Re: World broken in stage 1.1

On Wed, Aug 11, 2004 at 04:26:40PM -0600, M. Warner Losh wrote:
> In message: <20040811221348.GB93983_at_ip.net.ua>
>             Ruslan Ermilov <ru_at_FreeBSD.org> writes:
> : On Wed, Aug 11, 2004 at 04:04:20PM -0600, M. Warner Losh wrote:
> : > In message: <20040810231044.GA70020_at_xor.obsecurity.org>
> : >             Kris Kennaway <kris_at_obsecurity.org> writes:
> : > : More fallout from the wonderful new make(1) semantics? 
> : > 
> : > I think it is speficially related to the changes to src/Makefile and
> : > src/Makefile.inc1.  If I s/${_+_}//g on those two files, it appears
> : > that I can buildworld again (at least it doesn't die right away).
> : > 
> : How exactly does it die for you, please provide some logs.
> 
> The problem is due to the following in share/mk/sys.mk:
> 
> .if !empty(.MAKEFLAGS:M-n) && ${.MAKEFLAGS:M-n} == "-n"
> _+_             ?=
> .else
> _+_             ?=      +
> .endif
> 
> This should have a third clause:
> 
> .if ${MAKE_VERSION} >= 5200408030 && !empty(.MAKEFLAGS:M-n) && ${.MAKEFLAGS:M-n} == "-n"
> 
> so that _+_ is defined to be nothing for those versions of make that
> don't yet support this new feature of posix.
> 
Why?  It's not supposed that you try an old make(1) binary with
the new share/mk/sys.mk.  src/Makefile will upgrade make(1) if
necessary, then use it (or the old make if it's adequate) with
new share/mk stuff to call Makefile.inc1.  If you throw away
the -m from your "make" command in your build script, it will
work.

We want to be able to use the new make(1) features in our
share/mk files right after when they made available.  That was
one of the main ideas of the make(1) regression testing feature
in src/Makefile.

It's like we now handle backward issues in our C sources -- we
agreed not to pollute sources with __FreeBSD_version checks,
and we keep the necessary bootstrapping glue in a separate
place.  The same applies here -- one is free to use new make(1)
features in src/ makefiles and in share/mk support files.
src/Makefile and tools/regression/usr.bin/make guarantee that
we'll always use the make(1) binary compatible with the
feature set used in our .mk and makefiles.

Please update your build scripts...


Cheers,
-- 
Ruslan Ermilov
ru_at_FreeBSD.org
FreeBSD committer