function: don't pass unexpected marked values to TypeFunc

[jbardin]

ReturnTypeForValues was only checking an unmarking the top level value
for TypeFuncs that did not expect marked values. Use ContainsMarked and
UnmarkDeep to prepare the arguments.

[codecov]

Codecov Report

Merging #92 (904c1ed) into main (080b168) will increase coverage by 0.09%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #92      +/-   ##
==========================================
+ Coverage   70.85%   70.94%   +0.09%     
==========================================
  Files          79       79              
  Lines        6636     6636              
==========================================
+ Hits         4702     4708       +6     
+ Misses       1490     1486       -4     
+ Partials      444      442       -2     

Impacted Files	Coverage Δ
cty/function/function.go	48.24% <100.00%> (+10.52%)	⬆️
cty/convert/compare_types.go	92.59% <0.00%> (-7.41%)	⬇️

---

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 080b168...904c1ed. Read the comment docs.

[pselle]

I think this is good as far as safety I can think through on acceptable disclosure w.r.t. the types 🤔 I am surprised we haven't run into this as yet, so just so I can completely grok, do you have an example of where this is useful/what this is addressing?

[jbardin]

Hi @pselle , this particular case takes a combination of constructing values which contain marked values, and passing them to a function which needs to calculate the return type based on those values. The corresponding Terraform issue is: hashicorp/terraform#28049.

[apparentlymart]

This makes sense to me, so thanks! I think in retrospect this was a missing part of #72 that I neglected to catch during review.

The similar code for this in Call was doing some extra work to try to optimize for cases when work isn't required, which I think this new change defeats in some ways, but I'm going to merge this now because getting the correct behavior is more important than optimizing for performance, and we can always profile and optimize this further in later commits without changing the observable behavior.