esm: use undici/fetch data url parser

[KhafraDev]

Using the fetch parser, rather than a regex, should fix most of these edge cases.

It would probably be better to export the parser directly so we can use it for the sync parsing too. Just wanted to gauge how correct or welcome this change would be.

fetch should also be used for blob: urls, whenever support is added for them, as it handles edge cases regarding those as well. If someone ever brings back http/https imports, fetch should probably be used there as well. :)

Fixes #53775
Fixes #42890
Closes #51324

[nodejs-github-bot]

Review requested:

• @nodejs/loaders

[KhafraDev]

data:invalid,null is a valid data url afaik

[aduh95]

One of the problem of this approach is that Undici is not at all written in a way to be robust against prototype mutation, which seems to be a deal breaker when it comes to loading modules.

[KhafraDev]

that's true, maybe I could rip the data url parser from undici and implement it here? Could be useful for the node:util or something as well.

[KhafraDev]

I took undici's data url parser and added primordials, it's entirely possible I missed things or misused them. It's also likely possible to replace the mimetype parsing with the built-in one.

[mcollina]

@KhafraDev can you also port the tests for the tests for the data url parser?

[ljharb]

feel free to ignore these, but they might make things simpler and faster

[KhafraDev]

I took the WPTs for fetching data urls, if there's a more complete dataset, or if there's one specifically for imports, let me know.

[mcollina]

lgtm

[codecov]

Codecov Report

Attention: Patch coverage is 91.16022% with 32 lines in your changes missing coverage. Please review.

Project coverage is 87.61%. Comparing base (5949e16) to head (86d6382).
Report is 344 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/internal/data_url.js	91.76%	26 Missing and 3 partials ⚠️
lib/internal/modules/esm/load.js	70.00%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##             main   #54748    +/-   ##
========================================
  Coverage   87.60%   87.61%            
========================================
  Files         650      651     +1     
  Lines      182829   183289   +460     
  Branches    35379    35434    +55     
========================================
+ Hits       160173   160591   +418     
- Misses      15928    15950    +22     
- Partials     6728     6748    +20     

Files with missing lines	Coverage Δ
lib/internal/modules/esm/load.js	92.54% <70.00%> (-0.42%)	⬇️
lib/internal/data_url.js	91.76% <91.76%> (ø)

... and 34 files with indirect coverage changes

[jasnell]

Doesn't need to be in this PR, but having a benchmark for parsing these would be good. Would like to revisit to see if there's a more efficient approach than using the regex.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/62084/ 🟡

[aduh95]

Landed in 6c85d40

[targos]

I tried to include this in v20.x, but it broke HTTP imports (which were removed from main/v22.x)

See https://github.com/nodejs/node/actions/runs/11102823645/job/30843465228?pr=55170

[KhafraDev]

The test that failed is:

it('data: URL can always import other data:', async () => {
        const data = new URL('data:text/javascript,');
        data.searchParams.set('body',
                              'import \'data:text/javascript,import \'data:\''
        );
        // doesn't throw
        const empty = await import(data.href);
        assert.ok(empty);
      });

Where empty is 'data:text/javascript,?body=import+%27data%3Atext%2Fjavascript%2Cimport+%27data%3A%27'. The search gets parsed as part of the body (as data: urls cannot have search params). The test is faulty, there were some I changed in this PR that were similarly incorrect.