Re: A PREFIX_SEPARATOR bug in binutils 2.9

[Alan Modra <alan@spri.levels.unisa.edu.au>]

> 
>    From: hjl@lucon.org (H.J. Lu)
>    Date: Tue, 21 Apr 1998 12:34:46 -0700 (PDT)
> 
>    > Why not encourage people to write the prefix on a separate line, which
>    > works for both a.out and ELF?
>    > 
> 
>    That is a separate issue. Since PREFIX_SEPARATOR is wrong on ELF, we
>    should fix it. I think we can change it
> 
>    #define PREFIX_SEPARATOR '\\'
> 
>    so that everyone will be happy.
> 
> I am questioning the need for PREFIX_SEPERATOR at all.  I don't see
> any particular reason to use it on ELF if it is not required and if no
> other ELF assembler supports it.

The only reason I can think of is that we may want to treat prefixes
on a separate line differently to prefixes specified "with the
instruction"

eg.  We might find it desirable for
	ds
	lgdt	0
to emit the ds prefix while
	ds/ldgt 0  (or whatever the syntax is)
or
	lgdt	%ds:0
shouldn't. ie. gas realises that ds is the default segment.

Martynas (martynas@nm3.ktu.lt) has been working on a 16 bit capable
gas, and he's keeping track of address size prefixes as a means to
tell the assembler that the next instruction is a 16 bit one.

For instance `addr16; mov 0,%ebx' generates   67 8b 1e 00 00	
       while         `mov 0,%ebx' generates      8b 1d 00 00 00 00
So the `addr16' can't just be emitted; It considerably effects
following code, and not just offset sizes, but modrm bytes too.

I'm wondering if this is really a good idea.  Maybe it would be better
if addr16 didn't effect the code generated, and instead we require a
`.code16' directive.

How do other unix assemblers handle 16 bit code?  Or don't they?
Can someone with access to Unixware, SCO, etc. assemblers do some
experimentation please?