Re: QUOTES_IN_INSN

[Ian Lance Taylor <ian@airs.com>]

"Jan Beulich" <JBeulich@novell.com> writes:

> Currently, only ia64 defines this. In order to facilitate passing
> strings to macros I think the code protected by this needs to be
> generally enabled, otherwise stuff like
> 
> 	.macro m str
> 	 .asciz "\str"
> 	.endm
> 
> 	m "foo;bar"
> 
> doesn't work (whereas a plain
> 
> 	.asciz "foo;bar"
> 
> does obviously work).
> 
> Are there any contra-indications to such a change (i.e. are there
> targets that can't tolerate this)?

I would expect that this would work OK.

> In any case, shouldn't the code in read.c currently protected by
> this be dealing with '\n' specially (to at least increment the line
> counter)? What is the intended meaning of a quoted '\n' anyway?
> Shouldn't there rather be a warning about a missing closing quote
> (and the line be terminated there)?

I believe the MRI assembler allows you to quote a newline.  It's true
that bump_line_counters should be called in that case.  I don't know
about ia64.

> Additionally, as I'm seeing a number of similar issues in the macro
> handling code: Is there a formal specification somewhere about the
> expected behavior of both the assembler in general and the macro
> code in particular, so one could reasonably judge whether a certain
> construct being rejected or producing unexpected results is a bug

No.

Ian