Re: [Sbcl-devel] FSet bug/question

[Paul K. <pk...@gm...>]

On 26-Aug-08, at 5:34 PM, Scott L. Burson wrote:
> Quoting Paul Khuong <pk...@gm...>:
>>
>> Either way, is it so hard to (defvar *empty-vector* #())? I don't see
>> the point of using NIL instead of an empty vector; it doesn't even
>> save space!
>
> It doesn't save space, but comparing against NIL is very fast in all  
> implementations of which I am aware.  Dereferencing a special is  
> not, on the other hand, always fast.

Testing that a vector's length is zero should be pretty fast too, on  
any implementation that actually cares about execution speed.  
Especially these days, I'm fairly certain it's *at best* not a  
pessimisation to have a distinguished value instead of zero-length  
vectors. You're adding an, a priori, ill-predictable branch (unlike  
the type checks) to save a minuscule (nil?) amount of time.

>> I believe that special case is better asked for explicitly. The  
>> proper
>> way to achieve the desired effect would be to shadow CL:LENGTH and
>> define a new LENGTH *function* and declaim it inline or define a
>> compiler macro.
>
> FSet does in fact shadow CL:LENGTH, as Daniel said ("Where LENGTH  
> has been shadowed as ...").

The important point is to use a function, not a macro. If FSet did not  
shadow CL:LENGTH, asdf would have logged an ERROR, not a WARNING.

Paul Khuong