Re: efficient version of 'sort | uniq -c | sort -n'?

[Matthew Woehlke]

Paul Eggert wrote:
> Matthew Woehlke writes:
> > Is there an efficient implementation of 'sort | uniq -c | sort -n'? I
> > have a 4 GB core file I want to run 'strings' on, and the above is
> > really slow.
>
> See Jon Bently, Don Knuth, Doug McIlroy, "Programming pearls: a
> literate program", CACM 29, 6 (June 1986), 471-483
> <http://doi.acm.org/10.1145/5948.315654>.  Source code is included.  I
> think Knuth's solution will run rings around all the solutions
> proposed so far, if you tune it right (see below).
>
> > Is there a way already in coreutils to do this? If not, would there
> > be any interest in adding such a method?
>
> I dunno, it sounds pretty specialized.  Though there may be some
> interest in a combination "sort | uniq -c", I wouldn't think there'd
> be any interest in combining all three.

I would think 'sort --unique-with-count' would be the most valuable, 
although I'm not sure why you /wouldn't/ want the result sorted by 
number of occurrences (at least, that's the most obvious use case for 
the combination I can think of).

I'll put looking at the above-mentioned algorithm on my to-do list, IOW 
don't expect anything soon, if ever. :-) But thanks for answering the 
second question!

--
Matthew
When in doubt, duct tape!