error handler of same domain can be called several times when it throws

• Version: Current tip of master (66f45e7e5b), but probably applies to all versions.
• Platform: All platforms.
• Subsystem: domain.

The following code:

'use strict';

const common = require('../common');
const domain = require('domain');
const http = require('http');

const server = http.createServer((req, res) => {
  res.end();
});

let numDomainErrorListenerCalls = 0;

function performHttpRequestWithDomain(cb) {
  const d = domain.create();
  d.run(() => {
    const req = http.get({
      host: '127.0.0.1', port: server.address().port
    }, (res) => {
      res.on('data', () => {});
      res.on('end', () => {
        throw new Error('bang');
      });
    });

    req.end();
  });

  d.on('error', (domainErr) => {
    console.log(++numDomainErrorListenerCalls);
    throw new Error('boom');
  });
}

server.listen(0, '127.0.0.1', () => {
  performHttpRequestWithDomain(common.mustCall(() => {
    server.close();
  }));
});

gives the following output:

$ ./node test/parallel/test-http-req-domain-stack.js 
1
2
/Users/jgilli/dev/node/test/parallel/test-http-req-domain-stack.js:30
    throw new Error('boom');
    ^

Error: boom
    at Domain.d.on (/Users/jgilli/dev/node/test/parallel/test-http-req-domain-stack.js:30:11)
    at Domain.emit (events.js:188:13)
    at Domain.EventEmitter.emit (domain.js:430:20)
    at Domain._errorHandler (domain.js:216:23)
    at Domain._errorHandler (domain.js:244:33)
    at Object.setUncaughtExceptionCaptureCallback (domain.js:132:29)
    at process._fatalException (internal/process/execution.js:102:29)

The uncaught exception is expected. What is not expected as far as I understand is for the same domain's error handler to run more than once.

I believe the original intention of the domain's implementation is to pop the domains stack when a domain's error handler throw, so that the domain that handles that new error is the "parent" domain.

However, in the example above the same domain is pushed on the stack more than once. For instance, when an event is emitted from a nextTick callback, the same domain will be entered from the nextTick callbacks scheduler and then once again from the event emitter.

Pushing the same domain on the stack more than once makes sense so that the components that push a domain can pop it from the stack. However, I think we could probably replace the call to pop the stack once in the domain error handling code to remove all consecutive instances of that domain instead.

@nodejs/domains Thoughts?

Start with test/parallel/test-http-req-domain-stack.js and reproduce the repeated error-listener calls. Read lib/domain.js alongside internal/process/next_tick.js, especially the referenced stack-handling paths. Done means the regression is covered and the same domain's error handler is no longer called repeatedly when it throws.