Should the compiler signal error on (-) or (/ 1 0)? (At compilation time?)

[jpellegrini]

Not sure about this @egallesio . Tell me what you think..,.

If we turn off inlining,

stklos> (define (f) (/ 1 0))
;; f
stklos> (define (g) (-))
;; g

stklos> (disassemble f)

000:  PREPARE-CALL        
001:  ONE-PUSH            
002:  ZERO-PUSH           
003:  GREF-TAIL-INVOKE     0 2
006:  RETURN              
007:

stklos> (disassemble g #t)

000:  PREPARE-CALL        
001:  GREF-TAIL-INVOKE     0 0
004:  RETURN              
005:

But with the default (inlining turned on):

stklos> (define (f) (/ 1 0))
**** Error:
/: cannot make rational with null denominator
(type ",help" for more information)
stklos> (define (g) (-))
**** Error:
compiler-error: -: needs at least one argument
(type ",help" for more information)

And disassemble-expression signals an error:

stklos>  (disassemble-expr '(-) #t)
**** Error:
compiler-error: -: needs at least one argument

The previous one was a compilre error. The following is a generic error (because it was not caught in the ilining code - easy o fix actually):

stklos>  (disassemble-expr '(/ 1 0) #t)
**** Error:
/: cannot make rational with null denominator
	(type ",help" for more information)

So... Should the compiler crash when an inlinable expression is not calculable, or should we just, well, leave it there un-optimized, so it will crash when the code is executer? (I'd argue for signaling the error during compilation -- which is, leave it as it is -- as this is useful).