http: do not emit socket errors after complete response

[Archkon]

Fixes: #64272

[nodejs-github-bot]

Review requested:

• @nodejs/http
• @nodejs/net

[mcollina]

I fear we need to bake this a bit
or it might cause regressions.

[mcollina]

lgtm

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/74562/

[Archkon]

@mcollina @jasnell This pr would open for one week, should we ship this to main branch and not let the regession behavior which the issue describled last for long time?

[mcollina]

lgtm

[mcollina]

removed the baking for lts, given the bug is in 24, we might want to have it fixed there

[nodejs-github-bot]

Landed in 17163ea

[lpinca]

@nodejs/http fyi, all socket errors after req.res exists are now swallowed. I think this should be semver-major.

[pimterry]

Agreed @lpinca - this swallows errors for started but incomplete responses in problematic ways.

I think it's an easy fix: we just need to check for res.completed. @mcollina your note before merging in #64272 mentioned this specifically as a concern, is there a reason you decided to leave it & merge instead? Not sure if I'm missing something.

I think we have to to preserve reasonable error handling, I'll open a PR to fix ASAP.

[Archkon]

Agreed @lpinca - this swallows errors for started but incomplete responses in problematic ways.

I think it's an easy fix: we just need to check for res.completed. @mcollina your note before merging in #64272 mentioned this specifically as a concern, is there a reason you decided to leave it & merge instead? Not sure if I'm missing something.

I think we have to to preserve reasonable error handling, I'll open a PR to fix ASAP.

I do check res.completed in my first version of commit but that would cause some tests failed which maybe associated to macos platform difference so I only keep check whether only res exists and I don't remember the details completely now but thanks for continuing to accomplish this

[mcollina]

@lpinca should we revert?