Use NoTangent() instead of nothing to designate non-differentiability

[mzgubic]

Needed so that JuliaDiff/FiniteDifferences.jl#189 does not break ChainRules tests

[AlexRobson]

This is changing the length of the return - i.e. from (nothing, nothing...) to a single NoTangent(). Is this intended?

I'm surprised no tests had to change to handle this.

I think that this means that here given that the output of fd_cotangents) will only be length one, the zip will only loop over the first argument.

[mzgubic]

well caught, that's a mistake

[codecov-commenter]

Codecov Report

Merging #203 (be9dbdb) into master (e0f1171) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #203   +/-   ##
=======================================
  Coverage   90.94%   90.94%           
=======================================
  Files          11       11           
  Lines         287      287           
=======================================
  Hits          261      261           
  Misses         26       26           

Impacted Files	Coverage Δ
src/finite_difference_calls.jl	96.96% <100.00%> (ø)
src/testers.jl	92.85% <100.00%> (ø)

---

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 e0f1171...be9dbdb. Read the comment docs.

[willtebbutt]

I have no objection to this going in, but like @AlexRobson I'm a bit surprised that there's no changes needed to the tests. Do we just not have great coverage in this package?

[mzgubic]

I think the reason is that this is an implementation detail. It is not visible anywhere outside of the package. Annoyingly it didn't catch the length issue.