Allow window id to be passed to ImageGrab.grab() on macOS

[yankeguo]

Changes proposed in this pull request:

• add -l argument to screencapture for macOS, allowing capture a window

[radarhere]

Hi. Just for the record, how would a Python user get the window id?

[yankeguo]

here is a example script:

# pip install pyobjc
# pip install pyobjc-core
# pip install pyobjc-framework-quartz

import Quartz

windows = Quartz.CGWindowListCopyWindowInfo(
    Quartz.kCGWindowListExcludeDesktopElements, Quartz.kCGNullWindowID
)

for window in windows:
    window_id = window["kCGWindowNumber"]

    window_name = window["kCGWindowName"]

    window_x = window["kCGWindowBounds"]["X"]
    window_y = window["kCGWindowBounds"]["Y"]
    window_width = window["kCGWindowBounds"]["Width"]
    window_height = window["kCGWindowBounds"]["Height"]

    window_owner_name = window["kCGWindowOwnerName"]
    window_owner_pid = window["kCGWindowOwnerPID"]

user can determine which window id to use

there is also a tool:

https://github.com/smokris/GetWindowID

[radarhere]

There's an interesting problem here.

Originally, we did just crop the image using Pillow. However, that changed in #6152, because #6144 reported that it wasn't accounting for scaling on retina displays.

So if you pick a specific window, how would we know whether it was scaled or not? It might be on a retina display or it might not. If this were Swift, I'd say we want the backingScaleFactor for the window. I'm not convinced AppleScript could easily get information about a window just from the id.

[yankeguo]

I will look into whether there is any information about Retina in the EXIF data of the generated images.

[yankeguo]

seems no good solution, may be we should just put this on hold.

However, using the -l parameter when capturing a window does have value, as it can create a complete screenshot of an obscured window compared to taking a full screen capture and then cropping it.

[yankeguo]

I thought of a less elegant method, using -R0,0,100,100 to sample an image and then calculate the ratio, but I cannot ensure that the window obtained with -l and the image sampled with -R are on the same display.

Another option is to behave like the screencapture command line, ignoring the bbox parameter when passing the window parameter.

[radarhere]

I created yankeguo#2

One of the screencapture flags is

-o In window capture mode, do not capture the shadow of the window.

Excluding the shadow changes the size of the window. So if we take two screenshots of a window, we can measure the size of the shadow.

On a retina screen, the shadow is twice as wide as it is on a non-retina screen.

It is possible that a window doesn't have a shadow, but perhaps that can be considered an edge case.

[hugovk]

What's windowid? There's no other mention in the docs or code.

[radarhere]

I've pushed a commit to describe it as a CGWindowID instead.