argument checking: use "assume" or always call error?

[jpellegrini]

Hi @egallesio !

Some SRFIs use assume in arguments checking (like SRFI-214, flexvectors). Others always raise an error. Maybe this should be standardized? Either always one of them, or always the other? I can slowly change the SRFIs to always use assume if you think this is OK.

[egallesio]

The pro of assume is that the test is not done if you are not in debug mode (it's expansion is #void, so it's fast :smile:). The con is that if the user doesn't set the debug flag, its code could fail miserably. In this way, its very similar to the C assert function.

IMHO, assume should be used for internal functions to assume that a certain property is true at a given point, but if the test involves a data given by the user error is better. So both can probably cohabit in a program.

For instance:

(define (public-foo n)
  (unless (integer? n)
    (error "bad integer" n))
  (internal-bar (* n n)))

(define (internal-bar n)
  (assume (positive? n))
  .....
  .....)

This is just my opinion, conforming of what STklos is able to do with assume (SRFI 145 depicts situations the compiler cannot handle).

[jpellegrini]

Ok, I'll review some of the SRFIs later and see if we need to change anything