Allow super.method() calls in constructor

[arv]

The old code assumed that super always followed by a ( which is not
the case.

Also adds tests for invalid (at runtime) code with multiple super()
and super.m() / super.x / super.x = 1 before super().

Fixes #323

[arv]

@alangpierce PTAL

I added support and tests for multiple super() and super.a access before super() even though all those cases would lead to runtime errors. Is it worth dealing with this or can we just assume that we do not get this kind of invalid input code?

[codecov-io]

Codecov Report

Merging #324 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #324      +/-   ##
==========================================
+ Coverage   76.61%   76.62%   +<.01%     
==========================================
  Files          41       41              
  Lines        5456     5458       +2     
  Branches     1279     1279              
==========================================
+ Hits         4180     4182       +2     
  Misses        900      900              
  Partials      376      376

Impacted Files	Coverage Δ
src/util/getClassInfo.ts	85% <100%> (+0.25%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2cbe6fc...93c27c1. Read the comment docs.

[alangpierce]

Thank you!

Is it worth dealing with this or can we just assume that we do not get this kind of invalid input code?

I think it depends on how it affects the implementation. If invalid input makes the implementation harder, it's fine to call it out of scope and say the behavior is undefined. In this case, it's just the natural consequence of the straightforward approach that you took, so it's fine for Sucrase to handle it, and makes sense to add some tests as you did. Sucrase is intentionally not trying to check your program for errors (linters and typecheckers should be doing that), and a lot of parsing validation was intentionally removed for performance reasons, so there isn't any obligation to have Sucrase fail on the invalid cases that you tested for.