Re: Group sorting

[dreschs@mpd.tandem.com (Sten Drescher)]

larsi@ifi.uio.no (Lars Magne Ingebrigtsen) said:

LMI> dreschs@mpd.tandem.com (Sten Drescher) writes:
>> (setq gnus-group-sort-function '(gnus-group-sort-by-level
>> gnus-group-sort-by-unread))
>> 
>> This would first do a complete sorting by level, then a complete
>> sorting by unread - 2 complete sorts.  I'd want it to work in a
>> single sort - the sort would compare the unread keys, and _only_ if
>> they matched, compare the level keys.

LMI> Ok, that would be workable.  Gnus could transmogrify that list to

LMI> (lambda (g1 g2)
LMI>   (or (gnus-group-sort-by-unread g1 g2)
LMI>       (and (not (gnus-group-sort-by-unread g2 g1))
LMI> 	   (gnus-group-sort-by-level g1 g2))))

LMI> and give that as the predicate to sort.  Won't this give you the
LMI> same results as first using `gnus-group-sort-by-level' as a
LMI> predicate and then `gnus-group-sort-by-unread' as a predicate for
LMI> sort?

LMI> I have no idea whether it will be faster, though...

	I don't know, because I don't understand the elisp - what is it
doing?  Is it still sorting the entire list of groups between g1 and g2,
or does that fragment just compare two groups?  If each
gnus-group-sort-by-x sorts a list of groups, then no, it appears that it
will be worse, as you are doing three sorts instead of two!

	As it is now, two entire sorts are done with one key each -
every group is sorted by level, then every group is sorted by
unread. I'm proposing that one sort be done with two keys - every group
is sorted by unread, then ONLY IF the unread counts are equal are the
two groups being compared sorted by level.  This is a _significant_
difference in sort time, as only a small percentage of groups will have
identical unread counts.  This difference might be noticable with a
small number of groups, but even with only about 2500 groups I can tell
the difference between one and two sorts, and I'm dreading seeing even a
single sort with the 14,000 groups my ISP carries.  Remember, the time
for a bubble sort grows as the square of the number of sorted items.
However,

LMI> (`gnus-group-sort-by-level' should really be called
LMI> `gnus-group-level-less' or something since it is a `less' function
LMI> and not an actual sorting function...)

LMI> -- Home is where the cat is.



-- 
#include <disclaimer.h>				/* Sten Drescher */
To get my PGP public key, send me email with your public key and
	Subject: PGP key exchange
Key fingerprint =  90 5F 1D FD A6 7C 84 5E  A9 D3 90 16 B2 44 C4 F3