Steel Bank Common Lisp

Daniel Herring <dherring@...> writes:

> ; in: DEFUN TEST-TUPLE-OPERATIONS
> ;     (FSET::LENGTH FSET::TUPLE-KEYS)
> ; --> LET IF
> ; ==>
> ;   0
> ;
> ; caught WARNING:
> ;   This is not a (OR (SINGLE-FLOAT (0.0)) (DOUBLE-FLOAT (0.0d0)) (INTEGER 1)):
> ;     0
> ;   See also:
> ;     The SBCL Manual, Node "Handling of Types"

This is partly an instance of bug #117 (see the BUGS file) and partly
telling you something vaguely useful: if (LENGTH TUPLE-KEYS) returns
0, then the call to RANDOM will be an error (because (RANDOM 0) is not
valid code).  This shows up in this code, I think, because you have
shadowed LENGTH with something with an inline expansion.

Best,

Christophe

Daniel Herring <dherring@...> writes:
> Moved; new readers please find my first message on sbcl-devel.  For now, 
> the only confirmed bug is my confusion; the optimizer is doing something I 
> don't understand.
> 
> To rephrase the original questions,
> - is the FSET::LENGTH "optimization" really needed for SBCL?

Only you can decide whether it's really needed. The observation that
SBCL will not optimize CL:LENGTH for a type that's either nil or a
vector is true.

> - what's causing this warning?

SBCL thinks NKEYS can be 0, which would be an invalid input value to
RANDOM.

> - why does it go away with the optimize debug?

Because that disables enough optimizations to make sbcl no longer
think there's a reasonable chance of LENGTH returning 0.

-- 
Juho Snellman

>> ; in: DEFUN TEST-TUPLE-OPERATIONS
>> ;     (FSET::LENGTH FSET::TUPLE-KEYS)
>> ; --> LET IF
>> ; ==>
>> ;   0
>> ;
>> ; caught WARNING:
>> ;   This is not a (OR (SINGLE-FLOAT (0.0)) (DOUBLE-FLOAT (0.0d0))
(INTEGER 1)):
>> ;     0
>> ;   See also:
>> ;     The SBCL Manual, Node "Handling of Types"
> This is partly an instance of bug #117 (see the BUGS file) and partly
telling you something vaguely useful: if (LENGTH TUPLE-KEYS) returns 0,
then the call to RANDOM will be an error (because (RANDOM 0) is not
valid code).  This shows up in this code, I think, because you have
shadowed LENGTH with something with an inline expansion.


Thanks Christophe and Juho; your comments cleared things up nicely and
provide a path forward.

- Daniel