Re: Speeding up “guix pull”: splitting modules

["Gábor Boskovits" <boskovits@gmail.com>]

Hello zimoun,

zimoun <zimon.toutoune@gmail.com> ezt írta (időpont: 2020. jan. 10., P, 13:10):
>
> Hi Gábor,
>
> Thank you for the explanations.
>
> Below, I am thinking loudly. :-)
>
> On Tue, 7 Jan 2020 at 21:14, Gábor Boskovits <boskovits@gmail.com> wrote:
>
> > > > Gábor once suggested an iterative approach of identifying the most
> > > > important nodes in the package graph that should be moved to their own
> > > > modules, so that we would end up with a package graph that looks less
> > > > like a hair ball.  I think it would useful to get a better view on the
> > > > relationships between modules.
>
> [...]
>
> > I was suggesting the following idea:
> > The underlying package graph is actually a DAG, so by splitting modules it is
> > possible to achieve a state where the module graph also becomes a DAG.
> > The exact heuristic or algorithm of splitting was not defined.
>
> The modules graph (DAG) is already available. :-)

The main problem here is that the modules do not form a DAG.
There are circular dependencies between the modules.
If those were not, then modular build would be possible, but because of
the spaghetti we are forced to build these together.

> Modulo some glue code / tweaks, the tools are already in
> guix/graph.scm and guix/scripts/graph.scm, if I understand correctly.
> And note that the commit  ddd59159004ca73c9449a27945116ff5069c3743
> introduces topological sort -- for another topic: recursive import --
> but could be reused /adapted to detect cycles, IMHO.
>
> Would a weighted DAG help to detect which modules are in "bad shape"?
> Other said, mix somehow the packages DAG and the modules DAG?
>
> For example, the edge between the module m1 and the module m2 should
> be weighted by the ratio between the number of packages defined in the
> module m2 required in the module m1.
> So then, sorting this edges by weight value, it would give a better
> view of the relationship between modules.
>
> What do you think?
>
> We need to detect which "big" modules are imported for "few" packages, right?
>
>
> Aside, a naive question: does '#:select' improve the situation?

I don't know, but it is an interesting question.

>
>
>
> All the best,
> simon


Best regards,
g_bor
-- 
OpenPGP Key Fingerprint: 7988:3B9F:7D6A:4DBF:3719:0367:2506:A96C:CF63:0B21