Re: immediate strings

[Paul Eggert <eggert@cs.ucla.edu>]

On 11/26/11 06:30, Dmitry Antipov wrote:
> IIUC, string length can't be larger than most positive Lisp integer,
> which is smaller than BITS_PER_EMACS_INT - 1 size bitfield anyway.

Ah, sorry, I should have mentioned that I was thinking in the
context of the patch for bug#9874, which improves performance on
32-bit platforms configured --with-wide int (so EMACS_INT is 64
bits, ptrdiff_t 32) by shrinking struct Lisp_String as follows:
 
 struct Lisp_String
   {
-    EMACS_INT size;
-    EMACS_INT size_byte;
+    ptrdiff_t size;
+    ptrdiff_t size_byte;
     INTERVAL intervals;		/* text properties in this string */
     unsigned char *data;
   };

In this environment we can still steal one bit from size and from size_byte
without shrinking the maximum string length, but we'd need to use some
other way to mark a string as multibyte, since size_byte would become an
unsigned bitfield.

I'm more optimistic than Eli and Stefan in thinking that something along
the lines of a size+byte_size+intervals packing approach would yield a
performance win that's worth the maintenance hassle.  But of course all
this will require a reasonable amount of hacking and measuring.