using SendInput in Node-FFI

Question

I wanted to use the SendInput function from the windows Api in nodejs, using the FFI package.

My knowledge of C is limited so I can\'t really figure out what problem

Answer 1

I finally found a way to use node-ffi/node-ffi-napi to input key-presses using the SendInput function! (current code below uses node-ffi-napi, since node-ffi has been unmaintained/broken; see edit history for node-ffi version, the api is almost exactly the same)

However, note that there are two ways you can call the SendInput function, as seen here: https://autohotkey.com/boards/viewtopic.php?p=213617#p213617

In my case, I had to use the second (scan code) approach, because the first (virtual key) approach didn't work in the programs I needed the key simulation for.

Without further ado, here is the complete solution:

import keycode from "keycode";
import ffi from "ffi-napi";
import ref from "ref-napi";
import os from "os";
import import_Struct from "ref-struct-di";

var arch = os.arch();
const Struct = import_Struct(ref);

var Input = Struct({
    "type": "int",

    // For some reason, the wScan value is only recognized as the wScan value when we add this filler slot.
    // It might be because it's expecting the values after this to be inside a "wrapper" substructure, as seen here:
    //     https://msdn.microsoft.com/en-us/library/windows/desktop/ms646270(v=vs.85).aspx
    "???": "int",
     
    "wVK": "short",
    "wScan": "short",
    "dwFlags": "int",
    "time": "int",
    "dwExtraInfo": "int64"
});

var user32 = ffi.Library("user32", {
    SendInput: ["int", ["int", Input, "int"]],
    //MapVirtualKeyEx: ["uint", ["uint", "uint", intPtr]],
});

const extendedKeyPrefix = 0xe000;
const INPUT_KEYBOARD = 1;
const KEYEVENTF_EXTENDEDKEY = 0x0001;
const KEYEVENTF_KEYUP       = 0x0002;
const KEYEVENTF_UNICODE     = 0x0004;
const KEYEVENTF_SCANCODE    = 0x0008;
//const MAPVK_VK_TO_VSC = 0;

export class KeyToggle_Options {
    asScanCode = true;
    keyCodeIsScanCode = false;
    flags?: number;
    async = false; // async can reduce stutter in your app, if frequently sending key-events
}

let entry = new Input(); // having one persistent native object, and just changing its fields, is apparently faster (from testing)
entry.type = INPUT_KEYBOARD;
entry.time = 0;
entry.dwExtraInfo = 0;
export function KeyToggle(keyCode: number, type = "down" as "down" | "up", options?: Partial<KeyToggle_Options>) {
    const opt = Object.assign({}, new KeyToggle_Options(), options);
    
    // scan-code approach (default)
    if (opt.asScanCode) {
        //let scanCode = user32.MapVirtualKeyEx(keyCode, MAPVK_VK_TO_VSC); // this should work, but it had a Win32 error (code 127) for me
        let scanCode = opt.keyCodeIsScanCode ? keyCode : ConvertKeyCodeToScanCode(keyCode);
        let isExtendedKey = (scanCode & extendedKeyPrefix) == extendedKeyPrefix;

        entry.dwFlags = KEYEVENTF_SCANCODE;
        if (isExtendedKey) {
            entry.dwFlags |= KEYEVENTF_EXTENDEDKEY;
        }

        entry.wVK = 0;
        entry.wScan = isExtendedKey ? scanCode - extendedKeyPrefix : scanCode;
    }
    // (virtual) key-code approach
    else {
        entry.dwFlags = 0;
        entry.wVK = keyCode;
        //info.wScan = 0x0200;
        entry.wScan = 0;
    }

    if (opt.flags != null) {
        entry.dwFlags = opt.flags;
    }
    if (type == "up") {
        entry.dwFlags |= KEYEVENTF_KEYUP;
    }

    if (opt.async) {
        return new Promise((resolve, reject)=> {
            user32.SendInput.async(1, entry, arch === "x64" ? 40 : 28, (error, result)=> {
                if (error) reject(error);
                resolve(result);
            });
        });
    }
    return user32.SendInput(1, entry, arch === "x64" ? 40 : 28);
}

export function KeyTap(keyCode: number, opt?: Partial<KeyToggle_Options>) {
    KeyToggle(keyCode, "down", opt);
    KeyToggle(keyCode, "up", opt);
}

// Scan-code for a char equals its index in this list. List based on: https://qb64.org/wiki/Scancodes, https://www.qbasic.net/en/reference/general/scan-codes.htm
// Not all keys are in this list, of course. You can add a custom mapping for other keys to the function below it, as needed.
let keys = "**1234567890-=**qwertyuiop[]**asdfghjkl;'`*\\zxcvbnm,./".split("");

export function ConvertKeyCodeToScanCode(keyCode: number) {
    let keyChar = String.fromCharCode(keyCode).toLowerCase();
    let result = keys.indexOf(keyChar);
    console.assert(result != -1, `Could not find scan-code for key ${keyCode} (${keycode.names[keyCode]}).`)
    return result;
}

