Incorrect loop->active_handles count when nested setTimeout calls have same value

[abbr]

I maintain a c++ module deasync (originally written by @vkurchatkin) which essentially exposes uv_run to jsLand in order to address some use cases that cannot be handled by other async-2-sync modules. A problem was reported recently that when deasync is involved in nested setTimeout calls with same timeout value, the process hung. It is illustrated by code

var deasync = require('deasync');

function async(cb) {
  console.log('x');
  setTimeout(function () {
    console.log('y');
    cb(null, 'value');
    console.log('z');
  }, 8);
}

setTimeout(function () {
  console.log('A', deasync(async)());
}, 8); // changing 8 to, say 9, works

output

x
(hung)

Changing one of the timeout values of setTimeout yields expected output

x
y
z
A value

Problem is reproducible in latest Node version on at least Linux and Windows. I have run node in gdb session and found the problem is loop->active_handles has gone down to 0 after the inner setTimeout is called but before its handler is triggered, therefore the inner handler doesn't have a chance to run.

I think the cause is that timers of same timeout value are implemented using a linked list and only one handle is exposed to libuv. This handle is removed at the time the outer setTimeout handler is called.

By moving list.start(msecs, 0) in /lib/timers.js outside else block, I have verified the problem can be fixed. Will supply a PR later. What I am not clear is the impact of this change. Appreciate if someone can review the issue and my PR.

(P.S. I have read Node issue reporting guidelines and understand I am not supposed include dependencies. But deasync is simple, has no dependency by itself, and is easy to be re-factored into a dependency-less mixture of c++ and js code)

[tunniclm]

Could be related to #25763

[misterdjules]

@tunniclm Although it seems at first sight, this issue is not related to #25763 :) @abbr's analysis is spot on.

While deasync seems to rely too much on Node.js' internals, the fix suggested in #25832 actually looks good to me so far, and thus I think I would be inclined to merge it in.

Thank you both @abbr @tunniclm for your input!

[kjhangiani]

Has there been any progress on this? Does this fix have other potential consequences?