bisect_ppx

bisect_ppx copied to clipboard

Taking a quick stab at this: https://github.com/aantron/bisect-starter-rescript/issues/3

when you try rescript 10 with bisect_ppx:

 $ rescript
Dependency on bisect_ppx
rescript: [1/4] src/common/bisect_common-Bisect.cmi
FAILED: src/common/bisect_common-Bisect.cmi
File "/home/.../bisect-starter-rescript/node_modules/bisect_ppx/src/common/bisect_common.mli", line 68, characters 42-53:
Error: Unbound type constructor out_channel

I piggy-backed a flake.nix for a reproducible development environment, mostly for personal use. We can cherry-pick it out of the PR if we don't want it.

Confused why out_channel can be unbounded, since it's clearly in stdlib which apparently is open'd at the top of every file. https://v2.ocaml.org/releases/4.13/api/Stdlib.html#TYPEout_channel

@christianoc told me on https://github.com/rescript-lang/rescript-compiler/issues/5596 that some things to do with Pervasives (which is the same thing as Stdlib) were removed from rescript 10.0, I also notice in the rescript-compiler CONTRIBUTING.md file it says that ocaml 4.14 is used, but I don't think that's related.

The issue: https://github.com/aantron/bisect_ppx/issues/406

Would this same PR work with the older version of ReScript? The new version should only affect compiling ReScript projects. What part is currently failing?

Would this same PR work with the older version of ReScript?

I added rescript-9 and rescript-10 as two separate processes to test. I think I expect constraints like "if you're on rescript 9 then your bisect_ppx version has to be <=2.7.1" and "if you're on rescript 10 then your bisect_ppx version has to be >=2.9.1", ultimately.

The new version should only affect compiling ReScript projects. What part is currently failing?

Based on discussion in https://github.com/aantron/bisect_ppx/pull/400, I think CI gradually started failing because the resolver started pulling down later versions of ppxlib. Note that in the last commit to master in march, many jobs were failing.

Any quick fix for this for ReScript?

Great! Do you mind testing v11 beta too? It should just work, but just in case.

Thanks for the heads up! I've fixed some incompatibilities in https://github.com/aantron/bisect_ppx/commit/9439ba71f1c138ca38229affa180ffef6aa3b978, but I'm currently left with:

λ make -C test/js full-test

[...]

> node ./lib/js/hello.js

internal/modules/cjs/loader.js:818
  throw err;
  ^

Error: Cannot find module '/home/antron/code/bisect/bisect_ppx/test/js/lib/js/hello.js'
    at Function.Module._resolveFilename (internal/modules/cjs/loader.js:815:15)
    at Function.Module._load (internal/modules/cjs/loader.js:667:27)
    at Function.executeUserEntryPoint [as runMain] (internal/modules/run_main.js:60:12)
    at internal/main/run_main_module.js:17:47 {
  code: 'MODULE_NOT_FOUND',
  requireStack: []
}

This is the former location of the output, hello.js, with ReScript <= 10. As far as I can tell, there is no output file at all now:

λ cd test/js && find lib                                                                                                                                          2

lib
lib/bs
lib/bs/.bsdeps
lib/bs/.bsbuild
lib/bs/.compiler.log
lib/bs/build.ninja
lib/bs/.ninja_log
lib/bs/install.ninja
lib/bs/.sourcedirs.json
lib/js

Could you suggest what changed in ReScript 11? I wasn't able to spot a relevant change in the changelogs for the ReScript 11 pre-release versions.

Would a compiler error (type checking or other) leave some trace here? One difference could be partial application is not possible by default, but there are quite a few changes, so hard to say without looking at 1 individual compilation.

There were several compiler errors that I fixed along the way. As I understood your message, the file should be at the same place, but there is probably some silent error along the way that prevents it from being generated. I'm going to open a separate issue to look into that later. Thank you!

The issue was that ReScript no longer does anything with .re files, so there were no compilation units detected and thus no output.