lib: implement interface converter and use it for AbortSignal validation

[jazelly]

This PR contains 2 commits.

The first one implements interface converter as per the spec.
The one after uses the implemented interface converter to properly validate AbortSiganls in AbortSignal.any

Fixes: #54962

[aduh95]

Shouldn't we update validateAbortSignal – or remove it if we no longer need it?

[jazelly]

I tried updating it, but it breaks some tests, as it is used variously in other modules. Some modules treat EventTarget with { aborted: boolean; } as a valid AbortSignal, and its test is written in that way.

[codecov]

Codecov Report

Attention: Patch coverage is 96.15385% with 1 line in your changes missing coverage. Please review.

Project coverage is 88.03%. Comparing base (53ede87) to head (82da5e9).
Report is 37 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/internal/webidl.js	94.11%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #54965      +/-   ##
==========================================
- Coverage   88.07%   88.03%   -0.05%     
==========================================
  Files         652      652              
  Lines      183542   183778     +236     
  Branches    35862    35855       -7     
==========================================
+ Hits       161653   161783     +130     
- Misses      15144    15249     +105     
- Partials     6745     6746       +1     

Files with missing lines	Coverage Δ
lib/internal/abort_controller.js	97.90% <100.00%> (-0.01%)	⬇️
lib/internal/webidl.js	93.89% <94.11%> (-0.69%)	⬇️

... and 46 files with indirect coverage changes

[jazelly]

This error message change is not semver-major nor minor, as there is only one place using this converter and it's not using the context, i.e. the error message is not even displayable in the userland.

[atlowChemi]

I am marking this as semver-major PRs that contain breaking changes and should be released in the next major version. since as far as I understand, this would break for any user-land implementation of AbortSignal (https://www.npmjs.com/package/abort-controller etc)

[jazelly]

I am marking this as semver-major

I agree this would be a semver-major. Does it mean a CITGM is required?

this would break for any user-land implementation of AbortSignal

Would it be better to remove this change? This is unrelated to the issue that the PR is fixing and it can be avoided.

[KhafraDev]

Is there any documentation for supporting third party abortcontrollers? If not I could see this landing as a bug fix.

[jazelly]

There is only one benchmark test coverring abort_controller and it's only covering the AbortController.abort, which is, IMO, not affected by this PR.

@anonrig In that case, would it be better if I raise another PR to add benchmark tests to cover AbortSignal.any and aborted, which are the ones affected by this PR, before proceeding this?

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/62549/ 💛

[RedYetiDev]

@nodejs/tsc per nodejs/Release#1034

[benjamingr]

Not brand checking AbortSignal was an intentional design choice in order to enable userland to migrate to using Node's AbortSignal.

We can choose to break that compatibility but it would have to be semver-major and require @nodejs/tsc's opinion since it's an ecosystem concern (this is likely to break userland code)

[benjamingr]

To be clear - since this breaks userland I want to see how the TSC and namely people who were involved in the original discussion ( @jasnell @mcollina @ronag ) feel about this change.

I'm personally +0 (I'd like it but not at the cost of extensive breakage)

[KhafraDev]

We made the same decision in undici to drop third party AbortController/AbortSignal support.

[benjamingr]

Thinking about this more userland AbortController libraries have tens of millions of weekly downloads so I'm -0.5 (not for this change but still not blocking)

We made the same decision in undici to drop third party AbortController/AbortSignal support.

When did it happen (assuming it affects Node's fetch)? Was there any breakage?

[atlowChemi]

I am -1 for this change.
The top 4 user-land implementations on npm have almost 60 million weekly downloads

[KhafraDev]

When did it happen (assuming it affects Node's fetch)? Was there any breakage?

v7, which is releasing soon-ish. Breakage is unimportant here - it was undocumented behavior that broke compatibility with other implementations and the spec. If we implement a spec, we need to accept its burden too. If there are/were performance concerns, those need to be fixed in node core, not handed off to user land to create crappy (not to be offensive) abortcontroller implementations to be used instead.