ports/143529: [PATCH] math/py-numpy: does not build

[Li-Lun "Leland" Wang]

The following reply was made to PR ports/143529; it has been noted by GNATS.

From: "Li-Lun \"Leland\" Wang" <llwang at infor.org>
To: bf1783 at gmail.com
Cc: bug-followup at freebsd.org, amdmi3 at amdmi3.ru, gerald at freebsd.org
Subject: Re: ports/143529: [PATCH] math/py-numpy: does not build
Date: Mon, 15 Feb 2010 15:23:49 +0000

 On Mon, Feb 15, 2010 at 3:11 PM, b. f. <bf1783 at googlemail.com> wrote:
 > On 2/15/10, b. f. <bf1783 at googlemail.com> wrote:
 >> On 2/15/10, b. f. <bf1783 at googlemail.com> wrote:
 >>>>From: "Li-Lun Wang (Leland Wang)" <llwang at infor.org>
 >>>>To: bug-followup at FreeBSD.org
 >>>>Cc:
 >>>>Subject: Re: ports/143529: [PATCH] math/py-numpy: does not build
 >>>>Date: Mon, 15 Feb 2010 12:50:24 +0800
 >>>>
 >>>> -----BEGIN PGP SIGNED MESSAGE-----
 >>>> Hash: SHA1
 >>>
 >>>>The following patch will fix the problem. =C2=A0It first detects if
 >>>> binutils>=3D2.18.50 is installed. =C2=A0If it is, then it patches nump=
 y so
 >>>> that the new fenv.h will be used if __FreeBSD_version <=3D 900009.
 >>>
 >>> This fixes the build problem on i386 (although we're stuck with
 >>> possible run-time problems when numpy is built with the newer binutils
 >>> until the latest fenv.h changes are backported to all supported
 >>> versions of FreeBSD), but this code is machine-dependent, and fenv.h
 >>> needs to be selected based on ARCH.
 >
 > Oops: it seems that the r203441 only affected amd64 and i386, so the
 > latest headers shouldn't be used for the other architectures (there
 > may be problems lurking there, too, but that's another matter). =C2=A0So =
 my
 > sketch should be altered to something like:
 >
 >
 > ...
 > .if exists(${LOCALBASE}/bin/as) && ${OSVERSION} <=3D 900009 && \
 > (${ARCH} =3D=3D "i386" || ${ARCH} =3D=3D "amd64")
 > .if ${ARCH} =3D=3D "i386"
 > =C2=A0FP_ARCH=3D i387
 > .elif ${ARCH} =3D=3D "amd64"
 > =C2=A0FP_ARCH=3D ${ARCH}
 > .endif
 > MASTER_SITES+=3D http://svn.freebsd.org/base/head/lib/msun/%SUBDIR%/:fp
 > MASTER_SITE_SUBDIR+=3D ${FP_ARCH}/:fp
 > =C2=A0DISTFILES+=3D fenv.h:fp
 > .endif
 > .endif
 > =C2=A0...
 > =C2=A0pre-configure:
 > .if exists(${LOCALBASE}/bin/as) && ${OSVERSION} <=3D 900009
 > =C2=A0 =C2=A0 =C2=A0${CP} ${DISTDIR}/fenv.h
 > =C2=A0${WRKSRC}/numpy/core/include/numpy/fenv/fenv.h; \
 > =C2=A0 =C2=A0 =C2=A0${PATCH} ${PATCH_DIST_ARGS} < ${FILESDIR}/fenv.patch;=
  \
 > .endif
 > ...
 >
 > But again, this probably isn't necessary if lang/gcc44 is only using
 > the base system binutils, and indeed may fail if lang/gcc44 was built
 > when devel/binutils was not installed -- so we should see first what
 > action will be taken regarding lang/gcc44, or use a more refined test.
 
 I agree.  This really isn't a bug in the numpy port, but the
 unfortunate result of devel/binutils and bug in fenv.h.  Fixing in the
 numpy side is more of just a hack to make it build.
 
 -- llwang