Alexander Halser

Yes, that's the core issue. When loading an image, we must know - or be able to know - if it requires extra handling. Then we can make an informed decision. For myself, the issue is solved with the Jpeg-load-helper unit. Now I have to implement that in VCL as well... 🙂

Agreed. But this EXIF information is crucial (at least the orientation), from a practical point of view.

I wonder if anyone is interested in a solution... 😉 Just in case you are, here is what I did. Neither FMX nor VCL will honor the orientation information in the EXIF header inside the JPEG. For FMX: neither on Windows nor on macOS. And probably not on iOS, either. For Android and Linus, you need to test for yourself, maybe they do. The only solution is to read the EXIF header, extract the Orientation tag and apply the necessary transformations to the image after loading. There are several components for Delphi to read EXIF headers, some do work, some don't. The best known unit is CCR-Exif (https://github.com/Wolfcast/ccr-exif), which does everything and works quite well. It can read and write EXIF tags. I've created my own, because CCR-Exif seems like overkill (after all, I just want to know 1 single byte in the JPEG) If you want to display JPEG images correctly in your FMX application, I have published my little unit on SourceForge, It's as small and fast as possible with really minimal overhead. The first function JPEGRotationFromStream() will work for VCL as well! So you can adapt it and use it for your VCL app, too. But FMX has this nice little class called TBitmapSurface, which easily does the Rotate, Mirror and Flip transformations that are required to display the JPEG correctly. So, for FMX this unit goes the full length and delivers a TBitmap that's just right. The unit might be useful for VCL as well, but you have to go an extra mile to apply the transformations. SourceForge Download: https://sourceforge.net/projects/delphi-fmx-jpeg-loader-exif/

This is a blank FMX form in Delphi 11 with a TImage control, that loads a JPEG. The JPEG picture was created with a Samsung smart phone in portrait mode. Hence, the picture has a width of 2268 and a height of 4032 pixels. The JPEG's EXIF header states that orientation is portrait. That's obvious for everyone who looks at the picture, except for Delphi Firemonkey. TBitmap.LoadfromFile() ignores the EXIF orientation altogether and loads the JPEG with its physical dimensions (which are width 4032 and height 2268 pixels). It doesn't rotate the picture after loading. Has anyone solved this nasty problem for FMX?

Problem solved. It wasn't the permissions, but a native NSToolbar control. This toolbar contained two NSToolbarSidebarTrackingSeparator items, which were used to create a vertical separator bar on the toolbar. While this had worked before, it's illegal for macOS 15. Only one of this item type is permitted.

Signed and notarized, as explained above. Shouldn't that be sufficient?

I have a Delphi FMX app that runs fine on anything from macOS Big Sure to Sonoma. Just on the latest Sequoia release, it doesn't start. It seems, that this has to do with new sandbox restrictions. Any ideas what to change in the entitlements to make it run on Sequoia? The app is signed and notarized, compiled with Delphi 11.3. This is the crash log I get on Sequoia (from a different mac, not my development machine): 2024-11-07 08:43:02.786318 <Notice>: Last log repeated 1 times 2024-11-07 08:43:02.785852 (gui/501/application.SnipSVG.16305011.16305018) <Notice>: internal event: WILL_SPAWN, code = 0 2024-11-07 08:43:02.785996 (gui/501/application.SnipSVG.16305011.16305018) <Notice>: service state: spawn scheduled 2024-11-07 08:43:02.785999 (gui/501/application.SnipSVG.16305011.16305018) <Notice>: service state: spawning 2024-11-07 08:43:02.786053 <Notice>: Coalition Cache Hit: app<application.SnipSVG.16305011.16305018(501)> [985] 2024-11-07 08:43:02.786105 (gui/501/application.SnipSVG.16305011.16305018) <Notice>: launching: launch job demand 2024-11-07 08:43:02.787814 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: xpcproxy spawned with pid 1131 2024-11-07 08:43:02.787872 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: internal event: SPAWNED, code = 0 2024-11-07 08:43:02.787875 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: service state: xpcproxy 2024-11-07 08:43:02.787956 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: internal event: SOURCE_ATTACH, code = 0 2024-11-07 08:43:02.813870 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: service state: running 2024-11-07 08:43:02.813881 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: internal event: INIT, code = 0 2024-11-07 08:43:02.813884 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: job state = running 2024-11-07 08:43:02.814097 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: Successfully spawned SnipSVG[1131] because launch job demand 2024-11-07 08:43:02.845832 (pid/1131 [SnipSVG]) <Notice>: uncorking exec source upfront 2024-11-07 08:43:02.845882 (pid/1131 [SnipSVG]) <Notice>: created 2024-11-07 08:43:03.355264 (pid/1131 [SnipSVG]) <Notice>: shutting down 2024-11-07 08:43:03.355286 (pid/1131 [SnipSVG]) <Notice>: cleaning up 2024-11-07 08:43:03.355364 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: exited due to exit(217), ran for 567ms 2024-11-07 08:43:03.355370 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: service state: exited 2024-11-07 08:43:03.355381 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: internal event: EXITED, code = 0 2024-11-07 08:43:03.355386 (gui/501/application.SnipSVG.16305011.16305018 [1131]) <Notice>: job state = exited ... My entitlements list:

