tests: organize lang spec and generic related tests

[saem]

Summary

• what changed and how?
• why are we changing it?

Details

• info that couldn't fit into the summary
• relevant details; tricky parts
• anything else

---

TODO

• shift focus towards generics and keep going until tired

Notes for Reviewers

• test re-org is messy stuff, no real narrative

[saem]

I'm re-running the tests, I can't replicate the failure locally.

[zerbina]

Huh, that's strange. tjsonutils.nim consistently fails for the Windows and MacOS CIs, but not on the Linux CI. The only reason for this I can think of is that moving around, removing, or adding new tests changed the make-up of the megatest in a way that caused a latent bug to surface.

[saem]

Huh, that's strange. tjsonutils.nim consistently fails for the Windows and MacOS CIs, but not on the Linux CI. The only reason for this I can think of is that moving around, removing, or adding new tests changed the make-up of the megatest in a way that caused a latent bug surfacing.

Yup, that's what I'm thinking too. Locally, I get js failures for tjsonutils on line 119. My best guess is it has something to do with the define nimPreviewJsonutilsHoleyEnum. It's specified in tests/nim.cfg and jsonutils.toJson conditionally compiles on it within generic code. I wonder if it's being instantiated one way and then differently at some other time.

I'm not quite sure why fromJson is failing in CI on line 286, the only thing they have in common is isNamedTuple but it's been around for a while.

Hmm, given the common denominator is isNamedTuple right now, then I'll start by investigating that I suppose.

[saem]

Just a small update:

In terms of test ordering and population, the only difference is linux includes (tests/misc/tsizeof4.nim), while macos m1 does not -- it's disabled on arm64.

So I get local failures if I run tjsonutils for the js target, but no issues when doing a full run across all targets with megatest. I'm going to see if I can sort out the local failure issue and why it's not failing when run in a megatest context. Hopefully that'll help shed some light on things.

[saem]

i haven't quite figured out how, but there seems to be a js codegen issue for tjsonutils:119. the passed in parameter seems to have field positions swapped so the values are assigned incorrectly.

Here is a snippet from the generated js:

var Data5 = {jsonNodeMode: 1, enumMode: 0};
var Data6 = {jsonNodeMode: 2, enumMode: 0};
var Data7 = {enumMode: 2, jsonNodeMode: 0};

This corresponds to initializations of ToJsonOptions, the third item has the fields in the correct order. The first two the fields are in the incorrect order (not really a big deal), but the values are assigned incorrectly, jsonNodeMode should be 0 in both cases, while enumMode should have 1|2 assigned.

Just gotta track down where this is happening in jsgen.

Update: after some further debugging the issue might be earlier, in mir or transf.

[saem]

With the above issue fixed in the following PR, thanks to @zerbina, going to see if that also resolves the issue encountered in CI.