Re: individual software releases for third parties

[Alistair Crooks <agc%pkgsrc.org@localhost>]

On Thu, Aug 30, 2012 at 11:52:35AM -0400, James K. Lowden wrote:
> On Thu, 30 Aug 2012 14:20:41 +0100
> Patrick Welche <prlw1%cam.ac.uk@localhost> wrote:
> 
> > On Sat, Aug 25, 2012 at 06:53:21PM +0300, Aleksey Cheusov wrote:
> > > One more example of how NetBSD tools may become portable without
> > > autotools.
> > 
> > I don't see the point of avoiding autotools - {cross,} compiling on
> > a variety of architectures is precisely what it is designed to do.
> 
> Eh? Autoconf determines compile-time choices by probing the build
> host.  How is that a design for cross-compilation? 
> 
> The right way to approach portability has never been tried: to
> construct a database of alternatives, keyed approximately by os, libc,
> and compiler version. That list is orders of magnitude smaller than the
> number of build hosts, and that approach has the property of aiding our
> understanding instead of obscuring the problem.  

I think this is a good, albeit simplistic way of looking at things.
It does imply a static view of system software and characteristics.

It doesn't, however, address additional software which may or may not
be on a machine, and certainly doesn't address situations where
ncurses may be installed on a build host, but I wish to
(cross-)compile and (cross-)link with a different curses library.

It also assumes that installed software does not change over time. 
And that functions are available with the same API/ABI across the same
major version of libs (i.e.  it doesn't take openssl into account).
And that system characteristics are only dependent on system software.

In fact, the whole discussion is reminiscent of the calls to have a
central autoconf cache when building to speed things up (which fails
for the same reasons given above).

Regards,
Alistair