The error reporting solution from JustAddCode that you mention in your tutorial does work reasonably well. While the line numbers are pretty odd, it at least delivers the function name. I was able to solve the problem on macOS with this tool.

Bringing up this topic again... I have managed to get a stack trace report on macOS, but I am not entirely sure what to do with it. There is a Windows app called "addr2line.exe", which I assume is supposed to parse the file that contains the debug symbols (for macOS that would be MyAppName.dSYM, right?). Then I have to feed it the addresses that I got from the error log. And it should return unit and line number where the error actually occurred. However, addr2line.exe doesn't seem to understand the format of the .dSYM file. Is this because it expects a Windows executable? Probably yes. Do I have to do this on the Mac with a macOS version of this tool? If yes, which is it? My stack trace looks like this: OS Version: macOS (Version 11.6.7) Product Version: 1.0.0 Product Name: FmxApp Platform: OSXARM64 Config: Debug DateTime: 31_10_24 10_41_00 Log-File: FmxApp-31_10_24-10_41_00.log Stack trace: 00000001024C8000:000000010254298C: FmxApp 00000001024C8000:0000000102D1EC24: FmxApp 00000001024C8000:0000000102A20750: FmxApp 00000001024C8000:0000000102A85428: FmxApp 00000001024C8000:0000000102A20D98: FmxApp 00000001024C8000:0000000102C8A50C: FmxApp 00000001024C8000:0000000102C2A308: FmxApp 00000001024C8000:0000000102C2A66C: FmxApp 00000001024C8000:0000000102C17394: FmxApp 00000001024C8000:00000001024CB438: FmxApp 0000000192AA5000:0000000192C69DC8: AppKit 0000000192AA5000:0000000192C6915C: AppKit 0000000192AA5000:0000000192C68034: AppKit 0000000192AA5000:0000000192F35194: AppKit 0000000192AA5000:0000000192AD78A0: AppKit 00000001024C8000:00000001024CB260: FmxApp 00000001024C8000:0000000102C10CCC: FmxApp 00000001024C8000:0000000102C10C14: FmxApp 00000001024C8000:0000000102C7E464: FmxApp 00000001024C8000:0000000102D1EF24: FmxApp Modules: AppKit: FmxApp:304f03e2f8d59eb00757a868fa075d26 Title: Application Error Message: Access violation at address 0000000102D1EE30, accessing address 0000000000000000

The Windows D2D implementation in FMX in FMX.Canvas.D2D.pas always assumes a premultiplied alpha. This is probably the source of the problem. Anders Melander can perhaps comment on that with more expertise than I have. D2D1_ALPHA_MODE (dcommon.h) - Win32 apps | Microsoft Learn One option to solve it would be to write a TCustomBitmapCodec for raw PNG data and load/save your PNGs through this codec exclusively. If SKIA is an option for you, install SKIA and enable it for your app. SKIA effectively implements its own canvas and its own image codecs, leaving your PNG file as it is. I tested your example image with SKIA enabled and the bitmap stays the same, no matter how often it is loaded and saved. Obviously, the SKIA codecs do a better job.

