code base?  Example code is here:
>
>   https://github.com/google/reftable/blob/master/c/api.h#L153

Very exciting, and thanks for your work on this. I haven't taken a
close look at the code yet, so I'm sure taking this further involves a
much more careful examination.

I know that Christian Couder (who you CC-d on this thread) was also
working on this for either GitLab or Booking.com.

Christian -- are you still working on this for either one of those
entities? If so, is there some way to unify these two efforts?

> cheers!
> --

Thanks,
Taylor

Re: your mail

[Christian Couder]

Hi Taylor and Han-Wen,

On Wed, Nov 20, 2019 at 6:30 AM Taylor Blau <me@ttaylorr.com> wrote:
>
> On Tue, Nov 19, 2019 at 07:49:17PM -0800, Han-Wen Nienhuys wrote:
> >
> > I spent the last few weeks cobbling together an implementation of the
> > reftable format in C and in Go. I thought this would be cool to add to
> > git-core, but I doubt whether I will have enough time to see such an
> > effort through. Maybe some of you would want to try integrating it
> > into the Git-core code base?  Example code is here:
> >
> >   https://github.com/google/reftable/blob/master/c/api.h#L153

Interesting!

> Very exciting, and thanks for your work on this. I haven't taken a
> close look at the code yet, so I'm sure taking this further involves a
> much more careful examination.
>
> I know that Christian Couder (who you CC-d on this thread) was also
> working on this for either GitLab or Booking.com.
>
> Christian -- are you still working on this for either one of those
> entities? If so, is there some way to unify these two efforts?

Yeah, I started working on this last year for Booking.com, and I am
now slowly working on it for GitLab. It is not yet finished because it
has been low priority work.

I took a very quick look at Han-Wen's implementation and it looks very
different from mine from the outside. It seems more complete, but
might be less integrated into the Git code base.

Best,
Christian.

reftable (was: Re: none)

[Han-Wen Nienhuys]

On Tue, Nov 19, 2019 at 8:53 PM Junio C Hamano <gitster@pobox.com> wrote:
> > I spent the last few weeks cobbling together an implementation of the
> > reftable format in C and in Go. I thought this would be cool to add to
> > git-core, but I doubt whether I will have enough time to see such an
> > effort through. Maybe some of you would want to try integrating it
> > into the Git-core code base?  Example code is here:
> >
> >   https://github.com/google/reftable/blob/master/c/api.h#L153
> >
> > cheers!
>
> My initial impression was that the API overuses typedef.

I changed the code to use struct throughout.

> But the set of operations defined in the header file seemed at the
> right granularity in order to interface with the refs.h & refs/* API
> we have.  It however was unclear to me how transactional ref updates
> would work with it.

I implemented the reftable stack in Go, see

  https://godoc.org/github.com/google/reftable#Stack

which I could implement in C too. Would that work?

-- 
Han-Wen Nienhuys - Google Munich
I work 80%. Don't expect answers from me on Fridays.