Threadpool API should use handles, not pointers to structs

[kennykerr]

Types like PTP_WORK, PTP_TIMER, PTP_WAIT, PTP_IO are all handle types (opaque pointers) and should be modeled as such. While there are some Threadpool types that are defined as structs, like TP_CALLBACK_ENVIRON_V3, most should be considered opaque handles.

typedef struct _TP_WORK TP_WORK, *PTP_WORK;

So for example, TP_WORK should not be defined at all. Instead, there should be a PTP_WORK struct with the NativeTypeDef attribute and the InvalidHandleValue attribute.

[kennykerr]

All of these have an "invalid handle value" of 0.

[mikebattista]

Does the below look right? PTP_* are then remapped to the equivalent TP_* with a pointer.

  {
    "Namespace": "Windows.Win32.System.Threading",
    "Name": "TP_IO",
    "ValueType": "IntPtr",
    "InvalidHandleValues": [ 0 ]
  },
  {
    "Namespace": "Windows.Win32.System.Threading",
    "Name": "TP_TIMER",
    "ValueType": "IntPtr",
    "InvalidHandleValues": [ 0 ]
  },
  {
    "Namespace": "Windows.Win32.System.Threading",
    "Name": "TP_WAIT",
    "ValueType": "IntPtr",
    "InvalidHandleValues": [ 0 ]
  },
  {
    "Namespace": "Windows.Win32.System.Threading",
    "Name": "TP_WORK",
    "ValueType": "IntPtr",
    "InvalidHandleValues": [ 0 ]
  },

Windows.Win32.System.Threading.TP_IO :  => [InvalidHandleValue(0),NativeTypedef]    
Windows.Win32.System.Threading.TP_IO.Value added
Windows.Win32.System.Threading.TP_TIMER :  => [InvalidHandleValue(0),NativeTypedef] 
Windows.Win32.System.Threading.TP_TIMER.Value added
Windows.Win32.System.Threading.TP_WAIT :  => [InvalidHandleValue(0),NativeTypedef]  
Windows.Win32.System.Threading.TP_WAIT.Value added
Windows.Win32.System.Threading.TP_WORK :  => [InvalidHandleValue(0),NativeTypedef]  
Windows.Win32.System.Threading.TP_WORK.Value added

[riverar]

I wouldn't edit the name of these types. The P preface should just be considered historical and unreliable (for these particular types). All devs need to know is that these are just badly named handles (integers).

(In other words, undo the P removal and the rest looks good.)

[kennykerr]

Right, the docs only mention the PTP_X names - stick with that.

[mikebattista]

Could you take a look at https://github.com/microsoft/win32metadata/tree/mikebattista/threadingptptypes?

It looks to be consistent now with how PTP_POOL was implemented but given the odd definitions and changes to the [In][Out] attributes in the diff, it needs to be tested to make sure the dev experience is what you expect.

[kennykerr]

Sorry, I'm not sure how to check this without a winmd.

[mikebattista]

I sent you one offline.