Official git export

perryh at pluto.rain.com perryh at pluto.rain.com
Thu Sep 1 08:11:32 UTC 2011


Andriy Gapon <avg at freebsd.org> wrote:

> on 01/09/2011 10:03 perryh at pluto.rain.com said the following:
> > "Matthew D. Fuller" <fullermd at over-yonder.net> wrote:
> >> On Wed, Aug 31, 2011 at 07:30:34AM -0700 I heard the voice of
> >> perryh at pluto.rain.com, and lo! it spake thus:
> >>> Surely it would be "noticeably faster" to _download_ only (say)
> >>> /usr/src/sys than all of /usr/src, unless one has an uncommonly
> >>> fast link?  (It would also impose less load on the serving site.)
> >>
> >> In the context of most current-gen DVCSen, it's unlikely to be
> >> much (or in fact _any_) faster or less data to transfer.  It's
> >> just less data to blat into the working tree.
> > 
> > That makes a certain amount of sense _if_ the VCS considers the
> > entire base system to reside in a single repository, which is why
> > someone was suggesting splitting it into multiple repositories.
> >
> > The question remains:  does it really make sense that I must
> > download the entire VCS history for things like cddl, contrib,
> > crypto, games, and kerberos if I only plan to work on the kernel?
>
> As surprising as it may sound to you, in my opinion, the answer is
> closer to yes than to no.
>
[snip mention of building kernel-toolchain and userland for testing]
>
> Not everything is needed from userland bits, of course, and
> history may be not as useful as the source code itself, but once
> you need some bits from userland it's hard to separate them.

I don't doubt that one needs to have (whatever version of) the whole
source tree, but that was presumably installed from the distribution
ISO and kept up to date via csup.  I see little point in downloading
it all _again_ just to get VCS history -- and ability to check out,
branch, etc -- for areas that one doesn't plan to touch.  That's why
I referred specifically to downloading "the entire VCS history" of
such parts.

And yes, depending on what part(s) of the kernel I'm working on,
I may indeed need to touch some userland code.  That still doesn't
explain why I should have to import VCS awareness of code that I'm
_not_ modifying.


More information about the freebsd-arch mailing list