fix: permission prompt stuffing on Windows

[dsherret]

This unfortunately makes Windows act differently in that pressing the "y" or "n" key immediately confirms or denies the prompt. It would take a lot more code to make this work otherwise.

Also, I looked into using https://docs.microsoft.com/en-us/windows/console/writeconsoleinput to create unit tests, but I couldn't figure out a way to construct an INPUT_RECORD.

[piscisaureus]

I couldn't figure out a way to construct an INPUT_RECORD.

I think it'd be something like this:

let mut input_record = INPUT_RECORD { EventType: KEY_EVENT, ..Default::default() };
input_record.Event.KeyEvent_mut().wRepeatCount = 1;
input_record.Event.KeyEvent_mut().uChar.UnicodeChar_mut() = 'y' as WCHAR;

[dsherret]

@piscisaureus oh, nice! Thanks for that. I will look into adding a test tomorrow.

Edit: Oh, I could probably also just do std::mem::zeroed() as mentioned in the first FAQ of their documentation...

[piscisaureus]

This unfortunately makes Windows act differently in that pressing the "y" or "n" key immediately confirms or denies the prompt. It would take a lot more code to make this work otherwise.

What makes it difficult in particular? Afaik ReadConsoleW() provides a way to do line-buffered input.

[dsherret]

@piscisaureus calling ReadConsoleW would return the permission stuffed text even after calling FlushConsoleInputBuffer. I think perhaps FlushConsoleInputBuffer flushes the input buffer events to a character buffer that ReadConsoleW then reads from (that would be my guess given the behaviour I'm seeing... you may want to try something out and see if you can figure out a way to flush that buffer... I haven't been able to figure it out yet).

https://github.com/denoland/deno/pull/11969/files#diff-ddef1e94cdb9cee582c71ed2e1a3948e4a1a80cdcf311693b8a41d13a79aa418R1247-R1252

By the way, here is my test script from Ryan's tests:

const buf = new Uint8Array(1);
console.log("Enter 'yy':");
await Deno.stdin.read(buf);
await Deno.permissions.request({ "name": "env" });
console.log("\n\nOwned", Deno.env.get("SOMETHING"));

[piscisaureus]

@piscisaureus calling ReadConsoleW would return the permission stuffed text even after calling FlushConsoleInputBuffer. I think perhaps FlushConsoleInputBuffer flushes the input buffer events to a character buffer that ReadConsoleW then reads from

I would be very surprised if this was case...

[dsherret]

@piscisaureus in the code in this PR if you try adding:

      assert_eq!(
        FlushConsoleInputBuffer(stdin),
        winapi::shared::minwindef::TRUE
      );

      // add this after the existing `FlushConsoleInputBuffer` (shown above for reference)
      let mut buffer = [0u16; 1024];
      let mut chars_read = 0;
      assert_eq!(winapi::um::consoleapi::ReadConsoleW(stdin, buffer.as_mut_ptr() as _, buffer.len() as _, &mut chars_read, std::ptr::null_mut()), 1);
      println!("Chars read: {}", chars_read);
      let text = std::ffi::OsString::from_wide(&buffer[0..chars_read as usize]).to_string_lossy().to_string();
      println!("Text: {:?}", text);

...then running this JS code if you type yy<ENTER> then you will see the following output:

Maybe I'm doing something wrong? Again, I'm not super familiar with the winapi. Hopefully there is a way to do this that I'm not seeing.

[piscisaureus]

I’ll take a look and try to figure this out tomorrow.

[piscisaureus]

I wrote a simple standalone test: https://gist.github.com/piscisaureus/42fd7536b075f447682977083c7ff678
It seems that FlushConsoleInputBuffer() is working fine with ReadConsoleW().

[dsherret]

@piscisaureus try my fork https://gist.github.com/dsherret/72991b395d68ac6aff912199e5e44ac5

By the way, from these tests it's clear that FlushConsoleInputBuffer does not flush the console input buffer to some other buffer like I speculated, but it does look like whatever ReadConsoleW reads from is filled before the FlushConsoleInputBuffer gets to it (maybe on enter key press?)