On receiving SIGWINCH, only stdout has its rows/columns updated or resize event fired

[iarna]

I would expect that if both stdout and stderr are bound to /dev/tty then resizes would trigger on both tty objects, but this is not the case. This can be demonstrated with a simple script:

function setup (name, tty) {
  console.error(name, 'started with rows:', tty.rows, 'columns:', tty.columns)
  tty.on('resize', function () {
    console.error('\n',name, 'resized to rows:', tty.rows, 'columns:', tty.columns)
  })
}

setup('stdout', process.stdout)
setup('stderr', process.stderr)

console.error('sleeping for 10 seconds')
setTimeout(function () {}, 10000)

Run and try resizing the window before the timeout. You'll see update events from stdout but none from stderr.

[Fishrock123]

That is the case, yes, even clearly in the code: https://github.com/nodejs/io.js/blob/master/src/node.js#L626-L628

(Calls this: https://github.com/nodejs/io.js/blob/master/lib/tty.js#L76-L92)

And here it is from the blame'd commit: aad12d0#diff-6ff379484cbabad48301d485db111c08R329

I'm guessing it was just forgotten, since I assume stderr would have already existed.

[Fishrock123]

@iarna if you'd like to make a PR it definitely seems reasonable to fix. :)

[iarna]

oh hey that is pretty explicit huh =D

[Fishrock123]

Fixed in bf2cd22, thanks for finding this!

@iarna would you need this to be backported to 0.12 / 0.10, or is it just an annoyance?

[iarna]

It's just an annoyance. I've worked around it by using stdout's info for now, which mostly works.

Obviously if stdout is redirected and stderr isn't it means we can't handle resize events for stderr. But having seen the actual code, I can do better than that. Having it fixed in 0.12 / 0.10 doesn't actually stop me from having to work around the bug, which is why I'm meh on back porting the fix.