Subject: Re: CVS commit: pkgsrc/mk
To: Roland Illig <>
From: Dieter Baron <>
List: tech-pkg
Date: 05/12/2006 13:38:47
In article <> Roland wrote:
: Johnny C. Lam wrote:
: >>I cannot see any hint here. What about printing the warning (which is 
: >>currently in config.log) also to the terminal, so the user can see it 
: >>more directly?
: > 
: > 
: > I don't know how to do this portably, but please feel free to modify
: > the implementation.  Right now, there is enough information in the
: > files that pkgsrc directs you to look at to figure out what's wrong
: > and how to fix it.

: For me, "echo foo 1>/dev/tty" has worked on all systems I know. Are 
: there any counterexamples?

  I don't think that is a good idea: if you redirect the build output
(as e. g. the bulk builds do), this warning would not be redirected.
One shouldn't break the unix convention of writing to stdout/stderr
without a very good reason.

: >>By the way, I would prefer to only create the fail-wrappers when 
: >>PKG_DEVELOPER is set. For a typical user (even more for unexperienced 
: >>users) this is just annoying.
: > 
: > 
: > I do not see how this is "annoying".  If the package doesn't invoke
: > the C++ compiler, then nothing happens.  If it does, and it doesn't
: > properly set USE_LANGUAGES, then it fails just like it would if
: > PKG_CXX=${FALSE}.  Where is the "annoying" part?

: Before PKG_CXX had been set to ${FALSE}, the package built, although not 
: using the c++ wrapper. I think it would be best to continue this 
: behavior for non-PKG_DEVELOPER users, so they get working packages (in 
: most cases). I think that for PKG_DEVELOPER users, the current behavior 
: is the right thing.

  That was my thought also.  However, it's not that simple.  For
current NetBSD, the packages will likely build and work, but when
building with gcc from pkgsrc, the run-time dependency on gcc (needed
for the standard C++ shared library) will not be recorded.  So I think
this should be on, even for non-PKG_DEVELOPERs.

  However, the sweep through pkgsrc fixing USE_LANGUAGES should have
been done before turning this on for all users.  But that's largely
water under the bridge now.