fix(code_gen): incremental compilation

[Wodann]

Resolves all three side-effects of PR #97.

[codecov]

Codecov Report

Merging #101 into master will decrease coverage by 7.60%.
The diff coverage is 80.30%.

@@            Coverage Diff             @@
##           master     #101      +/-   ##
==========================================
- Coverage   76.84%   69.24%   -7.61%     
==========================================
  Files         132      133       +1     
  Lines       10643    10763     +120     
==========================================
- Hits         8179     7453     -726     
- Misses       2464     3310     +846     

Impacted Files	Coverage Δ
crates/mun_codegen/src/code_gen.rs	61.83% <0.00%> (-26.79%)	⬇️
crates/mun_codegen/src/ir.rs	55.76% <ø> (ø)
crates/mun_runtime/src/test.rs	19.09% <0.00%> (-76.99%)	⬇️
crates/mun_codegen/src/ir/abi_types.rs	86.58% <50.00%> (-13.42%)	⬇️
crates/mun_codegen/src/code_gen/symbols.rs	37.85% <62.96%> (-56.76%)	⬇️
crates/mun_codegen/src/ir/ty.rs	93.00% <75.00%> (-1.00%)	⬇️
crates/mun_codegen/src/ir/file.rs	79.31% <79.31%> (ø)
crates/mun_codegen/src/ir/body.rs	85.38% <86.20%> (-5.96%)	⬇️
crates/mun_codegen/src/ir/dispatch_table.rs	89.92% <93.33%> (-0.08%)	⬇️
crates/mun_codegen/src/ir/type_table.rs	93.71% <95.23%> (-3.91%)	⬇️
... and 38 more

---

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 0400406...8d5f1d5. Read the comment docs.

[baszalmstra]

Very nice! This also cleans up the generated IR by a lot.

• IR of the group_ir currently doesnt have a regression test
• Im missing documentation on what the group IR is.

Side note: did you try if incremental compilation works?

[Wodann]

• IR of the group_ir currently doesnt have a regression test

What do you think is the best solution:

• Output regression tests for the group_ir and module_ir?
• Output a single regression test for the linked (and potentially optimized) IR

Side note: did you try if incremental compilation works?

I don't currently have a solution for testing this. Do you have any ideas on how we could do it? This PR merely restores things to the status quo before PR #97.

[Wodann]

I've force pushed the requested additional comments. I also changed some of the file and struct names to better self-document, most notably module::ModuleIR -> file::FileIR. I felt this better reflects the changed structure.

I also changed the file::FileIR's return type to no longer contain FunctionValues, since those are only valid within the original LLVM Module, so it is unsafe to pass them around.

To facilitate this change, I also modified this and that.

[baszalmstra]

What do you think is the best solution:

* Output regression tests for the `group_ir` **and** `module_ir`?

* Output a single regression test for the linked (and potentially optimized) IR

Good question, I think both? But if I'd have to choose one, then probably the first.

I don't currently have a solution for testing this. Do you have any ideas on how we could do it? This PR merely restores things to the status quo before PR #97.

There is already one such test in https://github.com/mun-lang/mun/blob/master/crates/mun_hir/src/tests.rs . But I agree, maybe put that in a new PR. I do feel that it becomes more and more important as we go further. We want to make sure that we do not break incremental compilation. As we have seen that's very easy to do.