IMO the gestures are designed for real touchscreens and not for touchpads. I believe this has never worked with touchpads. Anyway, I have implemented a solution for myself by patching FMX.Platform.Mac, which was already patched anyway to work around the Sonoma scaling bug on MacOS. This is certainly not for everyone, but works like a charm. It uses the Angle parameter for an alternative zoom gesture (Angle is used in rotation gestures, for zoom it's always zero). Your app needs to respond to this accordingly. Most importantly, it doesn't break regular zoom gestures coming from real touch screens. procedure TFMXViewBase.magnifyWithEvent(event: NSEvent); var ... begin ... if FGestureControl <> nil then begin LTouches := event.touchesMatchingPhase(NSTouchPhaseTouching, NSView(Super)); if LTouches.count >= 2 then begin LTouchesArray := LTouches.allObjects; LTouch := TNSTouch.Wrap(LTouchesArray.objectAtIndex(0)); LDeviceSize := LTouch.deviceSize; FEventInfo.Distance := 0; //reset the distance // Find the greatest distance between the touches. for I := 0 to LTouches.count - 2 do begin LTouch := TNSTouch.Wrap(LTouchesArray.objectAtIndex(I)); LPoint := LTouch.normalizedPosition; for J := 1 to LTouches.count - 1 do begin LTouch := TNSTouch.Wrap(LTouchesArray.objectAtIndex(J)); LPoint2 := LTouch.normalizedPosition; Distance := Round(Sqrt(Sqr(LPoint.x * LDeviceSize.width - LPoint2.x * LDeviceSize.width) + Sqr(LPoint.y * LDeviceSize.height - LPoint2.y * LDeviceSize.height))); if Distance > FEventInfo.Distance then FEventInfo.Distance := Distance; end; FEventInfo.GestureID := igiZoom; if Supports(FGestureControl, IGestureControl, GestureObj) then GestureObj.CMGesture(FEventInfo); FEventInfo.Flags := []; end end {ECS/ALEX} else if LTouches.count = 0 then begin FEventInfo.Distance := 0; FEventInfo.Angle := event.magnification; FEventInfo.GestureID := igiZoom; if Supports(FGestureControl, IGestureControl, GestureObj) then GestureObj.CMGesture(FEventInfo); FEventInfo.Flags := []; end; end {ECS/ALEX} else //send the message up the responder chain NSView(Super).magnifyWithEvent(event); end;

No, I haven't. But the magnifyWithEvent comes reliably and this is what Delphi actually evaluates. I am not that deep into native MacOS development, so I rather stay with the methods that Delphi already implements.

The problem is in FMX.Platform.Mac: procedure TFMXViewBase.magnifyWithEvent(event: NSEvent); var ... begin ... if FGestureControl <> nil then begin LTouches := event.touchesMatchingPhase(NSTouchPhaseTouching, NSView(Super)); {ECS/ALEX Here comes our problem: LTouches.count is zero, that's why the gesture event is not fired. I assume that we have to deal with "event.magnification" to determine "FEventInfo.Distance". } if LTouches.count >= 2 then begin LTouchesArray := LTouches.allObjects; LTouch := TNSTouch.Wrap(LTouchesArray.objectAtIndex(0)); LDeviceSize := LTouch.deviceSize; ... I am currently experimenting with the event.magnification value (which is a float value), to match FEventInfo.Distance in a way that the example from the official docs keeps working the way it promises, but doesn't. However, I have a hard time to believe that this has never worked. If Emba has dedicated examples, there must have been a time when this did work properly. It perhaps depends on the MacOS version. My own version of MacOS that I'm using for testing is relatively old (Big Sur), with a second Mac used by one of my colleagues running Sonoma. The gesture is not fired on either machine, but may have been working on OS versions < Big Sur. If it has never worked with touchpad gestures, it might have worked and still work with a real touch screen. Is anyone competent to comment on that - does the igiZoom gesture work as advertised with touch screen Mac (we don't have them here, so I cannot test this)?

Yes, rotate works as described in the docs. Zoom and pan do not. There's no event at all for these. The pan gesture (swiping up or down with 2 fingers) obviously gets translated into a mouse wheel message.

So, it's officially broken, despite documentation says the opposite? Gestures in FireMonkey - RAD Studio (embarcadero.com)