PATCH to defer cpp_read_main_file and related changes

[Neil Booth]

Per Bothner wrote:-

> This patch does a couple of related things.
> It's more "cleanups" to bring the structure of the trunk
> closer to the compile-server, and I think make things
> a little more consistent.

[...]

> 2003-09-24  Per Bothner  <pbothner@apple.com>
> 
> 	* cpplib.c (_cpp_pop_buffer):  Do generate a _cpp_do_file_change
> 	callback even when popping the main file.
> 	* c-lex.c (fe_file_change):  Handle a NULL new_map.
> 	* fix-header.c (cb_file_change):  Likewise.
> 	* c-ppoutput.c (pp_file_change):  Likewise.
> 	* c-opts.c (c_common_post_options):  Don't call cpp_read_main_file yet.
> 	(c_common_parse_file):  No longer need to call cpp_read_main_file
> 	when file_index > 0 (as in multi-file or server compiation).
> 	(finish_options):  Change to <built-in> is an LC_ENTER, not LC_RENAME
> 	as this now happens before cpp_read_main_file.  Add a matching
> 	closing LC_LEAVE, with magic to defer push_command_line_include.
> 	Then call cpp_read_main_file when we're done with options.
> 	(push_command_line_include):  The LC_RENAME here is no longer needed.
> 	(fe_file_change):  Handle NULL new_map, and simplify.
> 	* cpplex.c (_cpp_get_fresh_line):  Revert my no-longer-needed
> 	08-28 change, since we're never called with a NULL buffer.
> 	(_cpp_lex_direct):  Likewise.
> 	* cpptrad.c (_cpp_read_logical_line_trad):  Likewise.
> 	Return false if buffer is NULL at end.

I'm happy with this if you are; it does seem to improve some things.
If you haven't already, could you document clearly somewhere what a
NULL map means?

Thanks,

Neil.