Re: [RFC] Hits/directions to using autoconf and friends in Debian
On Fri, 12 Oct 2001, Steve Langasek wrote:
> > (GNU makefile snippet):
> > export DEB_HOST_GNU_TYPE ?= $(shell dpkg-architecture -qDEB_HOST_GNU_TYPE)
> > export DEB_BUILD_GNU_TYPE ?= $(shell dpkg-architecture -qDEB_BUILD_GNU_TYPE)
> If this is the recommended way to call ./configure, should it any longer
> be recommended to import updated config.guess scripts from autotools-dev?
Well, I do update both in my packages simply because it reduces the
probability of bad surprises in the long run. The compressed diff increase
is not that big, most of the time.
> One still needs config.sub, but playing with config.guess just gives you
> larger diffs. I think it would be good to state this in the readme.
I have reworked the README and added a paragraph to that effect.
> Also, to canonicalize the host and build values, the above should be
> ./configure --build=$(DEB_BUILD_GNU_TYPE)-gnu --host=$(DEB_HOST_GNU_TYPE)-gnu \
> ....... (other configure options) ......
No. For i386, for example, the correct GNU arch string is i386-pc-linux-gnu;
Tacking -gnu to the tail of "i386-linux" as returned by dpkg-architecture is
not needed, as config.sub will expand that to the proper i386-pc-linux-gnu
or i386-unknown-linux-gnu anyway...
> since the -gnu isn't part of the value that dpkg-architecture returns.
I am not sure if this isn't actually a bug in dpkg-architecture. If it is
supposed to return the GNU architecture string, it might as well return the
proper canonical one. Just in case it gets fixed in the future, I'll leave
to configure/config.sub to find out the proper canonical name for the arch
as returned by dpkg-architecture.
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot