[Lib2geom-devel] performance tests

[Johan E. <jbc...@sw...>]

Hi all,
   I added a "perf" build target and two simple performance tests to trunk.
I highly recommend using Ninja for building instead of Make. Because we 
have so many subtargets, Ninja is much much faster when recompiling a 
piece of code. To generate Ninja build files: "cmake -G Ninja .....".
It is pretty fun to write some test code, tweak it, and meanwhile run 
"ninja perf" to see if your changes have any effect (as you all know, 
the result is generally the opposite of what you'd expect).

I'd like some input in how to improve the performance test framework. I 
believe we want separate .cpp files for each performance test. Each 
performance test should take a couple of 100 ms. And then each test 
should be done a couple of times (three times currently) to get some 
averaging going.
To reduce the amount of duplicated code, how about I make a 
PerformanceTest base class, and let each performance test derive from 
that? The PerformanceTest class would have boilerplate and a
       virtual clock_t run_test()
  method that returns the clockticks for that test. Then all outputting 
is done in the base class, and the main of each test would look like:
     int main() {
         DerivedTest test;
         test.run();
     }
We could even add some standard paths and other "inputs" inside the 
baseclass, ready to be used by performance test code.

I need your input on this. Yes, I am duplicating functionality of 
existing test frameworks. But I'd like to keep it simple and 
understandable. It doesn't have to be a very powerful system (imho).

Cheers,
   Johan