resize event not firing

[Flet]

• Version: v8.5.0 through v8.7.0
• Platform: Darwin
• Subsystem:

In versions v8.5.0 and higher, the resize event does not seem to fire on terminal window resize.

Test code:

process.stdout.on('resize', function () {
 console.log('resized')
})

setInterval(function () {
}, 0)

I was able to have success when trying v8.4.0, but v8.5.0 does not work.

Possibly related recent closed issue:
#13197

[refack]

/cc @nodejs/platform-macos

@Flet #13197 would have been my first guess too, but it was closed by a documentation change - ff07eaa
Later the underlying event mechanism was improved in libuv/libuv@6ad1e81 but that change was for Windows only, and your report states Darwin as the platform.

[bnoordhuis]

I'm not sure what broke exactly but it starts working again when you add a no-op process.on('SIGWINCH', () =>{}) listener. Manual bisecting indicates it's not the libuv upgrade.

lib/internal/process/stdio.js installs a SIGWINCH handler but it doesn't fire until another one is added because the 'newListener' handler for signals hasn't been installed yet.

My initial guess was that some of the inspector changes broke it because the problem goes away with this patch but I haven't pinpointed the cause yet:

diff --git a/lib/internal/bootstrap_node.js b/lib/internal/bootstrap_node.js
index e7676c0867..b9ac545dd1 100644
--- a/lib/internal/bootstrap_node.js
+++ b/lib/internal/bootstrap_node.js
@@ -36,6 +36,7 @@
 const browserGlobals = !process._noBrowserGlobals;
 if (browserGlobals){
 setupGlobalTimeouts();
+ _process.setupSignalHandlers();
 setupGlobalConsole();
 }

@@ -57,7 +58,6 @@
 _process.setup_cpuUsage();
 _process.setupMemoryUsage();
 _process.setupKillAndExit();
- _process.setupSignalHandlers();
 if (global.__coverage__)
 NativeModule.require('internal/process/write-coverage').setup();

[maclover7]

Did a bisect, and it seems like f34e0f9 is what introduced the new behavior.

[refack]

@maclover7 , @Flet can you devise a regression test for this? (something like: start a child node process in a terminal, then try to script a resize of that terminal)

Another good test to have is a unit test of the event on the node side (one that mocks the libuv behaviour)

[bnoordhuis]

Did a bisect, and it seems like f34e0f9 is what introduced the new behavior.

That commit was one of my suspects but its code does not exist anymore in v8.6.0, it was removed in 8ce0e9a (which is a somewhat suspect commit as well.)

[Fishrock123]

This sounds reasonable to me, we should set up the signal handlers before loading the console / process.{stdio}.

[bnoordhuis]

#16225