Joe Rickerby

I found that branch. The example script run looks like this: ```console $ python examples/context_api.py Start. scanning home dir... pyinstrument ........................................ . . Block at /Users/joerick/Projects/pyinstrument/examples/context_api.py:14 . . 0.548 main...

I'm a bit confused, do you think the CPython upgrade caused the performance regression? Or was it always there?

Another thing that has changed between these runs is the github actions runner image. But there could also be other factors, like the runner on GHA that actually executed the...

This is highly platform-dependent, I think. Running 1000 iterations on my Mac, I get ```shell $ python examples/sympy_perf.py This took 0:00:00.403314 $ pyinstrument examples/sympy_perf.py This took 0:00:00.845693 ``` Which admittedly...

Having this reproducer is really useful, thank you for that. I think there are some low-level things that can improve performance, once I know what platform you're on.

I'm assuming you're running this on Linux? I want to test a theory, could you run this code and tell me the output? ```c #include #include #include int main() {...

> I could perhaps try measuring performance of the `gettimeofday` in C to confirm it is indeed the culprit. Yeah, this would be the next step, I suppose. There are...

Here's a script: ```c #include #include #include typedef enum { CT_BASELINE, CT_GETTIMEOFDAY, CT_MONOTONIC, CT_MONOTONIC_RAW, CT_CLOCK, CT_REALTIME, CT_MONOTONIC_COARSE, } ClockType; static volatile double dummy_clock = 0.0; static double floatclock(ClockType clock_type) {...

I've tried the above script on a few machines. My conclusion is that it's not as simple as the existence of the VDSO. It always seems to be there in...

@J08nY I've started some work to improve this here #273 It's still very draft, but on the off-chance that you have time, I'm actually currently hitting a problem on Linux...