rfi: immutable strings

rfi: immutable strings

[Andy Wingo]

Hello,

Guile needs immutable strings.  String literals should be immutable, so
that its characters can come directly from objcode.  Also, the result of
(symbol->string 'foo) should be immutable; currently symbol->string
conses up a new shared substring every time.

In practice this means a flag on the stringbuf.

Takers? :)

Andy
-- 
http://wingolog.org/

Re: rfi: immutable strings

[Mike Gran]

> Hello,
> 
> Guile needs immutable strings.  String literals should be immutable, so
> that its characters can come directly from objcode.  Also, the result of
> (symbol->string 'foo) should be immutable; currently symbol->string
> conses up a new shared substring every time.

How is this different from read-only strings?

(BTW, you guys are doing some cool stuff right now.  I wish I could dig in,
but I'll be AWOL for a few more weeks trying to get updates of my other
projects out the door.)

-Mike

Re: rfi: immutable strings

[Ludovic Courtès]

Hi!

>> Guile needs immutable strings.  String literals should be immutable, so
>> that its characters can come directly from objcode.  Also, the result of
>> (symbol->string 'foo) should be immutable; currently symbol->string
>> conses up a new shared substring every time.
>
> How is this different from read-only strings?

Indeed, statically allocated stringbufs are immutable, for instance (see
‘SCM_IMMUTABLE_STRING’ in ‘snarf.h’.)

Read-only strings are also used elsewhere:

--8<---------------cut here---------------start------------->8---
scheme@(guile-user)> (string-set! (symbol->string 'foo) 2 #\x)
ERROR: In procedure string-set!:
ERROR: string is read-only: "foo"
--8<---------------cut here---------------end--------------->8---

Thanks,
Ludo’.

Re: rfi: immutable strings

[Andy Wingo]

On Fri 07 Jan 2011 19:52, Mike Gran <spk121@yahoo.com> writes:

>> Guile needs immutable strings.
>
> How is this different from read-only strings?

Seems to be the same.  We should have read-only string API though;
currently there is only substring API.  One should be able to allocate a
read-only string directly, like via
scm_from_static_latin1_{stringn,symboln}.  Also, strings loaded from
objcode should be read-only, and share storage with the objcode itself;
perhaps?

>> Also, the result of (symbol->string 'foo) should be immutable;
>> currently symbol->string conses up a new shared substring every time.

Though the result is immutable, it does cons up a shared substring,
which involves all kinds of mutex and allocation shenanigans.

Andy
-- 
http://wingolog.org/