Re: pkg/40663: math/py-numpy seems to miss the f2c requirement (at least when building under Linux)

[David Holland <dholland-pbugs%netbsd.org@localhost>]

The following reply was made to PR pkg/40663; it has been noted by GNATS.

From: David Holland <dholland-pbugs%netbsd.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: OBATA Akio <obache%netbsd.org@localhost>
Subject: Re: pkg/40663: math/py-numpy seems to miss the f2c requirement (at
 least when building under Linux)
Date: Sun, 12 Jun 2011 05:20:44 +0000

 On Tue, Feb 17, 2009 at 12:10:05PM +0000, OBATA Akio wrote:
  >  > ===> Creating toolchain wrappers for py25-numpy-1.1.0
  >  > ERROR: f2c>=20001205nb3 is not installed; can't buildlink files.
  >  > *** Error code 1
  >  
  >  It's known problem, same as PR#37536.
 [...]
 
  >  Background:
  >  math/py-numpy/Makefile added "fortran" to "USE_LANGUAGES".
  >  then include
  >    lang/python/extension.mk
  >  => lang/python25/buildlink3.mk
  >  ==> mk/bdb.buildlink3.mk
  >  ===> databases/db4/buildlink3.mk
  >  ====> mk/compiler.mk
  >  =====> comipler/f2c.mk
  >  ======> lang/f2c/buildlink3.mk
  >  
  >  Then f2c/buildlink3.mk recognizes that it is included by the other
  >  buildlink3 file, not depends directly, no need to add itself to
  >  DEPENDS.
 
 So, at this point there's only one buildlink3.mk file that includes
 compiler.mk; it's databases/db46/buildlink3.mk, which includes it so
 it can test PKGSRC_COMPILER.
 
 Is there any reason the inclusion of compiler.mk can't be moved
 outside of the BUILDLINK_TREE block? That would solve the problem
 properly.
 
 ...although I don't see why this failure mechanism would be dependent
 on OS type or version. I guess it's probably because of db being
 builtin.
 
 -- 
 David A. Holland
 dholland%netbsd.org@localhost