Environment variables are not inherited about pyperf HOT 8 CLOSED

from pyperf.

It is a deliberate choice to make results more reproducible between two computers.

How does blocking environment variables achieve that?

The only situation where such a blanket block of environment variables makes sense is for security reasons and that doesn't apply here. Another reason is changing/removing specific known-broken environment variables, but I don't think that this applies here either. In any case, the latter calls for a blacklist, not a whitelist.

I think that perf should run commands exactly the same way as in the usual Python environment. I don't see any reason to change environment variables at all, so whatever options are implemented, keeping the environment as-is should be the default.

from pyperf.

How does blocking environment variables achieve that?

For example, if you install a Python module on your machine and it's in the PYTHONPATH, the module can slowdown the benchmark, whereas the same benchmark on a similar machine without this module may be faster. Or it can be the opposite.

It's common that environment variables control optional features.

from pyperf.

@jdemeyer it's unclear to me if you still need to inherit the full environment or not. In case of doubt, I close the issue. If you want to move this forward, you can try to implement a change to add an option to implement that.

from pyperf.

I still think that it's a bug in perf that environment variables are blocked. But to be honest, I don't care enough to make a PR myself.

from pyperf.

Just one example. Depending on PYTHONPATH, you may or may not get additional .pth files which can change the behavior and impact performances.

from pyperf.

Depending on PYTHONPATH, you may or may not get additional .pth files which can change the behavior and impact performances.

Right, but that's a feature! I don't want code to suddenly behave differently if it's being profiled by perf. Instead, perf should just run the code that would normally be run by Python. If I set a PYTHONPATH, it means that I want to use it.

from pyperf.

(I reopen the discussion.)

perf helps you to identify what's unusal in your benchmark configuration to get reproducible results.

Removing environment variables idea comes from the old Python Grand Unified Benchmark Suite which late became http://pyperformance.readthedocs.io/ : new implementation (written by me) based on perf.

from pyperf.