Re: [Sbcl-devel] Patch: string-*-streams and file-position, take 3

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

Nikodemus Siivola <tsi...@cc...> writes:

> The attached string-*-stream patch provides for the following
> behaviour and corresponding tests:
>
>  MAKE-STRING-INPUT-STREAM
>
>  * Observe FILE-POSITION :START and :END, and allow setting of
>    FILE-POSITION beyond the end of string, signalling END-OF-FILE only
>    on read.
>
>  * If the input stream is resized or otherwise mutated the effects are
>    seen in the stream as well.

I'm not sure I like this one.  Most of the rest makes sense to me, but
this I think is a bit weird.  I'm at least slightly supported by
WITH-INPUT-FROM-STRING on this one, which refers to the "Traversal
Rules and Side Effects" bit.  On the other hand, I can see the
attraction of having mutations to the string's data take effect.

>  (defun make-string-input-stream (string &optional
>  					(start 0) end)
>    #!+sb-doc
>    "Return an input stream which will supply the characters of STRING between
>    START and END in order."
> -  (declare (type string string)
> +  (declare (type (vector character *) string)

Heh.  Nice try, but now I can't do
  (make-string-input-stream (make-array 0 :element-type 'nil))
and I'm sure Paul Dietz will write a test for that eventually...

>  	   (type index start)
>  	   (type (or index null) end))
> -  
> -  (internal-make-string-input-stream
> -   (coerce string 'simple-string)
> -   start
> -   (%check-vector-sequence-bounds string start end)))
> +  (let ((end (%check-vector-sequence-bounds string start end)))
> +    (internal-make-string-input-stream
> +     string
> +     start
> +     (if (or (adjustable-array-p string) (array-has-fill-pointer-p string))
> +	 (lambda () (min end (length string)))
> +	 (constantly end)))))

I have a feeling that this might hurt performance in a non-trivial
way: not only are we calling arbitrary functions at each read, but
each read will involve finding the underlying data.  Perhaps what we
want to do is something like

 (with-array-data ((string string) (start start) (end end))
  (internal-make-input-stream
   string ; now the simple, underlying data, so mutations will be reflected
   start
   end ; no need for functions, since we disallow changing the string structure
  ))

so that the stream string is now simple [ though not necessarily
(simple-array character (*)); it could be (simple-array nil (*)); I
think that's probably acceptable performancewise ]

Does that make sense?

Cheers,

Christophe
-- 
http://www-jcsu.jesus.cam.ac.uk/~csr21/       +44 1223 510 299/+44 7729 383 757
(set-pprint-dispatch 'number (lambda (s o) (declare (special b)) (format s b)))
(defvar b "~&Just another Lisp hacker~%")    (pprint #36rJesusCollegeCambridge)