[Openmcl-devel] Compiler warnings

[james anderson]

On 2009-10-19, at 23:29 , Tim Bradshaw wrote:

>
> On 19 Oct 2009, at 21:23, james anderson wrote:
>
>> but, if the implementation is of common lisp, it cannot do this.
>> if it does do do this, then all bets are off, as we have entered a
>> realm outside of the spec.
>
> I think we all know this.

actually, no, i don't.
at least if "know" is to mean, to understand the implications.
the tenacity of this confusion made me look. again. and observe how  
definitively outside the spec it is. and conclude, that i would  
actually be just as happy with a lisp which required one to use a  
definition form. a lisp which did not create non-constant value  
bindings with both indefinite scope and indefinite extent.
for the sake of clarity.

is there some use case which argues against such strictness?

> The point is that it is *useful* to be able to do this, and so  
> implementations support this (as extensions to the standard,  
> obviously).  They do it in various ways, of which CCL's is far from  
> being the worst.  But (as you say) people need to understand that  
> this is outside the standard.
>
> I'm pretty sure that there's never been a Lisp in the CL family  
> which did not support SETQ at the top level.  Apart from anything  
> else, if it's not supported people would immediately impement it in  
> terms of SET.
>
> I also think that implementations should clearly be allowed to  
> support top-level-SETQ-of-unbound-names while being conformant  
> otherwise, and I think they probably are because a lot of latitude  
> is allowed in the way the top-level works.

for better, or worse.