All committers have signed the CLA.

Coverage decreased (-0.4%) to 98.303% when pulling d4c92e5 on kalenikaliaksandr:kalenik-polyfill-remove-ring-tracing into ac534dc on uber:master.

Thank you for your submission. In the future, it might make sense to discuss in our Slack channel or in a GitHub issue before putting together a PR.

The reason we include the step of tracing the outlines is that the cells returned by polyfill may not be contiguous. The current polyfill algorithm only takes cells whose centers fall within the polygon, so there are many cases in which a simple flood fill won't work - imagine for example a "barbell" shape with two areas filled with cells, separated by a arbitrarily long "bar" with no cells. Depending on the scale of the shapes and the resolution of the cells, there may be many cases like this in a given dataset. Outline tracing was the best option we could come up with that handles all of these cases. We'd welcome other more efficient approaches here.

It looks like we don't have good test cases for these issues, however - I would have hoped the tests would fail in this PR, but they don't, which suggests a gap in our unit tests.