Exceptions in pure code

[ford-prefect]

Currently, if we have bad code that throws an exception despite being annotated as being pure, it seems to put Aff into a bad state where no more callbacks occur. I've written up an example at:

https://github.com/ford-prefect/purescript-sandbox

This is of course just bad code, but it might make sense to be defensive against this if there isn't a significant cost, as there are cases with a significant surface of JS interop, which can make for more common programming errors.

[ford-prefect]

@natefaubion you mentioned performance concerns w.r.t. adding try/catches to deal with such programming errors. Is this something you've quantified (or that we could try to)? Might help make a more informed decision about the cost.

[natefaubion]

There are some basic benchmarks in the test directory for pure binds. I would run it as is, apply a try/catch around the STEP_BIND case in the interpreter (where we apply the continuation), and then see what the difference is. If there is a significant performance degradation, then I would try lifting the try/catch out of the main loop like we do for runSync and benchmark that.

[ford-prefect]

Thanks @natefaubion -- I don't see a perf impact in the benchmark. pulp test takes ~4.8s to run with and without that change. I've made a PR for this (feel free to modify or ignore if there is a better way to do it, or I can do that too if you like).

[reactormonk]

This bug just cost me two days, so I wouldn't mind it being merged.

[ford-prefect]

Closed by #173