akiellor / bananas Goto Github PK

Traversal that may work well for infinite state machines. Biases towards transitions that create the biggest variation to the set of visited states.

• hash states from leaves to root, keeping track of all hashes
• keep a set of all these hashes for a given test suite
• select transitions based on which transition will add the most unique hashes to the visited set (maybe prefer hashes closest to the root, as they are in theory more impactful?)

Current tests a labeled based on their index, this is hardly useful.

Maybe something using the steps in the scenario could be better.

The current implementation of requirements is either:

• an explicit value {requires: {foo: 1}}
• or a predicate {requires: function(model) { return false; }}

We should have some higher abstractions for common predicates:
i.e. {requires: property('foo')}

Requirements predicates could do with some love.

At the moment we generate extra scenarios:
Consider transitions A, B, C:
C requires A

The current algorithm produces two scenarios:
ABC
BAC

We should maybe only have one:
BAC

We could describe the current algorithm as modeling all permutations satisfying the constraints, which we thought could be useful. In practice, this creates huge numbers of scenarios in even very simple cases.

rove and jaunt suggest a degree of aimlessness. What about the game Hopscotch?

A good model(TM), will pass when expected to pass and fail when expected to fail.