On Thu, Mar 17, 2016 at 01:07:19PM +0100, Marc Espie wrote:
> I should have been more clear in my comment.
> 
> Seeing ports failing to build is not really enough.
> 
> There is still a fairly nasty limitation in our make wrt matching paths
> between targets that don't have the same filename, but are the same in
> the filesystem.
> 
> This limitation prevents some ports to build in parallel, where make is
> the culprit that must be fixed (autoconf, for instance).
> 
> There are also various race conditions in some ports proper, which actually
> prevent parallel building.
> 
> Those are very different issues.  Just asserting a port doesn't build in
> parallel is not enough to take proper action.
> 
> If you can figure out what's going on, and make sure the port itself is
> broken (say, by checking that gmake exhibits the same broken behavior),
> then fixing the port/talking with upstream/marking as NO_PARALLEL is okay.
> 
> It involves rather specific skills.
> 
> As for that limitation, strategically, we would want to get rid of it
> eventually (it "just" involves me totally rewriting a large part of the
> core of make) rather than give in and use gpl code for building.
> 
> But marking code as NO_PARALLEL in an indiscriminate way does not help me
> doing that.
> 
> Also, in the large scheme of things, there are just a few ports that 
> realistically benefit from parallel building.  And those are often already
> tagged with DPB_PROPERTIES because they help the actual official 
> package building.
> 

I understand.

I might look at these ports at a later date, as well as a couple of
other ones I discovered today (devel/autoconf/2.61, as you mentioned,
and lang/python/3.4), to see if they show the same behaviour with gmake
outside of ports.  In the meanwhile I'll rely on DPB_PROPERTIES instead,
maybe in the way that Stuart suggested.

Thanks,
Andreas

-- 
Andreas Kusalananda Kähäri, Bioinformatics Developer, Uppsala, Sweden
OpenPGP: url=https://db.tt/2zaB1E7y; id=46082BDF
------------------------------------------------------------------------

Attachment: signature.asc
Description: PGP signature

Reply via email to