Re: broken packages for 2012Q3

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

On Sun, Sep 30, 2012 at 01:11:34PM +0200, Manuel Bouyer wrote:
> On Sun, Sep 30, 2012 at 09:18:05AM +0200, Joerg Sonnenberger wrote:
> > On Sat, Sep 29, 2012 at 11:23:46PM +0000, David Holland wrote:
> > > On Sat, Sep 29, 2012 at 10:30:32PM +0200, Manuel Bouyer wrote:
> > >  > I see that the TNF package builders have PKG_DEVELOPER=yes. They should
> > >  > probably not. AFAIK the goal of these systems is to produce binary 
> > > packages,
> > >  > not to debug pkgsrc. Also this may partially explain why they're so 
> > > slow
> > >  > building packages.
> > > 
> > > It doesn't do any good to produce broken binary packages, though.
> > > Maybe some of the PKG_DEVELOPER checks should be moved out of
> > > PKG_DEVELOPER. Particularly the PLIST check, at least when using
> > > destdir builds.
> > 
> > The PLIST check is relatively cheap with DESTDIR builds. It only touches
> > inodes, that would be needed in the next step (packaging) anyway. If
> > that creates a significant amount of IO, kern.maxvnodes should be
> > raised.
> 
> that's not what's causing most of the I/O; I suspect there are
> checks that look at the content of files, and this is what can be
> expensive.

Even those (e.g. the interpreter check, the rpath check) should not add
IO unless the machine is memory starved (which would waste time in a lot
of other places anyway), since the content is going to be read soonish
anyway.

Joerg