bytecode: Implement addition and subtraction

[Rossiar]

These simple operations for integer mathematics will
be expanded upon in the next change, which will add
the multiplication, division and modulo operations.
This change also standardises the compiler and vm test
layouts to increase their readability.

Co-authored-by: joshcarp [email protected]
Co-authored-by: pgmitche [email protected]

[github-actions]

firebase-deployment for ece2f03

• apex: https://evy-lang-stage--274-xegucm6m.web.app
• discord: https://evy-lang-stage-discord--274-7ab87065.web.app
• docs: https://evy-lang-stage-docs--274-38hytquy.web.app
• learn: https://evy-lang-stage-learn--274-molmdx00.web.app
• play: https://evy-lang-stage-play--274-9k39k8r5.web.app

[juliaogris]

Looks good to me just a couple of comments, mostly food for thought.
I'll wait for @camh- if he's got anything to add again.

[camh-]

mostly comments around tests - I'm sorry I didn't get around to reviewing the tests last time, some of this I could have picked up then.

[Rossiar]

fixed most issues, some outstanding conversations to follow up on.

What do we think of unifying vm_test and compiler_test? The inputs and test cases are the same and vm_test runs the compiler - would this simplify or complicate things?

[camh-]

What do we think of unifying vm_test and compiler_test? The inputs and test cases are the same and vm_test runs the compiler - would this simplify or complicate things?

Agree 100%. I was going to comment on this but felt I had said enough already. I'm glad you brought it up.

This is what we've done elsewhere - more of end-to-end testing that unit testing. As you point out, the test cases are the same and they will surface the same errors. If you plan to keep the compiler and vm in the same package, this is easy since it is logically valid to have a single test that covers both. I originally had a mental model of package bytecode containing the representation of the bytecode, bytecode/compiler that generates a stream of bytecode and bytecode/vm that executes a stream of bytecode. With that split it becomes a little weird that tests in the bytecode/vm pacakge would catch errors in the compiler. I'm not sure if you're considering a structure like this at all, but I guess we'll deal with that when the time comes if the time comes.

[Rossiar]

I have:

• merged the tests into vm_test
• set operand widths to nil instead of empty slice
• refactored the vm to use panics instead of errors and added ErrInternal
• changed the binary expression handling to avoid helper methods
• changed numVal to be a float64 (in doing so I had to remove Set because you can't have a pointer receiver on a float64, I think this is ok because you can now just write a = numVal(12) to reset the value)

[juliaogris]

looks pretty solid to me. Just a little bit around error grouping
sentinel errors.
All of this is foundational work for the bytecode package and this back and forth will hopefully soon drop away.

[juliaogris]

ok, i'm happy with this - Thank You - there is only one last unresolved comment:
#274 (comment)

[juliaogris]

ok, this is all good from my end now, but maybe @camh- has still something else to add? I'll leave it up to him.
Also, if you can please squash all fix-ups.