Re: Locale-Optimized Sort

[Jim Meyering]

Thanks for the heads-up.
Even without looking at the actual bug report, I admit
that this is probably a good opportunity for optimization.
Of course, it'd be a trade-off (memory for speed), but it
sounds reasonable.

If you dive in, be sure to use the latest test release
(available soon):

  ftp://alpha.gnu.org/gnu/fetish/textutils-2.0.11.tar.gz

Jason Bucata <address@hidden> wrote:
| This is regarding GNU sort in textutils 2.0, as distributed in Debian potato
| and woody (if that makes a difference).
|
| I was poking through Debian bug reports, when I noticed bug 62803 which says
| that sort is too slow for non-C locales.  It was put under libc6 since (it
| was asserted) the only way to speed up sorting was to speed up strcoll--you
| can't get the blazing speed of strcmp outside the C locale.
|
| Is there some reason why strxfrm can't be used here?  It seems like a
| perfect application of strxfrm, if you compute the strxfrm'd values for each
| of the keys once (or once each time the line is read from a tempfile), cache
| them in a linked list off of struct line, and compare those in your main
| sort routine with ludicrous speed.
|
| Just for grins I started tinkering with the source (dontcha just love being
| able to do that? ;) ) to see if I could make it do just that, but at 2 AM
| local time it's more effort than I want to sink into it.  Besides, I
| figured there had to be a legitimate reason why that hasn't been done yet,
| since there are plenty of great minds looking at this source.
|
| So then, Enquiring Minds Want To Know(TM).
|
| Jason B.