Re: sorting in C

[Andrew Cohen]

>>>>> "EZ" == Eli Zaretskii <eliz@gnu.org> writes:

[...]

    EZ> The 5-fold slowdown is a pain, IMO.  Can we do better in the
    EZ> worst case?
    >> 
    >> Maybe, but probably not too much. That is the point of these
    >> hybrids---they improve on some cases and do worse on others. Its
    >> a win only if the real-world data favors the improved case.
    >> 
    >> I played around a bit more and lowering the length threshold to
    >> 40 might be a better compromise. I am working on a more thorough
    >> set of tests first.

    EZ> OK, thanks.  Let's see the improved numbers, and can you also
    EZ> show absolute times?  If they are short enough, the slowdown
    EZ> might not be significant, since if the data structure is larger,
    EZ> the slow method will not be used, right?  Or did I
    EZ> misunderstand?

Yes, I will show absolute numbers in the future (for reference, it is
microseconds). And you did not misunderstand---the "slow" method (which
is much faster for sorted lists :)) isn't used for longer cases.

[...]

    EZ> But for showing the performance, the license is not important,
    EZ> is it?

Right, that was sort of a joke---just buying time until I can produce
some nice looking data.  I'm trying to automate their production a bit
more.

    >> Shouldn't be that difficult (the algorithm itself is
    >> well-documented) but I'm not sure how much time I have to finish
    >> it.

    EZ> From my POV, take all the time you need.  Emacs 29 is far from a
    EZ> release.

Great, thanks.

-- 
Andrew Cohen