ports/143529: [PATCH] math/py-numpy: does not build
Gerald Pfeifer
gerald at pfeifer.com
Mon Feb 15 20:50:02 UTC 2010
The following reply was made to PR ports/143529; it has been noted by GNATS.
From: Gerald Pfeifer <gerald at pfeifer.com>
To: bf1783 at gmail.com
Cc: "Li-Lun Wang (Leland Wang)" <llwang at infor.org>, bug-followup at freebsd.org,
amdmi3 at amdmi3.ru
Subject: Re: ports/143529: [PATCH] math/py-numpy: does not build
Date: Mon, 15 Feb 2010 21:41:17 +0100 (CET)
On Mon, 15 Feb 2010, b. f. wrote:
> This problem is occurring with lang/gcc44 from Ports, when
> devel/binutils is installed, because then the latest gcc44 snapshots
> are automatically preferring devel/binutils to the base system
> binutils.
No-no. lang/gcc44 does not "automatically" prefer things. I just
verified that.
What you describe only happens when an admin _willfully_ selects
the tools from devel/binutils over the system ones. I assume in
this case this happened by setting PATH accordingly, though there
are other ways. And of course, when someone arbitrarily replaces
system tools by others, or even just newer versions, all sorts of
unforeseen things can happen.
The proposed patch to check for the version of devel/binutils is
not correct, since it is not granted that lang/gcc44 would use that
version of binutils. For example, as b.f. notices, we could force
it to use the system tools even when devel/binutils is installed.
You may want to use '$CC -print-prog-name=as' to identify the
version of as that GCC is using and go from there.
That said, given the increasing problems that are arising due to
the rotten binutils that are in the base system, I had planned to
also have lang/gcc44 use devel/binutils for a while, after seeing
how this pans out with lang/gcc45. I am testing a patch to that
end right now and plan to commit it within the next 24 hours, when
I will also update to the latest snapshot of GCC 4.4.
Still, as far as math/py-numpy goes, please do check for the
version of binutils that is used by GCC instead of hardcoding
things. For example, if we later make this an OPTION, things
could break again otherwise.
Gerald
More information about the freebsd-python
mailing list