WORD_SWITCH_TAKES_ARG cleanup

This patch, relative to a tree with
<http://gcc.gnu.org/ml/gcc-patches/2010-10/msg02102.html> (pending
review) applied, cleans up uses of WORD_SWITCH_TAKES_ARG.

With this patch applied, all options taking a single separate argument
are now known to .opt files; WORD_SWITCH_TAKES_ARG remains solely for
those Darwin options taking more than one argument, until the .opt
machinery has support for such options added.

FreeBSD accepted several options (passed through to the linker by
specs), which go in a new freebsd.opt file.  The FreeBSD
-dynamic-linker option, which was accepted by the
WORD_SWITCH_TAKES_ARG, had no corresponding spec to pass it through to
the linker, so I did not include it in the .opt file.

(There are a lot of *broken* specs for different targets of the form
"%{!dynamic-linker:-dynamic-linker /lib/ld.so}".  These treat
-dynamic-linker as a no-argument option whose effect is to disable the
passing of a -dynamic-linker option to the linker.  The intended
effect is clearly to avoid passing such an option if the user
specified a dynamic linker explicitly - but it doesn't have that
effect as there are no specs to pass through an explicit
-dynamic-linker option and its argument.  Thus, all these specs should
be simplified in a subsequent cleanup to remove the
"%{!dynamic-linker:}" parts.)

Some targets, including FreeBSD, accepted -rpath in
WORD_SWITCH_TAKES_ARG.  These were made to use a new rpath.opt.  For
the definition in alpha.h, alpha-dec-osf5.1 appeared to be the
relevant target, since the only Alpha targets with relevant specs are
FreeBSD and OSF.

Three Darwin options -arch, -arch_only, -precomp-trustfile had no
corresponding specs that would accept them; thus they are not added to
darwin.opt.  I suppose the inclusion in WORD_SWITCH_TAKES_ARG was
mistakenly added to the FSF tree without the actual implementations of
those options.

There was some CRIS code to deal with a -target option, apparently
working around some problem with Sun make.  It makes no sense for such
a workaround to go in architecture-specific code; in some cases we
accept options for compatibility with a system compiler, but that's
OS-specific and the concept of such a system compiler to be compatible
with doesn't make sense for CRIS.  Thus, this workaround is removed by
this patch.

Bootstrapped with no regressions on x86_64-unknown-linux-gnu.  Also
tested building cc1 for crosses to: alpha-dec-osf5.1 i686-interix3
mips-sgi-irix6.5 i686-freebsd8.1 cris-elf i686-darwin.  (As with the
previous patch, i686-interix3 fails for unrelated reasons.)  OK to
commit?

2010-10-27  Joseph Myers  <joseph@codesourcery.com>

	* config/alpha/alpha.h (WORD_SWITCH_TAKES_ARG): Remove.
	* config/cris/cris.h (WORD_SWITCH_TAKES_ARG): Remove.
	(CC1_SPEC): Don't handle -target.
	* config/darwin.h (WORD_SWITCH_TAKES_ARG): Remove single-argument
	options.
	* config/darwin.opt (Zallowable_client, Zbundle_loader,
	Zdylib_file, Zexported_symbols_list, Zfn_seg_addr_table_filename,
	Zimage_base, Zinit, Zinstall_name, Zmultiply_defined,
	Zmultiplydefinedunused, Zseg_addr_table, Zsegs_read_only_addr,
	Zsegs_read_write_addr, Zumbrella, Zunexported_symbols_list,
	Zweak_reference_mismatches, client_name, compatibility_version,
	current_version, dylinker_install_name, pagezero_size,
	read_only_relocs, seg1addr, segs_read_only_addr,
	segs_read_write_addr, sub_library, sub_umbrella): New.
	* config/freebsd-spec.h (BSD_WORD_SWITCH_TAKES_ARG): Remove.
	* config/freebsd.h (WORD_SWITCH_TAKES_ARG): Remove.
	* config/freebsd.opt: New.
	* config/interix.h (WORD_SWITCH_TAKES_ARG): Remove.
	* config/mips/iris6.h (WORD_SWITCH_TAKES_ARG): Remove.
	* config/rpath.opt: New.
	* config.gcc (*-*-freebsd*): Use rpath.opt and freebsd.opt.
	(alpha*-dec-osf5.1*, i[34567]86-*-interix3*, mips-sgi-irix6.5*):
	Use rpath.opt.

WORD_SWITCH_TAKES_ARG cleanup

Commit Message

Comments

Patch