Re: [Sbcl-devel] [PATCH] compile-time type errors

[William H. N. <wil...@ai...>]

On Thu, Sep 12, 2002 at 10:41:20AM +0400, Alexey Dejneka wrote:
> This patch removes :ERROR continuation type check type. Forms with
> compile-time type errors are replaced with call of ERROR.

I think I like the basic idea, but I have some quibbles with the
details of the current patch.
 
With the patch, my system doesn't pass tests/run-tests.sh, failing in
defstruct.impure.lisp. That test code isn't especially clear, but it
looks to me as though it should be legal. It fails on the
(TEST-VARIANT VANILLA-STRUCT :BOA-CONSTRUCTOR-P T) case, and it looks
to me as though in that case the OPTIONAL-TEST-P flag should be true,
so the REFCOUNT will default to the OPTIONAL-TEST2 value, i.e. 1,
which satisfies the type constraints.

> Drawback:
>   Unhelpful error message:
> 
>   (defun foo (x)
>     (declare (type integer x))
>     (1+ (the float x)))
> 
>   * (foo 14)
>   debugger invoked on condition of type SIMPLE-TYPE-ERROR:
>     Execution of a form with compile-time type error.

Also, I think all instances of TYPE-ERROR are supposed to support
TYPE-ERROR-DATUM and TYPE-ERROR-EXPECTED-TYPE, and the TYPE-ERRORs 
that I encountered didn't have those slots bound.
 
> Advantage:
>   Internals are cleaner. :ERROR combination kind is only used for
>   invalid syntax of a call (e.g. invalid number of arguments).

As before, this sounds OK.

-- 
William Harold Newman <wil...@ai...>
"Unofficially, I advise that we erase the records and forget the whole thing."
  -- <http://www.terrybisson.com/meat.html>
PGP key fingerprint 85 CE 1C BA 79 8D 51 8C  B9 25 FB EE E0 C3 E5 7C