This release substantially improves our internal distribution for
generating integers. This release has the most visible effect on
"integers()", but may incidentally improve other strategies which draw
integers internally.
Our integers distribution had two problems. First, it had jagged
discontinuities at certain values where we switched sampling
approaches. Second, it used a different distribution for bounded and
unbounded ranges, which resulted in "st.integers()" and
"st.integers(-264, 264)" producing very different distributions
despite being semantically similar.
We now use a smooth distribution for both "st.integers()" and
"st.integers(a, b)", which fixes both of these issues. This should
substantially improve our testing power in certain cases.
The only way this release should be user-visible is that it finds more
bugs! If this release is user-visible in other ways - for example,
because it is slower, or produces a worse distribution in some cases -
please open an issue.
The canonical version of these notes (with links) is on readthedocs.