fix(terminal): report terminal window size in pixels

[lumynou5]

Window size in pixels is required by some programs, reporting it can help those programs work inside NeoVim terminals.

Closes #8259.

However, this patch doesn't make icat work. Instead, it just makes the program show another error, because the problem is not only related to window sizes in pixels but how escape sequences (specifically, APC) are handled. See #4349, #20176.

[justinmk]

Nice, thanks. If no one else has comments, ping me if this isn't merged in a couple days

[justinmk]

I wonder if this fixes any of the log messages that were showing in Windows CI

neovim/test/testutil.lua

Line 95 in cf9f002

['Warning: noted but unhandled ioctl'] = 1,

[lumynou5]

Currently the test is failed. When stdout is redirected, ioctl calls fail with errno ENOTTY, and the program crashes due to invalid division.
Adding a check to set pixel sizes to 0 in case of failed calls can help, but it seems not a desirable solution?

  int res = ioctl(1, TIOCGWINSZ, &ptyproc->winsize);
  ptyproc->winsize = (struct winsize){
    ptyproc->height,
    ptyproc->width,
    res ? 0 : (ptyproc->winsize.ws_xpixel / ptyproc->winsize.ws_col) * ptyproc->width,
    res ? 0 : (ptyproc->winsize.ws_ypixel / ptyproc->winsize.ws_row) * ptyproc->height,
  };

[zeertzjq]

It seems this doesn't actually solve anything, and may be incorrect. For example, when running Nvim with --headless and attaching another TUI with --remote-ui, this will use the pixel size of the terminal running the --headless instance instead of the one running the --remote-ui instance.

[lumynou5]

It seems this doesn't actually solve anything, and may be incorrect. For example, when running Nvim with --headless and attaching another TUI with --remote-ui, this will use the pixel size of the terminal running the --headless instance instead of the one running the --remote-ui instance.

Thanks. I didn't think about remote. I'll try to solve it.