Add global variable mapping and preserve primitives

[JSAbrahams]

Relevant issues

• Resolves #404
• Resolves #426
• Might have resolved #400 We use output instead of ok to check the output of Command::main_binary()
• Made note of #431
• Made note of #432

Summary

Quite a big PR as we had to change some fundamental behaviour. More fundamentally how we deal with constraint generation. Diff slightly larger due to some renaming of variables or consistent change in behaviour, namely how we reference var_mapping.

Multiple constraint sets

We now change the behaviour of the ConstrBuilder such that we always add constraints to multiple sets in parallel. This allows us to generate constraints for all possible execution paths (for instance, after an if or match).

We can then either:

• Exit a set locally, thus preventing the constraints here from interacting with constraints further down the line.
• Not exit the set, thus increasing for now the amount of constraints being added to. One can use bookmarks (usize) so one exits to the correct level at a later stage without knowing how many sets are being added to in parallel.

Pushing types

Get rid of unnecessary unions by updating logic of how push_ty operates. Whether one has a union of types is determined there and whether any of the Names is interchangeable.

• Streamline further some Expect logic which was redundant, which got rid of primitives by directly substituting them. Instead, all constraints should be generated in the generate stage, including those for primitives. This further simplifies the logic of primitives and makes them more like any other class, which they essentially are (like any good OOP-like language).
• Fix is_superset_of logic in Name. Logic before was erroneous, but not caught due to faulty test.

Tests

Happy

• Modify arithmetic test to get rid of unnecessary union.
• Modify factorial test to get rid of unnecessary union of input function output. Sad
• Is superset of Name where other is interchangeable but none of self is super of other
• Like previous, but with same class but other nullable (so self is not super)

[codecov[bot]]

Codecov Report

Merging #416 (14b36a8) into develop (9723d5f) will decrease coverage by 0.10%. The diff coverage is 95.39%.

:exclamation: Current head 14b36a8 differs from pull request most recent head f497055. Consider uploading reports for the commit f497055 to get more accurate results

@@             Coverage Diff             @@
##           develop     #416      +/-   ##
===========================================
- Coverage    88.43%   88.33%   -0.11%     
===========================================
  Files          110      110              
  Lines        12040    12000      -40     
===========================================
- Hits         10648    10600      -48     
- Misses        1392     1400       +8     

Impacted Files	Coverage Δ
src/check/constrain/unify/mod.rs	75.00% <20.00%> (+12.93%)	:arrow_up:
src/check/constrain/constraint/iterator.rs	86.20% <66.66%> (-0.46%)	:arrow_down:
src/lib.rs	82.89% <75.00%> (+1.19%)	:arrow_up:
src/check/result.rs	82.22% <83.33%> (ø)
src/check/constrain/constraint/builder.rs	86.27% <85.71%> (-2.19%)	:arrow_down:
src/check/constrain/generate/env.rs	94.44% <96.15%> (-0.56%)	:arrow_down:
src/check/constrain/constraint/expected.rs	88.11% <100.00%> (-1.61%)	:arrow_down:
src/check/constrain/generate/call.rs	88.20% <100.00%> (+0.10%)	:arrow_up:
src/check/constrain/generate/class.rs	82.14% <100.00%> (-6.99%)	:arrow_down:
src/check/constrain/generate/collection.rs	96.77% <100.00%> (+0.03%)	:arrow_up:
... and 23 more

[JSAbrahams]

Before merging need to document why this works, because the underlying mechanics are starting to become a bit unclear.

[JSAbrahams]

Something went wrong after rebase on develop. So turns out that what we were doing here was actually quite dangerous, could've known.

[JSAbrahams]

It seems that we are now also accidentally allowing any argument, regardless of type, to be passed to functions. Possibly something wrong with how the any boolean in Name affects is_superset_of. See #426

[JSAbrahams]

Must investigate significant drop in coverage first.

One thing which I'm not sure about is what if we shadow a variable with another one of a different type. I say this because I'm not quite sure how the environment and global mapping interact. But I can't think of a definitive way to test this, but a good place to start would be to shadow variables inside a function:

• Shadow function argument
• Shadow class field
• Shadow outer field

In each test, make sure that we call methods only of that specific class.

[JSAbrahams]

Streamlined quite some logic. After this, implementing dictionary should hopefully be more straightforward.