Re: Scaling Mercurial at Facebook

["Perry E. Metzger" <perry%piermont.com@localhost>]

On Sun, 19 Jan 2014 10:43:45 +0100 Martin Husemann
<martin%duskware.de@localhost> wrote:
> On Sat, Jan 18, 2014 at 09:19:01PM -0500, Perry E. Metzger wrote:
> > I substantially prefer Mercurial to git. However, I think at this
> > point one should be reasonably realistic and accept that git has
> > won.
> 
> I completely agree with the first sentence, however, I fail to see
> how "winning" anything is important to our choice.

The tooling available for git is, at this point, overwhelming, and
more and more people are familiar with git and expect it to be used
if they're going to be comfortable contributing to something and
patching.

I think this is much like the decision to use X11, which at the time
was not the only or even the best window system. Frankly, X11 has
always been more than a little crappy, but the ecosystem was just too
compelling and it became a monopoly of sorts. Now many years later
people are gearing up to move to Wayland of course, but at the time
NetBSD picked it the choice to standardize on something else would
have been unreasonable.

The decision to pick git is also not one that need be irreversible --
the open source community is getting good at building tools for
converting between different systems (such as Reposurgeon). I think
any such decision should be made with the full belief that if a
better alternative later rises in three or five or ten years, that git
can and should be abandoned without looking back. It is a tool, not
part of the operating system itself, and there is no need to be
sentimental about it.

> To repeat what has been said before: for any serious candidate, we
> need a short paper describing the future workflow and a plan for
> the server side setup (which includes mail-index and gnats
> integration).

I think that might lead to some self-deception. It is not possible to
plan such a thing perfectly. I would suggest that, instead,
experiments be done with live mirrors of some portion of the source
tree. Experience will inevitably produce information that no amount of
planning will generate.

If an initial experiment works, then it will be possible to put
together a broader plan, and then probably some small subset of the
tree (like localsrc or some such) should be converted first, and if
that is successful, then something larger (like pkgsrc).

Perry
-- 
Perry E. Metzger                perry%piermont.com@localhost