import pkg_resources can be very slow

[charleseidsness]

Running on a little imx6ul processor "import pkg_resources" can take 6 seconds or more. It looks like all it is doing is loading the version. I've patched it on my system to remove it, something you may want to consider.

[svinota]

Got it, thanks for the issue. I'll try to look what may be wrong there.

[svinota]

Related:

• __version__ support #576
• Include __version__ in the pyroute2 package #580
• pkg_resources breaks some pyroute2 use case #584

[svinota]

@ffourcot @celebdor

Looks like pkg_resources seriously affects startup time on platforms like ARM, and I can confirm that after tests.

I'm going to drop pkg_resources and provide __version__ by something like generating __version__.py file on make dist. Any objections?

[crosser]

Not an objection, just some considerations to keep in mind: semantics will be slightly different.

At runtime, version from pkg_resources will show the version of installed package, while version from a generated file will show the version of the package that is currently running. It may be different if you e.g. use an instance from the current directory. And in this case, generated file will show "more correct" information, but on the other hand, if you try to use an instance with not yet generated file, it may crash if import of this file is unconditional.

[ffourcot]

Not an objection at all, we don't use it and we internally avoid usage of pkg_resources for the same reasons.

@celebdor was author of the commit adding pkg_resources, it's probably a good idea to check with him if this semantic change is ok for him.

[celebdor]

Sounds good

[svinota]

So be it — pkg_resources import is dropped. The version file import is enveloped in try/except, should not fail on import errors.