Performance of DeepZipper Selection Needs Significant Improvement about anti-xml

[ncreep]

Do we have some predefined quick performance tests?
Something that runs within a minute, just to get a hint to whether an optimization has the desired effect.

[djspiewak]

The best performance test to run would be test-perf shallowSelectSmall (from the SBT shell). Unfortunately, this test takes a really long time right now because a single shallow select operation on a 7 MB tree requires two and a half minutes (and the test is run about a dozen times). So, that's a little long. However, if the performance were better, the test would complete quite quickly (on the master branch, I believe it only takes a few seconds).

[ncreep]

Blind optimizations FTW!

[djspiewak]

I'm terribly sorry! :-) It would be pretty easy to just whip up a new performance test with a smaller tree, something that runs a little more quickly. Alternatively, the performance is so dramatically slow right now that you can probably just check timing in the REPL and see differences.

[josharnold52]

Of course, there's always the tried and true method of "break into the debugger and see what's on the call stack". When the performance is really out of whack, chances are you will break into the bottleneck :-)

[djspiewak]

With the very latest changes from ncreep, the performance is good enough that I wouldn't consider it a blocker any longer. It could still stand some improvement, but I would feel comfortable cutting a release without further improvement.

[djspiewak]

With the latest changes from ncreep, I think we can consider this closed!