Re: [Sbcl-devel] STRING is too hairy for sequence functions.

[Christophe R. <cs...@ca...>]

David Lichteblau <da...@li...> writes:

> the following happens with 0.8.13.77.character.32:
>
> CXML(16): (deftype test () 'string)
>
> TEST
> CXML(17): (concatenate 'test "foo" "bar")
>
> debugger invoked on a SIMPLE-TYPE-ERROR in thread 6411:
>   STRING is too hairy for sequence functions.

I believe this is not a bug.  (The error message is definitely bad,
and it's possible that we should be lenient and accept this, but I
don't think that it is defined to work.)

From the page for CONCATENATE:

  If the result-type is a subtype of vector, then if the
  implementation can determine the element type specified for the
  result-type, the element type of the resulting array is the result
  of upgrading that element type; or, if the implementation can
  determine that the element type is unspecified (or *), the element
  type of the resulting array is t; otherwise, an error is signaled.

STRING is not, in general, a fully-determined type; even in mainline
CVS, it is a union of (VECTOR NIL) and (VECTOR CHARACTER), and even in
mainline your test doesn't work.  Why does 
  (concatenate 'string "foo" "bar") 
work, then?  Well, under "System Class STRING", it is specified that
STRING means (VECTOR CHARACTER) when used as an argument to sequence
creating functions (such as CONCATENATE).  So we special-case the
string symbol.  In case this motivates you to ask for your TEST type
to be likewise recognized, consider what you would expect to happen
with
  (deftype test2 () '(or (vector nil) (vector character)))
  (concatenate 'test2 "foo" "bar").

Cheers,

Christophe