Any thoughts on verifying that a profiler is returning valid results? In
particular, I'm thinking of my stepping variant on Ian's profiler. When I
last tried it, it got numbers that looked somewhat reasonable, but, it would
probably be easy to get confused about the stack frames and produce
nonsense. Any clever tests?
> Any thoughts on verifying that a profiler is returning valid results? In
> particular, I'm thinking of my stepping variant on Ian's profiler
I can't suggest Unit tests, but for normal sanity checking it should be
sufficient to cross-check against Ian's original profiler shouldn't it ? If
your new tool has significantly better temporal resolution than Ian's, then
compensate for that by running Ian's profiler on a correspondingly longer
run of the target code.
In the most intensive use I've made of the existing profiler I was using
runs in the 10s of minutes -- which was long enough to "resolve" down to the
time taken *sending* messages (as opposed to execting the bodies of the
methods). Creates masses of data, though...
Locked
