Issue with RGFW_window_blitSurface on macos

[derinil]

Hi, there appears to be an issue with RGFW_window_blitSurface on macos. I just now cloned this repo, ran make, and ran "./examples/surface/surface". Here is what it looks like (I removed the transparency here):

This is the patch I applied to fix it:

--- a/RGFW_nonpatch.h
+++ b/RGFW_nonpatch.h
@@ -12047,17 +12047,21 @@ RGFW_bool RGFW_createSurfacePtr(u8* data, i32 w, i32 h, RGFW_format format, RGFW
 void RGFW_surface_freePtr(RGFW_surface* surface) { RGFW_FREE(surface->native.buffer); }
 
 void RGFW_window_blitSurface(RGFW_window* win, RGFW_surface* surface) {
-	int minX = RGFW_MIN(win->w, surface->w);
-	int minY = RGFW_MIN(win->h, surface->h);
+	// int minX = RGFW_MIN(win->w, surface->w);
+	// int minY = RGFW_MIN(win->h, surface->h);
+    // In the example, surface is always size of monitor, so do this
+	int minX = surface->h;
+	int minY = surface->w;
 
-	RGFW_copyImageData(surface->native.buffer, surface->w, minY, surface->native.format, surface->data, surface->format);
+	// RGFW_copyImageData(surface->native.buffer, surface->w, minY, surface->native.format, surface->data, surface->format);
 
 	id image = ((id (*)(Class, SEL))objc_msgSend)(objc_getClass("NSImage"), sel_getUid("alloc"));
 	NSSize size = (NSSize){(double)surface->w, (double)surface->h};
 	image = ((id (*)(id, SEL, NSSize))objc_msgSend)((id)image, sel_getUid("initWithSize:"), size);
 
 	id rep  = NSBitmapImageRep_initWithBitmapData(&surface->native.buffer, minX, minY, 8, 4, true, false, "NSDeviceRGBColorSpace", 1 << 1, (u32)surface->w  * 4, 32);
-	RGFW_copyImageData(NSBitmapImageRep_bitmapData(rep), minX, minY , RGFW_formatRGBA8, surface->native.buffer, surface->native.format);
+	// No need to copy image twice I think
+    RGFW_copyImageData(NSBitmapImageRep_bitmapData(rep), minX, minY , RGFW_formatRGBA8, surface->data, surface->format);
 	((void (*)(id, SEL, id))objc_msgSend)((id)image, sel_getUid("addRepresentation:"), rep);
 
 	id contentView = ((id (*)(id, SEL))objc_msgSend)((id)win->src.window, sel_getUid("contentView"));

After fixing:

Now not exactly sure what is causing the issue, maybe some issue in the RGFW_copyImageData? Because ultimately that is the only real big difference. Or maybe the example is out of date?

I removed the first call to RGFW_copyImageData because it seemed a little redundant to me.

[derinil]

Now I just added back the first RGFW_copyImageData call, which still works so I guess minX and minY is the issue. But this happens when resizing:

some curious memory bug

[ColleagueRiley]

minX and minY exist because you don't want to copy the whole buffer if you are only rendering the window size.

[ColleagueRiley]

That's not how the example looks when I test it. What version of MacOS are you using for context?

[derinil]

m1 pro 14.4.1
Apple clang version 15.0.0 (clang-1500.3.9.4)
Target: arm64-apple-darwin23.4.0

[derinil]

did a fresh clone, ran make and ran the surface example again:

also the window dies of OOM.

[derinil]

This also fixes the sizing issue, but I guess only because surface w and h are smaller than w and h. I don't have much time right now to investigate

i32 w, h;
	RGFW_window_getSizeInPixels(win, &w, &h);
	printf("%d %d - %d %d\n", w, h, win->w, win->h);
	int minX = RGFW_MIN(w, surface->w);
	int minY = RGFW_MIN(h, surface->h);

[derinil]

RGFW_window_getSizeInPixels:

[ColleagueRiley]

I'm using MacOS 15.7 on a late 2012 MacMini

Can I have some information about your monitor? Perhaps the window size is being scaled by the OSX, which is why the window size isn't matching.. ...I'm currently using a TV as a monitor at 768p, so that would be why I'm not having the issue.

The crash/memory leak is worrying, I don't remember experiencing the issue, so maybe it was created after some change. I thought all the resources were freed with NSRelease at the end of the function but maybe I'm missing something.

