[Patch debug] Do not emit debug sections until they are required.

[IainS]

This is part 2 of a campaign to get rid of differences in section  
ordering between debug & non-debug cases.

This moves the three remaining switch_to_section()  from  
dwarf2out_init () and places them directly before the code that emits  
the respective sections.

In order to do this, macinfo data are saved in a vec, and emitted from  
dwarf2out_finish () rather than 'on the fly'.

Whilst the benefit is mainly for targets that emit sections in the  
order they are encountered...
...  I also expect it is of help to anyone who has occasion to compare  
debug/non-debug asm files.

bootstrapped on i686-darwin{8,9,10} and x86_64-unk-linux.
OK for trunk?
Iain

gcc:

	* dwarf2out.c (macinfo_entry): New struct.
	(output_comp_unit): Note that we will emit an info section.
	(output_pubnames): Only try to refer to the info section if it is  
emitted.
	(dwarf2out_start_source_file): Save data in a macinfo entry rather than
	emitting directly.
	(dwarf2out_end_source_file): Likewise.
	(dwarf2out_define): Likewise.
	(dwarf2out_undef): Likewise.
	(output_macinfo): New.
	(dwarf2out_init): Do not emit debug section switches here.
	(output_indirect_string): If any strings are emitted, then emit the  
section
	switch first, and once only.
	(dwarf2out_finish): First switch to debug_abbrev_section here.
	debug_line_section,  debug_macinfo_section, Likewise.

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: 166173-dwarf-to-end-pt2.txt
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20101102/1b5f6dd8/attachment.txt>
-------------- next part --------------



===

After this, there are tthree remaining issues (AFAICT);
1. frame_debug emitted before eh_frames (from toplev.c rather than  
dwarf2out - is there some special reason for this?)
2. darwin's indirection stubs are emitted from end_file () and they  
really need to be between the _eh frames and the debug.
3. the lto marker is emitted after everything else.
I had patches for these .. but they've got some bit-rot ..
.. this is not a mega-high-priority - but it would be good to shake  
down some of the darwin debug hassles for 4.6.