[Kde-scm-interest] [Proposal] Package splitting with thin meta-repos

[Oswald Buddenhagen]

On Sat, Jan 30, 2010 at 04:09:14PM +0100, Thiago Macieira wrote:
> Em Sábado 30. Janeiro 2010, às 12.19.20, Oswald Buddenhagen escreveu:
> > such a limitation will bite us sooner or later. well, in fact, probably
> > it would right from the start - some modules are plugins for applications
> > which live "somewhere closer to the core".
> 
> You don't link to applications. You link to libraries.
> 
> This includes plugins: plugins link to libraries.
> 
uhm ... so?
- the application needs to provide headers which define the interface
- the application may provide an internal library to link against

> > - the side effects of fixing screwups (i.e., forced pushes) are way more
> >   limited. and don't tell me "no forced pushes". there is no way i will
> >   accept it as a irreversible fact if some idiot pushes a 250mb core file
> >   into some repository (sounds ridiculous? well, it happened at
> >   trolltech).
> 
> Well, it wasn't a 250 MB core file at Trolltech. It was a 50 MB GIF at Qt 
> Development Frameworks, Nokia.
> 
that was just the "minor" case. the core thing really happened as well,
multiple times even (though i admit the actual size is extrapolated
from the typical size of a creator core dump). like for many other of our
projects, we cut the history before integration into mainline (a pretty
bizzare way to handle things, but anyway), so no permanent harm was done
once more ... but we (both qt and kde) really ought to have pre-commit
and pre-receive hooks to catch such crap.

> Whole world of difference :-)
> 
:D

> In any case, people will notice trying to upload that much data when
> you're not on the same LAN as the server.
> 
i wouldn't bet on that. people commonly have 10 or even 100 mbit in
dorms, and in some cities that's even common-place home connections.
also, when someone starts a push and goes afk immediately, even a
considerably slower connection will be sufficent to allow for such a
screwup.
also, after the experiences at qtdf, i'll readily deny the existence of
"common sense" when it comes to most(?) people using git ...