I linked another issue that was open since both issues will require me to review this function, which I have been working on even before this issue was opened.

[derinil]

I'm using a macbook pro from 2021 with no external monitor. Seems like my monitor scaling is 2x seeing how the getSizeInPIxels returns exactly 2x win->w

[ColleagueRiley]

That's not scale exactly, that's pixel density. Retina displays have higher pixel density so the pixel ratio is multiplied by 2. That could be the source of the issue.

[randall-fulton]

@ColleagueRiley I also have this issue, only on the built-in Retina display. I've gotten around it in my project by adding calls to RGFW_window_getMonitor within the macOS implementation of RGFW_window_blitSurface, so that the higher DPI screen is handled correctly, without causing issues for standard screens..

I have been meaning to revisit that, and open an issue/PR here. Haven't thus far because it caused some crazy memory leaks that I had to fix with the band-aid fix of another NSAutoreleasePool. I'm sure there is probably a better way, without fetching the monitor every time, but haven't had time to dig in any deeper yet.

Here is the very rough fix I've been using to get by, in case it's helpful (though it sounds like you already understand the problem):

  void RGFW_window_blitSurface(RGFW_window* win, RGFW_surface* surface) {
+   id eventPool = objc_msgSend_class(objc_getClass("NSAutoreleasePool"), sel_registerName("alloc"));
+   eventPool = objc_msgSend_id(eventPool, sel_registerName("init"));
	  
	  RGFW_copyImageData(surface->data, surface->w, RGFW_MIN(win->h, surface->h), surface->native.format, surface->data, surface->format);
	  
    size_t depth = (surface->format >= RGFW_formatRGBA8) ? 4 : 3;
	  id image = ((id (*)(Class, SEL))objc_msgSend)(objc_getClass("NSImage"), sel_getUid("alloc"));
	  NSSize size = (NSSize){(double)surface->w, (double)surface->h};
	  image = ((id (*)(id, SEL, NSSize))objc_msgSend)((id)image, sel_getUid("initWithSize:"), size);
	  
-   int minX = RGFW_MIN(win->w, surface->w);
-   int minY = RGFW_MIN(win->h, surface->h);
+	  RGFW_monitor mon = RGFW_window_getMonitor(win);
+	  int minX = RGFW_MIN(win->w, surface->w) * mon.pixelRatio;
+	  int minY = RGFW_MIN(win->h, surface->h) * mon.pixelRatio;
	  
	  id rep  = NSBitmapImageRep_initWithBitmapData(&surface->data, minX, minY, 8, (i32)depth, (depth == 4), false, "NSDeviceRGBColorSpace", 1 << 1, (u32)surface->w  * (u32)depth, 8 * (u32)depth);
	  RGFW_copyImageData(NSBitmapImageRep_bitmapData(rep), minX, minY , RGFW_formatRGBA8, surface->data, surface->format);
	  ((void (*)(id, SEL, id))objc_msgSend)((id)image, sel_getUid("addRepresentation:"), rep);
	  
	  id contentView = ((id (*)(id, SEL))objc_msgSend)((id)win->src.window, sel_getUid("contentView"));
	  ((void (*)(id, SEL, BOOL))objc_msgSend)(contentView, sel_getUid("setWantsLayer:"), YES);
	  id layer = ((id (*)(id, SEL))objc_msgSend)(contentView, sel_getUid("layer"));
	  
	  ((void (*)(id, SEL, id))objc_msgSend)(layer, sel_getUid("setContents:"), (id)image);
	  ((void (*)(id, SEL, BOOL))objc_msgSend)(contentView, sel_getUid("setNeedsDisplay:"), YES);
	  
	  objc_msgSend_bool_void(eventPool, sel_registerName("drain"));
	  
	  NSRelease(rep);
	  NSRelease(image);
  }

[ColleagueRiley]

@randall-fulton Thanks for your input on the issue.

The fix you shown is the exact same as running the new RGFW_window_getSizeInPixels, so I'm glad we all came to the same conclusion.

without fetching the monitor every time

Preferably monitor support will be fixed to reduce fetching in general.

Haven't thus far because it caused some crazy memory leaks that I had to fix with the band-aid fix of another NSAutoreleasePool

Sadly Autoreleasepool seems necessary since I am releasing all the objects I allocate, so it could be some weird objective-c hidden allocation I can't free manually?

[ColleagueRiley]

This PR should fix all the issues mentioned.

#445