Would be nice if 'typep' signalled on invalid type specifiers

[slburson]

[Previously filed on gitlab.common-lisp.net, but that doesn't seem to have been the right place.]

AFAIK, CL provides no portable way to determine whether a symbol has been deftyped. The only semi-portable way I know to do it is to try passing it to typep and see whether I get an error:

(defun is-type? (x)
  (handler-case 
       (progn (typep nil x) t) 
     (error () nil)))

This works on SBCL, CCL, ECL, CLASP, Allegro, and LispWorks, but fails on ABCL:

> (typep nil 'garbage)
NIL

It is true that the spec for typep says

The consequences are undefined if the type-specifier is not a type specifier.

but it would be nice if ABCL would signal an error, as other implementations do. Wouldn't it make sense to do this in normalize-type?

[slburson]

I tried following my own suggestion and modifying normalize-type, but it didn't work easily. Building ABCL turned up several places where the current behavior, returning unknown types unchanged, was being depended on. I gave up.

I did find another way, though, that seems to work:

(defun type-specifier-name? (x)
  (let ((x (if (listp x) (car x) x)))
    (or (get x 'system::deftype-definition)
        (find-class x))))

Feel free to close this.

[easye]

Hmm. I'm going to keep this open if you don't mind: being able to determine whether a given symbol has been mentioned to type specifier system in the current session is something we should have anyways. Until we have a suitably formal post-ANSI specification process, most common behavior is best practice.

After looking a bit this morning, the Bear's typep mechanism could use some love anyways.

Thanks for the followup.