To use it, call:

KeyTap(65); // press the A key

Or, if you're using the keycode npm package:

import keycode from "keycode";
KeyTap(keycode.codes.a);

Answer 2

Here's a working example that presses the a key. It employs ref-struct-napi and ref-union-napi to accurately represent the INPUT structure.

const FFI = require('ffi-napi')
const StructType = require('ref-struct-napi')
const UnionType = require('ref-union-napi')
const ref = require('ref-napi')


const user32 = new FFI.Library('user32.dll', {
  // UINT SendInput(
  //   _In_ UINT cInputs,                     // number of input in the array
  //   _In_reads_(cInputs) LPINPUT pInputs,  // array of inputs
  //   _In_ int cbSize);                      // sizeof(INPUT)
  'SendInput': ['uint32', ['int32', 'pointer', 'int32']],
})

// typedef struct tagMOUSEINPUT {
//   LONG    dx;
//   LONG    dy;
//   DWORD   mouseData;
//   DWORD   dwFlags;
//   DWORD   time;
//   ULONG_PTR dwExtraInfo;
// } MOUSEINPUT;
const MOUSEINPUT = StructType({
  dx: 'int32',
  dy: 'int32',
  mouseData: 'uint32',
  flags: 'uint32',
  time: 'uint32',
  extraInfo: 'pointer',
})

// typedef struct tagKEYBDINPUT {
//   WORD    wVk;
//   WORD    wScan;
//   DWORD   dwFlags;
//   DWORD   time;
//   ULONG_PTR dwExtraInfo;
// } KEYBDINPUT;
const KEYBDINPUT = StructType({
  vk: 'uint16',
  scan: 'uint16',
  flags: 'uint32',
  time: 'uint32',
  extraInfo: 'pointer',
})

// typedef struct tagHARDWAREINPUT {
//   DWORD   uMsg;
//   WORD    wParamL;
//   WORD    wParamH;
// } HARDWAREINPUT;
const HARDWAREINPUT = StructType({
  msg: 'uint32',
  paramL: 'uint16',
  paramH: 'uint16',
})

// typedef struct tagINPUT {
//   DWORD   type;
//   union
//   {
//     MOUSEINPUT      mi;
//     KEYBDINPUT      ki;
//     HARDWAREINPUT   hi;
//   } DUMMYUNIONNAME;
// } INPUT;
const INPUT_UNION = UnionType({
  mi: MOUSEINPUT,
  ki: KEYBDINPUT,
  hi: HARDWAREINPUT,
})
const INPUT = StructType({
  type: 'uint32',
  union: INPUT_UNION,
})

const pressKey = (scanCode) => {
  const keyDownKeyboardInput = KEYBDINPUT({vk: 0, extraInfo: ref.NULL_POINTER, time: 0, scan: scanCode, flags: 0x0008})
  const keyDownInput = INPUT({type: 1, union: INPUT_UNION({ki: keyDownKeyboardInput})})
  user32.SendInput(1, keyDownInput.ref(), INPUT.size)

  const keyUpKeyboardInput = KEYBDINPUT({vk: 0, extraInfo: ref.NULL_POINTER, time: 0, scan: scanCode, flags: 0x0008 | 0x0002})
  const keyUpInput = INPUT({type: 1, union: INPUT_UNION({ki: keyUpKeyboardInput})})
  user32.SendInput(1, keyUpInput.ref(), INPUT.size)
}

pressKey(0x1E)

If you want to perform a single call to SendInput that includes multiple key presses, construct an array of INPUT structs:

const pressKey = (scanCode) => {
  const inputCount = 2
  const inputArray = Buffer.alloc(INPUT.size * inputCount)
  const keyDownKeyboardInput = KEYBDINPUT({vk: 0, extraInfo: ref.NULL_POINTER, time: 0, scan: scanCode, flags: 0x0008})
  const keyDownInput = INPUT({type: 1, union: INPUT_UNION({ki: keyDownKeyboardInput})})
  keyDownInput.ref().copy(inputArray, 0)
  const keyUpKeyboardInput = KEYBDINPUT({vk: 0, extraInfo: ref.NULL_POINTER, time: 0, scan: scanCode, flags: 0x0008 | 0x0002})
  const keyUpInput = INPUT({type: 1, union: INPUT_UNION({ki: keyUpKeyboardInput})})
  keyUpInput.ref().copy(inputArray, INPUT.size)
  user32.SendInput(inputCount, inputArray, INPUT.size)
}

Answer 3

The "1" tells you that 1 event was inserted, not what the event actually is. I don't know about FFI but it seems to me that keyboardInput has some invalid type definitions. wVK and wScan must be 16-bit integers (hence the 'w' for WORD). Since they are typed the same as dwFlags (an 'int') that's cause invalid input values.