Add a low level window class

[JeremyKuhne]

Adds WindowClass for testing windows messages.

Adds InternalsVisibleTo so we can expose helpers for Interop testing from internal test utility.

Ref to SWF in internal test utility should be removed. This is one step. Tests using AxHost should refactor AxHost helper methods into SWF.Primitives (don't move the whole AxHost down).

Fix a bug in Cursor interop.

Microsoft Reviewers: Open in CodeFlow

[codecov]

Codecov Report

Merging #3010 into master will increase coverage by 0.01157%.
The diff coverage is 100.00000%.

@@                 Coverage Diff                 @@
##              master       #3010         +/-   ##
===================================================
+ Coverage   62.14366%   62.15523%   +0.01157%     
===================================================
  Files           1257        1257                 
  Lines         449428      449428                 
  Branches       39227       39227                 
===================================================
+ Hits          279291      279343         +52     
+ Misses        164660      164606         -54     
- Partials        5477        5479          +2     

Flag	Coverage Δ
#Debug	62.15523% <100.00000%> (+0.01157%)	⬆️
#production	33.34748% <100.00000%> (+0.02061%)	⬆️
#test	98.98504% <ø> (ø)

[hughbe]

should we extract this to enum IDC

[JeremyKuhne]

Thinking it over a bit. They're technically native integers (IntPtr) which is giving me some pause. Probably overthinking. :)

[weltkante]

Actually, no, they aren't IntPtr constants, they are 2 byte integers. You are supposed to pass them through MAKEINTRESOURCE which takes a UInt16. Of course if you drop the distinction between integer resource ids and string resource names then both can be treated as IntPtr (basically inlining MAKEINTRESOURCE into the definition)

From an API point of view I'd prefer providing overloads for enum resource ids vs. string resource names rather than merging both into IntPtr (i.e. inline MAKEINTRESOURCE into the called function overload instead of having the caller do it)

[JeremyKuhne]

Well, technically the numbers themselves are int literals, and the define usage (e.g. IDC_NO, etc.) always effectively gives you a char* or w_char* "literal". 😛

I do agree with you, however, that passing them around as int is better in this case as there aren't a lot of API entry points. In some other cases (such as HWND_TOPMOST), I am inclined to expose statics.

[weltkante]

Not trying to nitpick but my point was that any literal integer not fitting in a 2 byte integer wouldn't be allowed because it'd be either cut off when passed through MAKEINTRESOURCE or, if you would put it directly into an IntPtr, it would be interpreted as a pointer to a resource name, not a numeric resource id. So for all practical purposes the numeric range of resource ids is UInt16.