[M3commit] CVS Update: cm3

[Tony Hosking]

On 25 Sep 2010, at 02:16, Jay Krell wrote:

> CVSROOT:	/usr/cvs
> Changes by:	jkrell at birch.	10/09/25 02:16:13
> 
> Modified files:
> 	cm3/m3-sys/m3cc/gcc/gcc/m3cg/: parse.c 
> 
> Log message:
> 	tighten up some asserts
> 	size and align >= 1, not just >= 0
> 	er, no, size can be 0, for empty records
> 	or fields of their type, so leave it alone and align >= !!size
> 	
> 	work in progress
> 	types/debugging for enums and typedefs
> 	(when you say TYPE a = {b,c};
> 	{b,c} is anonymous enum
> 	and a is a typedef/typename for.
> 	consider you can say:
> 	TYPE a = {b,c};
> 	TYPE a2 = {b,c};
> 	it is equivalent to typedef enum {b,c} a, a2;)
> 	not yet working, not yet enabled
> 	
> 	don't widen return types
> 	leave the mechanism though because
> 	this might be where we adjust struct vs. struct*, which
> 	is another thing configure -enable-checking complains about,
> 	though that is specific to module initializers I think,
> 	not all functions in general
> 	
> 	set the alignment of bitfields 1 bit, other fields to BITS_PER_UNIT (8 bits)
> 	part of attempted work in progress to generate decent trees
> 	instead of obviously poor ones; aka to actually maybe finish
> 	this backend instead of leaving it very incomplete for years
> 	(not yet enabled)
> 	
> 	Do a better job of skipping the prefix of module/segment names.
> 	M_Main => "Main" instead of "ain".
> 	Instead of always skipping 3, skip up to the first
> 	underscore, if it is at position 0, 1, or 2.
> 	Right? Or did I regress this somewhere since release?

This must have been a regression!

> 	
> 	layout_decl(var, 0)
> 	=> layout_decl(var, 1)
> 	alignment = 0 is bad, causes divide by zero with plain -g