win, tty: fix problem of receiving unexpected SIGWINCH

[erw7]

Fix an issue where WINDOWS_BUFFER_SIZE_EVENT occurs and unexpected SIGWINCH is received before calling uv__tty_console_resize_message_loop_thread().

We encountered this problem in neovim functional tests.

Ref. neovim/neovim#10978 (comment)

[bnoordhuis]

@bzoz You're probably the most qualified to review this.

[saghul]

Since this was caught with a test, any chance one can be written so we can catch it if it comes back in the future?

[bzoz]

LGTM

[erw7]

Since this was caught with a test, any chance one can be written so we can catch it if it comes back in the future?

To write a test that will catch this problem reliably, we need to call uv__tty_console_resize_message_loop_thread() after the first WINDOWS_BUFFER_SIZE_EVENT. I think it is possible to implement such behavior only during testing, but it will be too complicated.

1beb739 tried to add assert so that it can be caught when this problem recurs in the future. How about this?

[saghul]

That assert would trip if GetConsoleScreenBufferInfo had returned error, right? Is that possible? We might want to abort() if we are in a "the code is broken" situation, IMHO.

[erw7]

That assert would trip if GetConsoleScreenBufferInfo had returned error, right? Is that possible? We might want to abort() if we are in a "the code is broken" situation, IMHO.

@saghul I'm not sure, but if the GetConsoleScreenBufferInfo() on line 177 returns an error, the assert will not fail because GetConsoleScreenBufferInfo() on line 2369 also returns an error.

[saghul]

Gotcha. LGTM.

[saghul]

Landed in 5013042, thank you!