I took out the check for darcs-2 executables and managed to run
darcs-benchmark just fine with a darcs-1 executable.  I ran the
benchmark on a large darcs-1 format repo.  It took four days.

I will update my darcs-benchmark to the new version that does not
do repetitions when the command takes more than 10 seconds, then
run on a hashed and a darcs-2 version of the same repository.  If
those are fast enough, I may re-run the benchmark on the darcs-1
format repository.

The results are interesting.  The release candidate does better
than darcs-2.3.1 with plain 'whatsnew'.  Every other benchmark
seems to be about the same or much worse, including 'whatsnew -l'.
None of the darcs-2 benchmarks are faster than darcs-1 except for
'pull'.  It will be interesting to compare these benchmarks
against the hashed and darcs-2 format benchmarks (might take a
few days to run).  

cap-1
-----

==================  =============  ===============  ================
                      darcs-1.0.9      darcs-2.3.1      darcs-2.3.99
==================  =============  ===============  ================
        get (full)   51.2s, 39.0M    1975.3s, 0.0M     1857.9s, 0.0M
   get (lazy, x10)           -, -   22097.6s, 0.0M    22197.2s, 0.0M
          pull 100  116.8s, 49.0M     85.5s, 73.0M    516.0s, 672.0M
            wh x50    60.3s, 0.0M   13297.1s, 0.0M     9872.3s, 0.0M
        wh mod x50    60.4s, 6.0M  13316.4s, 63.0M   9861.1s, 496.0M
         wh -l x20   158.4s, 0.0M     112.7s, 0.0M    15114.0s, 0.0M
    record mod x10   26.7s, 16.0M     39.3s, 62.0M   2258.5s, 515.0M
    revert mod x50   117.4s, 7.0M    239.5s, 28.0M  19743.4s, 594.0M
(un)revert mod x10    49.2s, 7.0M    107.7s, 76.0M   8815.1s, 960.0M
             check  189.0s, 42.0M  1123.8s, 589.0M  1093.6s, 1394.0M
            repair  166.7s, 42.0M  1121.4s, 589.0M  1250.1s, 1394.0M
          annotate           -, -             -, -              -, -
         pull 1000  301.9s, 99.0M   169.4s, 107.0M    972.3s, 680.0M
==================  =============  ===============  ================

The 'annotate' benchmark failed with the error:

  error: user error (darcs failed with error code 2
  saying: Stack space overflow: current size 8388608 bytes.
  Use `+RTS -Ksize' to increase it.)

I am surprised that the 'whatsnew -l' is listed as using 0.0M of
memory.  I watched the memory indicator in my panel go from about
20% to more than 80% for each iteration of that benchmark.

I tried to add a benchmark for 'darcs add', but so far have been
unsuccessful.  I can send my attempt if someone wants to (gently)
mentor me in haskell.

-kolibrie

Attachment: signature.asc
Description: Digital signature

_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users

Reply via email to