Problems with cursor position escape

[cracyc]

Windows Terminal version

1.22.2912.0

Windows build number

10.0.22631.4196

Other Software

https://github.com/cracyc/msdos-player/blob/vt/msdos.cpp#L1088

Steps to reproduce

Start it in a new console window or run in an existing one and execute a command with lots of output.

Expected Behavior

No response

Actual Behavior

When the program starts it immediately gets the console window size and the cursor position which hangs forever. If run from an existing console window it starts fine but if say command.com is run and dir then it will randomly hang waiting for the reply to ESC [6n, although it will resume after a bit or a key is pressed. This does not happen on terminal version 1.21.2911.0, in legacy conhost or if GetConsoleScreenBufferInfo is used instead.

[shaymuiruri]

I would love to work on this issue.

[carlos-zamora]

Keeping the Triage tag on, but this may be of interest to @lhecker

[carlos-zamora]

Thanks for filing. A few questions from the team:

• Could you capture an ETL trace (see https://github.com/microsoft/terminal/wiki/Troubleshooting-Tips#capturing-a-debug-etl-trace) of the bug?
• Is there a simpler way to repro this that doesn't require the attached ms-dos repo?

[j4james]

@carlos-zamora I'm not sure if this is exactly the same thing, but I can reproduce similar behavior with a batch file like this:

@echo �[6n
@set /p "x=RESPONSE:"

NB: The � character is meant to be an escape.

If you run that from within an open terminal window, you'll get the expected DSR response looking something like this:

RESPONSE:^[[13;1R

But if you try and run the the script by double clicking on the file in Windows Explorer, it'll open up in the terminal, and you'll see the RESPONSE: prompt, but no actual DSR response.

Interestingly the problem appears to be specific to the CPR request. Other report sequences I tried (e.g. DA1) worked fine. I suspect we're sending out own CPR request as part of the conpty startup sequence and that possibly ends up grabbing the script's response as well.

[cracyc]

This will hang it for me:

#define NOMINMAX
#define WIN32_LEAN_AND_MEAN
#include <Windows.h>
#include <stdio.h>

int main() {
    const HANDLE input = GetStdHandle(STD_INPUT_HANDLE);

    DWORD input_mode_old;
    GetConsoleMode(input, &input_mode_old);
    SetConsoleMode(input, ENABLE_PROCESSED_INPUT | ENABLE_VIRTUAL_TERMINAL_INPUT);

    for (unsigned long long i = 0;; i++) {
        printf("%llu \x1b[6n", i);

        char byte;
        DWORD chrs;
        do {
            if (!ReadConsoleA(input, &byte, 1, &chrs, NULL) || chrs == 0) {
                goto exit;
            }
            if (byte < 0x20 || byte >= 0x7f) {
                printf("\\x%02x", byte);
            } else {
                printf("%c", byte);
            }
        } while (byte != 'R');

        printf("\n");
    }

exit:
    SetConsoleMode(input, input_mode_old);
    return 0;
}

[lhecker]

That is an excellent repro! It really hangs almost immediately.
I've rewritten your example slightly though to make Ctrl-C work and get the terminal into the previous input mode when it exits. I also added some logic to make sure all possibly non-printables are escaped (just in case). Finally, I added a counter so that it's easier to spot when it stops.