Add CoreGraphics bindings and macOS getAllWindows() implementation

[dbwiddis]

Adds CoreGraphics interface to c.s.j.p.mac providing JNA bindings for the macOS Quartz Window Services and Quartz Display Services APIs.

Also implements getAllWindows() in MacWindowUtils, giving macOS parity with the existing Windows implementation in WindowUtils. (Note that I had to use the Win32 HWND class to be compatible with the DesktopWindow class.)

Window Services

Function	Description
CGWindowListCopyWindowInfo	Enumerate windows with their properties
CGRectMakeWithDictionaryRepresentation	Parse bounds dictionary into CGRect

Constants: kCGWindowListOption* (filtering), kCGWindow* (dictionary keys), sharing types, backing store types.

Display Services

Function	Description
CGMainDisplayID	Get main display ID
CGDisplayBounds	Get display bounds
CGDisplayPixelsWide / CGDisplayPixelsHigh	Display dimensions
CGGetActiveDisplayList / CGGetOnlineDisplayList	Enumerate displays
CGDisplayRotation	Display rotation angle
CGDisplayIsActive / CGDisplayIsMain / CGDisplayIsBuiltin	Display state queries
CGDisplayVendorNumber / CGDisplayModelNumber / CGDisplaySerialNumber	Display identification

Structures

CGPoint, CGSize, CGRect (with ByValue for return types)

MacWindowUtils.getAllWindows()

Uses CGWindowListCopyWindowInfo to enumerate windows, filtering to normal windows (layer 0). Extracts title, owner name, window ID (as HWND), and bounds, using the DesktopWindow class used by the Windows implementation.

Test Output

On my mac:

[junit] Testsuite: com.sun.jna.platform.mac.CoreGraphicsTest
[junit] Tests run: 6, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.349 sec
[junit] 
[junit] ------------- Standard Output ---------------
[junit] First window bounds: x=193.0 y=789.0 w=9.0 h=18.0
[junit] On-screen windows (excluding desktop): 57
[junit] Main display rotation: 0.0
[junit] Main display bounds: 3440.0x1440.0
[junit] Active displays: 2
[junit]   Display 3: 3440x1440
[junit]   Display 1: 1728x1117
[junit] Total windows: 376
[junit] First window owner: CursorUIViewService
[junit] First window PID: 2323
[junit] Main display ID: 3
[junit] ------------- ---------------- ---------------

and

➜  jna git:(add-coregraphics-bindings)   jshell --class-path build/jna.jar:contrib/platform/dist/jna-platform.jar -R-XstartOnFirstThread <<< '
  import com.sun.jna.platform.*;
  var windows = WindowUtils.getAllWindows(true);
  System.out.println("Found " + windows.size() + " windows");
  for (var w : windows) System.out.println(w.getTitle() + " | " + w.getFilePath() + " | " + w.getLocAndSize());
  /exit'
|  Welcome to JShell -- Version 25.0.2
|  For an introduction type: /help intro

jshell> 
jshell> 
jshell> windows ==> [com.sun.jna.platform.DesktopWindow@54a097cc, com ... orm.DesktopWindow@97e1986]

jshell> Found 17 windows

jshell>  | Terminal | java.awt.Rectangle[x=140,y=295,width=1332,height=791]
 | Google Chrome | java.awt.Rectangle[x=944,y=30,width=1534,height=1288]
 | Sublime Text | java.awt.Rectangle[x=1607,y=238,width=1710,height=1002]
 | Eclipse | java.awt.Rectangle[x=278,y=52,width=2301,height=1274]
 | Microsoft Outlook | java.awt.Rectangle[x=1129,y=104,width=1728,height=994]
 | Messages | java.awt.Rectangle[x=304,y=191,width=1024,height=768]
 | Slack | java.awt.Rectangle[x=111,y=122,width=1353,height=1016]
 | Finder | java.awt.Rectangle[x=1804,y=285,width=1556,height=851]
 | Code | java.awt.Rectangle[x=1674,y=249,width=1567,height=848]
 | Code | java.awt.Rectangle[x=1665,y=217,width=1584,height=883]
 | Numbers | java.awt.Rectangle[x=1090,y=30,width=1260,height=1201]
 | zoom.us | java.awt.Rectangle[x=194,y=271,width=1280,height=960]
 | Preview | java.awt.Rectangle[x=4,y=34,width=708,height=328]
 | Preview | java.awt.Rectangle[x=4,y=34,width=708,height=328]
 | Preview | java.awt.Rectangle[x=4,y=34,width=708,height=328]
 | Safari | java.awt.Rectangle[x=1185,y=286,width=1324,height=940]
 | Print Center | java.awt.Rectangle[x=1270,y=443,width=900,height=450]

[matthiasblaesing]

In general the implementation makes sense to me. The usage of HWND is the general API in unfortunate, but the approach taken here is IMHO a good match.

I left two inline comments regarding the data types.

[matthiasblaesing]

Hm - the declaration is a bit more involved:

// https://developer.apple.com/documentation/coregraphics/cgdisplaypixelswide(_:)?language=objc
extern size_t CGDisplayPixelsWide(CGDirectDisplayID display);
// https://developer.apple.com/documentation/kernel/size_t?language=objc
typedef typeof (sizeof(int)) size_t;
// https://developer.apple.com/documentation/coregraphics/cgdirectdisplayid?language=objc
typedef uint32_t CGDirectDisplayID;

If I remember correctly sizeof results a value of type size_t. The size of that depends on the architecture and can be 32bit or 64bit. Could you please check if it makes sense to reuse com.sun.jna.platform.unix.LibCAPI.size_t. That type is already used in SystemB for mac os.

[dbwiddis]

That type is already used in SystemB for mac os.

And that was one of my earliest mappings here.

MacOS has been 64-bit capable since 10.6 (Snow Leopard) and 64-bit only since 10.15 (Catalina) in 2019. We had a discussion about 32-bit support here and it looks like you're only building 64-bit binaries for it recently.

OpenJDK dropped 32-bit support with JDK 25 (probably to avoid providing SIZE_T support in FFM 😉).

I can certainly map it that way, although I have a feeling I've already made a 64-bit assumption in a JNA contribution elsewhere.

[dbwiddis]

OK, size_t is definitely correct here and I looked at all my other contributions here and haven't made any 64-bit assumptions... except one. I added this in #1131 7 years ago and it didn't break anything.

jna/contrib/platform/src/com/sun/jna/platform/mac/SystemB.java

Line 583 in 9b39daf

int host_page_size(int hostPort, LongByReference pPageSize);

That's a vm_size_t which is essentially a NativeLongByReference. Should I fix that in another PR?

[matthiasblaesing]

To keep compatibility I'd deprecate this method and add the one with the correct mapping. In the end yeah mac os is 64bit only and I have some serious hopes, that by the time 128bit becomes mainstream project panama obsoleted JNA ;-)

[dbwiddis]

I have some serious hopes, that by the time 128bit becomes mainstream project panama obsoleted JNA ;-)

FYI I and another maintainer have essentially replicated most of the JNA mappings I use with Panama mappings. Wish there was a repo I could share them...

[matthiasblaesing]

I can't really test, but looks sane and I assume you tested on your end. Thank you.