quickcheck

When generating recursive structures, users will often use the size of the generator to bound the recursion depth. Dividing the size by the number of sub-structures appears to be a common practice. However, modifying the size of the generator passed to an implementation of would also affect the generation of sister structures if not restored after the generation of child nodes. Instead, users may choose to create a new generator instance with the target size. The new function makes use of seed from the original generator rather than entropy. In addition to bounding recursive structures, this also allows to re-create those structures based on some initial seed. This PR is based on #278 (in order to make use of ).

This type was originally motivated by observing that a class of properties will involve a pseudo-identity followed by a check for equivalence between its input and output. A generalization of would be properties which are defined by the equivalence check, only. See #280. For this class, we usually want to know _how_ those two values differ, rather than only that they do. Test-authors may thus write tests like the following in order to include those values in a failure report: This change introduces a convenience type which encapsulates the equivalence check as well as the error message generation. Using it, the above test could be written as:

Arbitrary s are generated based on arbitrary s by adding or subtracting from . If we happen to generate big s, this could cause an overflow. This change avoids this problem by resorting to ever smaller durations in case of an overflow. Fixes #321.