Solaris 8/SPARC libgcj bootstrap broken by recent fix-header change

[Rainer Orth]

Neil Booth writes:

> > This declaration is at line 246, conflicting with the original
> > 
> > extern int kill(pid_t, int);
> > 
> > at line 57.  This didn't happen in my last successful mainline bootstrap on
> > 20030224, where signal.h isn't modified by fix-headers.
> > 
> > Since you recently made several patches affecting fix-header, I suspect
> > that one of them is the culprit.
> > 
> > I'll start hunting down the exact one right now.
> 
> Maybe, though I can't think how this might have been caused.

me neither, but I've just tracked down the patch introducing this
regression: 

2003-03-15  Neil Booth  <neil@daikokuya.co.uk>

	* Makefile.in: Update.
	* c-common.h (cb_register_builtins): Rename c_cpp_builtins.
	* c-lex.c (init_c_lex): Register builtins hook is dead.
	* c-opts.c (COMMAND_LINE_OPTIONS, missing_arg): Handle -A, -D and -U.
	(c_common_decode_option): Don't call cpp_handle_option.
	Handle -A, -D and -U.
	(handle_deferred_opts): Simplify.
	(finish_options): Define builtins and command line macros.
	* c-ppoutput.c (init_pp_output): Register builtins hook is dead.
	* cppinit.c: Don't include intl.h.
	(init_builtins): Rename cpp_init_builtins.  No hook to call.
	(init_library): Don't need to sort options.
	(cpp_create_reader): Don't set pending.
	(cpp_destroy): Don't free pending.
	(struct pending_option, cl_directive_handler, struct cpp_pending,
	APPEND, free_chain, new_pending_directive, parse_option, opt_comp,
	cpp_finish_options, COMMAND_LINE_OPTIONS, DEF_OPT, struct cl_option,
	cl_options, cpp_handle_option): Remove.
	* cpplib.h (struct cpp_pending, register_builtins, cpp_handle_option,
	cpp_finish_options): Remove.
	(cpp_init_builtins): New.
	* fix-header.c (read_scan_file): Update to handle -D.  Fix
	handling of -I.  Replace call to cpp_finish_options.

Before this patch, fix-header is only applied to curses.h, afterwards also
signal.h, stdlib.h, and time.h are `fixed', breaking bootstrap as
described.

	Rainer