Re: sourceware and git

[Christopher Faylor <cgf-use-the-mailinglist-please@sourceware.org>]

On Wed, Jul 30, 2014 at 08:57:20PM +0000, Joseph S. Myers wrote:
>On Wed, 30 Jul 2014, Christopher Faylor wrote:
>
>> On Wed, Jul 30, 2014 at 05:02:47PM +0000, Joseph S. Myers wrote:
>> >I think having them in one git project (that includes libgloss and the 
>> >relevant toplevel) would be fine.  I'm generally dubious of things such as 
>> >submodules;
>> 
>> Your dubiousity doesn't have any bearing if it isn't backed up by facts.
>
>It's backed up by the list of requirements I gave for git to work smoothly 
>in the presence of submodules as if there were a single repository, none 
>of which (according to the git-submodule manpage) git submodules meet.

You mentioned that you were dubious of "things such as submodules" which
sounds a lot like an unreasoning prejudice rather than an informed
opinion.  And, you specified a list of what you think is needed but
didn't back it up with "...and here's why submodules won't work".

From what I've read on the internet it is possible to make a lot (if not
all) of the submodule stuff happen automatically.  But, even if it
wasn't we're talking about Cygwin here which I'm willing to bet is not
something that you even have occasion to build.

>We have many years' experience of the src repository being problematic
>because of the use of a special feature of CVS (modules files for
>partial checkouts of a directory in the repository) that doesn't work
>well with other features of CVS (cvs update -d; the interaction of
>modules file changes with tagging and branching).  That also provides a
>basis for caution about such features in other version control systems,
>and for preferring an ordinary git repository whose clones work with
>ordinary git commands just like those commands work with clones of any
>other random git repository.

Hopefully you aren't trying to one-up me on "devo" experience.

You're overplaying the problems with CVS modules but since git is
obviously intended to be several orders of magnitude improvement over
CVS (which is the point of converting) there is no reason to
automaticaly avoid something called a "module" in git because you didn't
like how CVS implemented them.

CVS also had pecularities with branching and tagging but we're still
talking about branching and tagging in git.

As far as tagging is concerned, I don't see any reason why newlib
should be tagged when cygwin is tagged.  Cygwin shouldn't care about
newlib tags.