Re: [Sbcl-devel] odd behaviour of MAP over STRINGs

[Anton K. <an...@sw...>]

Gabriel Dos Reis <gd...@in...> writes:

> We do not use directly TYPEP or SUBTYPEP.  But, we do rely
> heavily on function type proclamations involving those type specifiers.
> Is such usage also not-recommended?

Everything is fine unless you really need 'STRING for some of your
declarations (in the sense of accepting/returning both base-strings and
non-base strings and not being sure). 

Note that you may be working with base-strings unwittingly, even if you
are not requesting their creation anywhere in your code:

 (type-of (symbol-name 'list)) => (SIMPLE-BASE-STRING 4)

Default stream element type for SBCL's OPEN is BASE-CHAR too, so another
possibly-unexpected source of base-strings in your code is READ-LINE.

The declaration using a type that expands to STRING is the right one for
code that may receive a base-string; if you replace it with a type
expanded to (VECTOR CHARACTER), then base-string will cause a failure
(btw, for any level of SAFETY but the most reckless one, SBCL will
normally detect a type error in this case, instead of doing something
dangerous).

> OK.  So, just to make sure I understand: it is OK to use the type
> specifiers as long
> as they expand to (VECTOR something), but not to the type specifier STRING
> or SIMPLE-STRING.  Is that correct?

Yes. BASE-STRING and SIMPLE-BASE-STRING are fine too: their
specialization is fixed and they aren't "hairy", because there is
nothing like VERY-VERY-BASE-STRING specialized for some
VERY-VERY-BASE-CHAR subset of possible BASE-CHARs.

For /declarations/, any type (be it deftyped or predefined) is acceptable,
so when you've figured out the right type to declare, feel free to use a
deftype'd name.

-- 
Regards, Anton Kovalenko
+7(916)345-34-02 | Elektrostal' MO, Russia