Closed mario-bucev closed 1 year ago
What's the rationale for this? I mean, the new behavior sounds simpler to understand, but was the past one causing actual problems? Any test cases?
funnyarith3
for example is a good test case (which sometimes fails depending on how symbols are sorted, affecting the order of the matching as well).
Before this PR, the strategy was to generate the maximum number of allowed matching (default of 16) and then perform pruning on it. Some picked matching may get pruned, leaving us with less than the limit which is not something we would like. This PR changes the behaviour to pick matching (passing the pruning test) as long as we don't go over the limit.