[blyxxyz]

• Add PySAT support. This is an optional dependency. If installed it's used by default for satisfiability and solving, except for very small sentences. The overhead is low and it seems reliable.
  It's not used by default for model enumeration, because PySAT's implementation is naive and ends up slower than the native solver in many cases.
  Resolves Optionally call pysat if it is installed #10.
• Add a general configuration system. It uses a few more weird Python tricks than ideal, but I think it ended up pretty robust and pleasant to work with.
• Implement .solve(). Resolves Method that returns an arbitrary model (.solve()) #15.
• Make all implementations of model enumeration lazy. The one for DNNF ended up nicer than before, but the one for d-DNNF uses some more awful Python tricks that I'm not entirely happy with.
• Remove the Builder class. It didn't actually solve the problem it was supposed to.
• Various other tweaks and fixes.

[haz]

Seems like the tests are failing for 3.4 / 3.4, ya?

Seems pretty good to me. Magic # of 5 on the sentence size for auto choosing the SAT backend? ;). What were the performance details you found?

[blyxxyz]

Magic # of 5 on the sentence size for auto choosing the SAT backend? ;). What were the performance details you found?

I played around with generated CNF sentences, and this was the relatively worst performance for the native backend on small sentences I found:

(number of clauses on the x axis, time in seconds on the y axis)

I could move the intersection to the right with different parameters, but not further left.

[blyxxyz]

When/right before releasing 0.3.0 the README should be updated to recommend pip install nnf[pysat], to pull in PySAT.

If we do it too soon then the instructions people find on Github don't work yet and if we do it after then the outdated instructions will end up on PyPI.

[haz]

I think I'm about ready to pull the trigger on 0.3.0. Any last-minute changes you wanted to throw in?

[blyxxyz]

No, looks good to me.

[haz]

Hope it was done right, but we should be all set.