From: Mark Jamsek <mark@jamsek.com>
Subject: Re: special case 'got fetch -b <branch>' + regress
To: Game of Trees <gameoftrees@openbsd.org>
Date: Tue, 14 Feb 2023 00:09:07 +1100

On 23-02-13 12:49PM, Stefan Sperling wrote:
> On Mon, Feb 13, 2023 at 02:23:01AM +1100, Mark Jamsek wrote:
> > At present, we cover 'got fetch' from:
> > 
> >   - repo: got.conf branch that exists on remote (fetches got.conf only)
> >   - repo: -b branch that exists on remote (fetches -b only)
> >   - repo: -b branch that doesn't exist on remote (fetches none)
> >   - repo: got.conf branch that doesn't exist (fallback to remote HEAD)
> >   - wt: got.conf that doesn't exist, wt that does exist (fetches wt only)
> >   - wt: got.conf and wt branches that both exist (fetches conf and wt)
> >   - wt: -b branch that does exist on the remote (fetches -b only)
> >   - wt: -b branch that doesn't exist on the remote (fetches none)
> 
> Thanks! It is nice to see more test coverage added in particular.
> 
> I think you should commit this and we'll keep iterating on top.
> There were ideas about "always fetching HEAD" (specifics pending), which
> this patch does not seem to implement. But we should keep building on top
> of this diff in any case.
> Without good test coverage it would be harder to decide what we really want.

Thanks!

Yes, that was my intent. I'm one of the always-fetch-HEAD proponents,
but I think we need to make some comparisons. So if we can run for a few
days with HEAD fetched only as a fallback, and then try always fetching
HEAD irrespective of whether got.conf or work tree branches are
successfully fetched, we should have a good idea of which mode we
prefer.

I suspect that the remote's HEAD branch will invariably be set in
got.conf (or git's config) when the repository is cloned; as such, it
will more often than not be fetched in the default configuration.
However, if we move to the always-fetch-HEAD mode, users won't
inadvertently miss it when/if HEAD is changed. So ultimately I think
that will be the better option. OTOH, as we briefly discussed, it might
seem like a misfeature to fetch a branch that the user hasn't requested.

> > As a related change, I'll follow up (tomorrow) with a diff that improves
> > the output when requesting a branch with 'got fetch -b' that does not
> > exist. Currently, we report:
> > 
> >   got-fetch-pack: could not find any branches to fetch
> >   got: could not find any branches to fetch
> 
> Ah, yes, agreed. It would be good to reinstate this change.

Is the man page change ok to commit, too, or would you rather we leave
that till we're settled on the fetch behaviour?

-- 
Mark Jamsek <fnc.bsdbox.org|got.bsdbox.org>
GPG: F2FF 13DE 6A06 C471 CA80  E6E2 2930 DC66 86EE CF68