Re: Dropping tsort for archive creation

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Fri, Jul 03, 2015 at 01:25:40PM +0000, Christos Zoulas wrote:
> In article <20150702220349.GA8752%britannica.bec.de@localhost>,
> Joerg Sonnenberger  <joerg%britannica.bec.de@localhost> wrote:
> >-=-=-=-=-=-
> >
> >Hi all,
> >in the old days, sorting archive members topologically helped to improve
> >build time. I don't think it provides a significant advantage anymore as
> >most systems will keep the archives in memory during linking and lookup
> >is generally helped by the index. Newer linkers like gold, lld or
> >mclinker or designed to build an in-memory index as well to allow
> >single pass processing of archives. As such, I want to propose getting
> >rid of the complexity. Trying it on my laptop saves 30s (of about 1h),
> >so change is pretty much in the noise. Comments?
> 
> I don't see the win here. All the complexity is hidden, and if it aint
> broken... I don't even remember when it was the last time I had to worry
> about lorder or tsort. What's the next step, to remove those tools? What's
> the grand plan here?

It removes complexity from the build system. It was triggered by a
discussion in #pkgsrc since many systems lack lorder and sort; the use
is one of the divergence points in the pkgsrc rule sets. If there is no
real gain from the sorting, why do it in first place?

Joerg