http: always call response.write() callback

[lpinca]

Ensure that the callback of OutgoingMessage.prototype.write() is
called even when writing empty chunks.

Fixes: #22066

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

[nodejs-github-bot]

Lite-CI: https://ci.nodejs.org/job/node-test-pull-request-lite-pipeline/3527

[lpinca]

This is an attempt to fix #22066.
The idea is to treat empty chunks like all other chunks and write them to the socket. In this way

• The callback is called in and called the correct order.
• We can return the correct boolean value, the one returned by socket.write(), instead of blindly returning true.

The downside is that internal empty writes like

node/lib/_http_outgoing.js

Line 714 in 53bef42

this._send('','latin1',finish);

now trigger a socket.write() call which in turn calls into C++. This could have an impact on performance if writev is not supported. A possible way to mitigate the issue is to not call writeGeneric() if the chunk is empty here

node/lib/net.js

Line 696 in 53bef42

req=writeGeneric(this,data,encoding,cb);

If we don't care about the callbacks order, we can simplify everything and just call the callback of OutgoingMessage.prototype.write() in the next tick when an empty chunk is written.

[addaleax]

I think this is my preferred approach as well 👍

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/23181/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/23193/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/23195/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/23214/

[addaleax]

Landed in 0df581c

[awwright]

What's the test case that this fixes?

This PR seemed to introduce a bug where the callback for OutgoingMessage#end is not always called, since the callback for Writable#write doesn't always call if passed an empty string (where as before this PR, OutgoingMessage#_writeRaw did guarantee this).

But this PR was supposed to fix such an issue, so I'm not sure what's going on here.

[lpinca]

@awwright see #22066 (comment) and included tests.

[awwright]

But #22066 (comment) is incorrect, Stream#write has no guarantee in the code that the callback will always be called, see #29649 for an example