Speedup uses of object.__setattr__

[davfsa]

Summary

I was playing around seeing if I could speedup the creation of frozen slot classes for a project of mine since we instantiate a lot of attr classes. Ideally they would be frozen, but we didnt go with it because of the speed implications it had.

To try and speed it up I tried different things:

• Make use of objects.__setattr__ during the whole of the init and then place the frozen one at the end
  • The idea behind this is to use the default __setattr__ during the whole __init__ function and then setting it to the frozen one after, which I thought would bring speedups as well as improve the UX, since they can could set attributes normally in the init. This unfortunately didnt work nicely with slots and slowed it down even further when attempting a workaround
• Playing around a bit with python internals and trying to figure out the ideal way to implement this
  • This is the change this PR adds does

Overall I wasn't really successful, but I did find this small performance gain of "1.1x", which isnt much, but I think it might be worth having.

Bellow you can see the pypref benchmarks on my laptop (to simulate the ones that were in the documentation)

Before (attrs == 21.4.0)

$ pyperf timeit --rigorous -s "import attr; C = attr.make_class('C', ['x', 'y', 'z'], slots=True, frozen=True)" "C(1, 2, 3)"
.........................................
Mean +- std dev: 499 ns +- 26 ns

After

$ pyperf timeit --rigorous -s "import attr; C = attr.make_class('C', ['x', 'y', 'z'], slots=True, frozen=True)" "C(1, 2, 3)"
.........................................
Mean +- std dev: 450 ns +- 17 ns

Pull Request Check List

• Added tests for changed code.
  Our CI fails if coverage is not 100%.
• New features have been added to our Hypothesis testing strategy.
• Changes or additions to public APIs are reflected in our type stubs (files ending in .pyi).
  • ...and used in the stub test file tests/typing_example.py.
  • If they've been added to attr/__init__.pyi, they've also been re-imported in attrs/__init__.pyi.
• Updated documentation for changed code.
  • New functions/classes have to be added to docs/api.rst by hand.
  • Changes to the signature of @attr.s() have to be added by hand too.
  • Changed/added classes/methods/functions have appropriate versionadded, versionchanged, or deprecated directives.
    Find the appropriate next version in our __init__.py file.
• Documentation in .rst files is written using semantic newlines.
• Changes (and possible deprecations) have news fragments in changelog.d.

[hynek]

Hmmm 10% more performance is nice and he solution is simpler too, which is nice!

Calling @Tinche who seems to be the daddy of the old code – any edge cases we're missing here?

[Tinche]

This essentially saves us an attribute lookup at the cost of an additional argument in a couple function calls?

If the tests pass, LGTM.

[hynek]

@davfsa can you rebase please? you didn't give us write permissions to your branch

[hynek]

Thanks!

[davfsa]

Rebased and turned on edits in case anything else is needed :)