consistently use NoInverse instead of exceptions

[aplavin]

Currently, some inverse(non-invertible f) return NoInverse, some throw an exception. (the latter is code from my earlier PRs, sorry for that!)
Here, I make it consistent by always returning NoInverse.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (d397fef) 100.00% compared to head (7856164) 100.00%.

Additional details and impacted files

@@            Coverage Diff            @@
##            master       #40   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            5         5           
  Lines          121       121           
=========================================
  Hits           121       121           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[oschulz]

Hm, one the one hand, this makes things more consistent. But it does introduce type instabilities, and the functions in question may well be used in time-critical code.

I would suggest to keep throwing an exception here for practical reasons. True, we'll then get a failure on certain single points, e.g. at zero. But on the other hand, if an application encounters a zero in such places, it's very likely that something is wrong, and throwing an exception would make sense.

The idea behind NoInverse is to enable code to check if a function is invertible, and then take a different code paths if it's not. But I don't think this should come at the cost of type instability.

[aplavin]

Yeah, tend to agree with the type stability concerns! I didn't manage to write any small self-contained benchmark that demonstrates any performance difference between NoInverse vs exception, but agree that type stability should be kept whenever possible.