Re: RFC: Do not install specs file

[Alexandre Oliva <aoliva at redhat dot com>]

On Dec 20, 2004, Mark Mitchell <mark@codesourcery.com> wrote:

> Does anyone object to this change (which I have tested on
> i686-pc-linux-gnu, but not installed)?

Yes, it will break mips64-linux-gnu.  config/mips/t-linux64 says:

# This arranges for us to extract the built-in specs of the compiler
# into specs.install.  It's this file that we're going to install as
# specs.  Howver, at toolchain build time, we must use specs that
# enforce the defaults that the compiler expects, namely, the o32 ABI,
# instead of n32, that we want to use as the default post install.
SPECS = specs.install
specs: specs.install
	# Changing the build-time default ABI from n32 to o32
	# The post-install default ABI will remain n32
	@if test "x`grep '!mabi=\*:-mabi=n32' specs.install`" = \
		"x%{!mabi=*:-mabi=n32}"; then \
	  sed 's,\(!mabi=\*:-mabi\)=n32,\1=32,' < specs.install > $@ ; \
	else \
	  echo failed, gcc/config/mips/t-linux64 needs updating >&2; \
	  false; \
	fi


I'm ok with enclosing the changes between ifeq ($(SPECS), specs) /
endif

Hopefully we'll some day be able to get rid of this hack.  I'm told we
can already, but I haven't had a chance to verify this.  The
requirements IIRC are that the o32 libgcc's soname must be the same as
that of a mips-linux toolchain and that libgcc, libstdc++ and any
other libs for o32 be in the GCC internal lib dir, not in lib32 (the
dir used for the n32 ABI, the default in mips64-linux).

-- 
Alexandre Oliva             http://www.ic.unicamp.br/~oliva/
Red Hat Compiler Engineer   aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist  oliva@{lsd.ic.unicamp.br, gnu.org}