add win32 DLL ordinal value function support

[kobevaliant]

add win32 DLL ordinal value function support
if funname is like '#123', a ordinal value will be used instead of a string representation

[twall]

Looks reasonable. Tests?

[dblock]

Check for NULL, too.

[kobevaliant]

it's already checked in function 'Java_com_sun_jna_Native_findSymbol'

[dblock]

Until someone removes it or calls this from another scope. The repercussions of security vulnerabilities caused by this in case that happens are pretty major, it costs nothing at runtime and is basic defensive programming, just my 0.02c.

[matthiasblaesing]

Is it safe to assume, that no ordinary function name begins with the #-mark? If (even just in theory) a function name could clash with this - I see the need for either an escape mechanism or a different implementation.

[dblock]

Is there a way to write this to avoid any ambiguity? Like @FuncNumber(1206)?

[matthiasblaesing]

I think this could be handled completly without touching the dispatching mechanism, purely on the java side:

First I modified Kernel32.java to include the only necessary new function:

Pointer GetProcAddress(HMODULE hmodule, int ordinal);

This way I'll get access to the pointer leading me to the function itself. That used in the sample:

public class InvokeByOrdinal {
    static NativeLibrary user32;

    public static void main(String[] args) {
        // ensure user32 is loaded and hold it static
        user32 = NativeLibrary.getInstance("user32");
        // get module handle needed to resolve function pointer via GetProcAddress
        HMODULE module = Kernel32.INSTANCE.GetModuleHandle("user32");
        // resolve function pointer
        //
        // From link.exe /dump /exports c:\\Windows\\System32\\user32.dll:
        //    ordinal hint RVA      name
        //       1822  13A 00039220 GetDesktopID
        //       1823  13B 000176B0 GetDesktopWindow
        //       1824  13C 0002E710 GetDialogBaseUnits
        Pointer funcPointerGetDesktopWindow = Kernel32.INSTANCE.GetProcAddress(module, 1823);
        // Wrap Pointer into Function
        Function funcGetDesktopWindow = Function.getFunction(funcPointerGetDesktopWindow);
        // Invoke GetDesktopWindow via manually resolved ordinal lookup and output pointer value
        Pointer hwndPointer = funcGetDesktopWindow.invokePointer(new Object[]{});
        System.out.println(Pointer.nativeValue(hwndPointer));
        // Invoke GetDesktopWindow via JNA bindings
        HWND hwnd = User32.INSTANCE.GetDesktopWindow();
        System.out.println(Pointer.nativeValue(hwnd.getPointer()));
    }
}

leads to the same outputs. I also tried ShowCursor, but learned, that that does not effect the global cursor visiblity and without a window the effect is underwhelming.

This could be used for an InvocationHandler, that resolves the function pointers on first call/greedy and dispatches them through the bound functions.

[kobevaliant]

I think it is elegant and consistent with my intuition by using '#', and it is implemented easily (A single line of code).
There's same usage in .NET, so I think it is somehow safe about ambiguity.
https://msdn.microsoft.com/en-us/library/f5xe74x8(v=vs.110).aspx

Surely I expect more elegant way using @Anotation, while it needs more code and change.

[matthiasblaesing]

@kobevaliant please have a look at this different approach:

matthiasblaesing@5f1c191

While the naming is discussable, I think this is safer - there are no side effects for other architectures and no it is still reasonablely short.

What I'd like to know: Is there a real life sample, that demonstrates the need for this?

[kobevaliant]

Nice job, It meets my need enough.
Sometimes the dll owner don't want to expose the minimal information about the dll functions, so they choose to hide their function names entirely for privacy.

[matthiasblaesing]

@kobevaliant please have a look at this PR: #705 that PR implements the necessary functions for the OrdinalValueInvoker. These are general enhancements. For your concrete problem feel free to implementation from https://github.com/matthiasblaesing/jna/tree/ordinalvalue_invoker (I added a license statement as MIT, so feel free to integrate).

If someone makes a case why this is of general interest, I'll reevaluate pulling this into JNA itself.

If you agree, I'll merge #705 and close this request.

[matthiasblaesing]

Ok - I cleaning up. Feel free to contact me if you need help with the OrdinalValueInvoker.