Use __slots__ in Partition

[pjrule]

(I'm not sure if this is worth it yet, but figured I'd throw up a PR and we can do some benchmarks.) I was just reminded that using __slots__ is known to improve memory usage (and sometimes performance) of Python classes (see here), at least when the classes are small. I suspect our memory usage is dominated by large assignment dictionaries, but maybe this is a step in the right direction. __slots__ also protects against abuse of the Partition class by preventing the use of additional fields, though this may break some people's code.

[InnovativeInventor]

@pjrule I'm going to give this a rebase and run some performance benchmarks. If we see a sizable improvement in replay speeds, I think it'd be worth merging in.

[InnovativeInventor]

With the benchmarking utils I wrote a while back to look for performance regressions, I can confirm that this simple change results in a massive speedup in replay times. I estimate this gives a ~2-3x speedup in replay performance.

Replay performance without __slots__:

10 loops, best of 5: 1.42 sec per loop

Same chain replay performance with __slots__:

10 loops, best of 5: 610 msec per loop

We should totally merge this in!

[InnovativeInventor]

Excellent speedup in replay times!

[InnovativeInventor]

With the benchmarking utils I wrote a while back to look for performance regressions, I can confirm that this simple change results in a massive speedup in replay times. I estimate this gives a ~2-3x speedup in replay performance.

So . . . turns out this is not true. I was benchmarking off an old commit in main (before #367 was merged in), so the amended benchmarks on 10k replays are as follows:

• 194 it/s (Parker's slots)
• 189 it/s (main branch)

It's still worth it, though (every little bit counts).

[InnovativeInventor]

Note that once we merge this in, our next release should be a major one (we should follow semver).