Re: [Sbcl-devel] bug[?] compiler/interpreter disagreement

[Raymond T. <to...@rt...>]

>>>>> "Alexey" == Alexey Dejneka <ade...@co...> writes:

    Alexey> Brian Spilsbury <br...@de...> writes:
    >> I have an odd compiler error, it looks like it is inferring
    >> that eltype cannot be a cons, but that's definitely weird since
    >> the known-def give it as returning a type-specifier which can
    >> be a list.
    >>
    >> The function works fine while interpreted, but compiling it
    >> causes it to break.
    >>
    >> Perhaps someone can see something I've overlooked (this is in
    >> sbcl-0.6.13)
    >>
    >> (defun pixarray-element-size (pixarray) (let ((eltype
    >> (array-element-type pixarray))) (cond ((eq eltype 'bit) 1)
    >> ((and (listp eltype) (eq (first eltype) 'unsigned-byte))
    >> (second eltype)) (t (error "Invalid pixarray: ~S." 
    >> pixarray)))))
    Alexey> Optimizer thinks PIXARRAY is of type (ARRAY *), therefore
    Alexey> ELTYPE is *, i.e. of type (MEMBER *), or, more generally,
    Alexey> of type SYMBOL. Here is the patch.

I note that SBCL has incorporated my array-element-type and coerce
optimizers.  Unless you've made some changes to it, I think the coerce
optimizer is buggy and will cause bad code to be generated.

(I'd do a diff, but I don't have the code anymore).

I also don't have my test case that demonstrates this bug,
unfortunately.  But it was sufficiently buggy for me to delete it from
my working sources. 

Ray