Link to issue number:

None

Summary of the issue:

There are 4 functions wrappers related to screen, client, logical and physical coordinate conversions.

• windowUtils.logicalToPhysicalPoint
• windowUtils.physicalToLogicalPoint
• winUser.ClientToScreen
• winUser.ScreenToClient

These functions take a window handle and a point structure and return 1 on success and 0 on failure. The user32 functions also set the last error when the functions fail. The point in the structure is used to calculate a new point, and on success, the coordinates in the structure are replaced. On failure, the coordinates stay the same.

However, the wrappers of these functions silently return whatever is in the structure, thereby silently ignoring whether the underlying function succeeded or failed.
As far as I've seen, for the user32 functions this is only a problem when giving it a wrong window handle. Converting coordinates falling out of the client area of the window works just fine. However, for physicalToLogicalPoint and logicalToPhysicalPoint, the functions also fail when the given coordinates fall outside the window. While normally, providing the wrong coordinates to these functions should not happen, not being able to see whether the underlying function succeeded or failed is problematic for debugging reasons.

Description of how this pull request fixes the issue:

1. The winUser wrappers will now raise a WindowsError on failure. Note that this does not occur when providing it coordinates that fall outside the client window.
2. the windowUtils functions raise a RuntimeError. Unfortunately, when these functions fail, they do not set the last error, so there's no way to see why the function failed.
3. The corresponding calls in locationHelper do not catch errors, so calling screenToClient on a point or rectangle will raise a WindowsError on failure, etc.
4. Places that call the mentioned functions, particularly displayModel, will now properly catch errors and log them.
5. It removes Windows XP code from windowUtils

Testing performed:

This is pretty hard to test. I stumbled upon this bug when testing the highlighter in the vision framework (#9064) as that converts physical to logical rectangle coordinates. It silently seemed to fail the conversion of the top left coordinates where it succeeded in converting the bottom right coordinates, resulting in a malformed rectangle.

Known issues with pull request:

None

Change log entry:

• Changes for developers
  • The following functions now raise an exception on failure instead of a silent return: (Fix window point conversion functions not exposing failure #9269)
    • RuntimeError: windowUtils.logicalToPhysicalPoint and windowUtils.physicalToLogicalPoint
    • WindowsError: winUser.ClientToScreen and winUser.ScreenToClient
    • Note that the wrapper methods in the locationHelper module will now also raise these exceptions on failure.