Ncurses cannot receive resize events after crossterm read

[samtay]

Describe the bug
Ncurses cannot receive window resize events after using crossterm::event::read.

To Reproduce
Using cursive to demonstrate

crossterm::event::read().unwrap(); // blocks and continues after any event, e.g. key press
let mut siv = cursive::ncurses();
siv.add_layer(cursive::views::Dialog::text("foo"));
siv.run();

You'll notice that it doesn't auto resize to keep the dialog centered, because ncurses can't detect window resizing. If you remove the crossterm call, then it works just fine.

Expected behavior
Expect ncurses to be able to operate as normal after using crossterm.

OS
At least Arch Linux.

Terminal/Console
At least kitty and terminator.

Context
See the chat in the discord server. User MrCool seemed to be tracking down a cause.

[sigmaSd]

The issue is here https://github.com/crossterm-rs/crossterm/blob/master/src/event/source/unix.rs#L54
Signals struct needs to be dropped to un-register the signals https://docs.rs/signal-hook/0.1.16/src/signal_hook/iterator.rs.html#98
But even after adding the dropping the issue persists.

strace attests that the un-registering happens: close syscall is called on the fd used with sendto syscall and sendto is never called again after.
currently close is never called and sendto syscall persist with each sigwinch indicating that the signal is still registered

I think:

1- crossterm needs to unregister this signals to free all resources, Since there is no crossterm struct or guard I think the only way is to add a fn clean() to the API that does this

2 - this issue might be in part to ncurses / singal_hook

code to reproduce:

use signal_hook::iterator::Signals;
fn main() {
   let mut signals = Signals::new(&[signal_hook::SIGWINCH]).unwrap();
   signals.close();
   drop(signals);
       
   ncurses::initscr();
   loop {
            if ncureses::getch() == 410 { // 410 is resize event
                      panic!("this wont happen") 
           }        
       }
}

[sigmaSd]

Based on vorner/signal-hook#42 (comment) this is probably ncurses fault.
I think a pragmatic solution might be to re-export signal_hook_cleanup::cleanup_signal and the signals we use like SIGWINCH or just add a note about this interaction in the documentation.