implement sync formdata parser

[KhafraDev]

Fixes #2890

Based on https://andreubotella.github.io/multipart-form-data/

import { Request } from 'undici'

const request = new Request('http://localhost', {
  method: 'POST',
  headers: {
    'Content-Type': 'multipart/form-data; boundary=----formdata-undici-0.6204674738279623'
  },
  body:
    '------formdata-undici-0.6204674738279623\r\n' +
    'Content-Disposition: form-data; name="fiŝo"\r\n' +
    '\r\n' +
    'value1\r\n' +
    '------formdata-undici-0.6204674738279623--'
})

const fd = await request.formData()
// > FormData { [Symbol(state)]: [ { name: 'fiŝo', value: 'value1' } ] }

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 83.51449% with 91 lines in your changes are missing coverage. Please review.

Project coverage is 93.62%. Comparing base (fc4e766) to head (2f69f5d).
Report is 10 commits behind head on main.

Files	Patch %	Lines
lib/web/fetch/formdata-parser.js	80.67%	91 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2911      +/-   ##
==========================================
- Coverage   93.86%   93.62%   -0.25%     
==========================================
  Files          85       86       +1     
  Lines       23418    23861     +443     
==========================================
+ Hits        21981    22339     +358     
- Misses       1437     1522      +85     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[KhafraDev]

Alternatively we could just take Deno's parser.

[mcollina]

lgtm

[mcollina]

Does this increase our wpt conpliance?

[KhafraDev]

Does this increase our wpt conpliance?

Yes, there are 4 (out of 14 tests wrt .formData) that now pass and there are some untested behaviors/behaviors we had to workaround that are now correct:

• We actually follow the steps in the spec.
• .formData() is now a function that returns a Promise instead of being async (I asked in the WHATWG matrix chat and this is the expected behavior)
• We were able to get rid of hacks in the code for a WPT: "Fetch aborted & connection closed when aborted after calling response.formData()" in fetch/api/abort/general.any.js

        // Wait a tick before checking if the request has been aborted.
        // Otherwise, a TypeError can be thrown when an AbortError should.
        await Promise.resolve()

[KhafraDev]

@Uzlopak ptal

[Uzlopak]

LGTM

[KhafraDev]

We can either land this as-is or I can spend the next few days/week (its midterm week for me so it might take longer) porting the busboy tests and fixing edge cases.

[Uzlopak]

how about a new branch and port the tests, and then when they are ported we merge it into main?

Can you port a test from busboy so that i can assess how much work it is?

[KhafraDev]

Adding the tests is very simple for the sole reason that they don't use any test runner and rely completely on node:assert. I ported the tests in my old formdata parser branch: https://github.com/KhafraDev/undici/tree/undici-formdata-parser/test/busboy

However, I don't think there's much benefit in porting something without also fixing bugs.

[KhafraDev]

It might be a little harder, given that we don't implement any limits, nor does the parser extend EventEmitter much less emit events, but it shouldn't take more than an hour?

[Uzlopak]

ah you mean the original busboy.

[KhafraDev]

I assumed that @fastify/busboy would have the same tests, guess I was wrong. Busboy's tests seem more relevant here since they test the actual parsing/output, where as fastify/busboy seems to test more so the actual api (which obviously we'd fail).

[Uzlopak]

actually busboy (original) mixed dicer and co into it. @fastify/busboy vendored them.

Anyhow i think that our @fastify/busboy tests were kind of more reasonable.

What about these tests?

https://github.com/fastify/busboy/blob/master/test/types-multipart.test.js

[KhafraDev]

No preference, but I don't like how expected parts removed the objects for arrays, now I have no idea what each index means.

[mcollina]

lgtm

[gurgunday]

looks good to me