libui/darwin/main.m

// 20 april 2019
#import "uipriv_darwin.h"

@interface uiprivApplication : NSApplication
@end

@implementation uiprivApplication
@end

@interface uiprivApplicationDelegate : NSObject<NSApplicationDelegate>
@end

static uiprivApplication *uiprivApp;
static uiprivApplicationDelegate *uiprivAppDelegate;

@implementation uiprivApplicationDelegate

#if 0
- (NSApplicationTerminateReply)applicationShouldTerminate:(NSApplication *)app
{
	// for debugging
	NSLog(@"in applicationShouldTerminate:");
	if (uiprivShouldQuit()) {
		canQuit = YES;
		// this will call terminate:, which is the same as uiQuit()
		return NSTerminateNow;
	}
	return NSTerminateCancel;
}
#endif

- (BOOL)applicationShouldTerminateAfterLastWindowClosed:(NSApplication *)app
{
	return NO;
}

@end

static pthread_t mainThread;

bool uiprivSysInit(void *options, uiInitError *err)
{
	uiprivApp = [uiprivApplication sharedApplication];
	if (![NSApp isKindOfClass:[uiprivApplication class]])
		// TODO verify if Info.plist (both in a proper .app bundle and embedded into a standalone binary) can override this (or if that only happens if you use NSApplicationMain() instead), and, if so, add that tidbit to this message
		return uiprivInitReturnErrorf(err, "NSApp is not of type uiprivApplication; was likely already initialized beforehand");

	// don't check for a NO return; something (launch services?) causes running from application bundles to always return NO when asking to change activation policy, even if the change is to the same activation policy!
	// see https://github.com/andlabs/ui/issues/6
	[uiprivApp setActivationPolicy:NSApplicationActivationPolicyRegular];

	uiprivAppDelegate = [uiprivApplicationDelegate new];
	[uiprivApp setDelegate:uiprivAppDelegate];

	mainThread = pthread_self();
	return true;
}

void uiMain(void)
{
	if (!uiprivCheckInitializedAndThread())
		return;
	[uiprivApp run];
}

void uiQuit(void)
{
	if (!uiprivCheckInitializedAndThread())
		return;
	@autoreleasepool {
		NSEvent *e;

		[uiprivApp stop:uiprivApp];
		// stop: won't register until another event has passed; let's synthesize one
		// TODO instead of using NSApplicationDefined, create a private event type for libui internal use only; if we can't, set up data1 and data2 so that we can intercept and discard this event if it was accidentally triggered
		e = [NSEvent otherEventWithType:NSApplicationDefined
			location:NSZeroPoint
			modifierFlags:0
			timestamp:[[NSProcessInfo processInfo] systemUptime]
			windowNumber:0
			context:[NSGraphicsContext currentContext]
			subtype:0
			data1:0
			data2:0];
		[uiprivApp postEvent:e atStart:NO];		// let pending events take priority (this is what PostQuitMessage() on Windows does so we have to do it here too for parity; thanks to mikeash in irc.freenode.net/#macdev for confirming that this parameter should indeed be NO)
	}
}

// thanks to mikeash in irc.freenode.net/#macdev for suggesting the use of Grand Central Dispatch for this
void uiprivSysQueueMain(void (*f)(void *data), void *data)
{
	// dispatch_get_main_queue() is a serial queue so it will not execute multiple uiQueueMain() functions concurrently
	// the signature of f matches dispatch_function_t
	dispatch_async_f(dispatch_get_main_queue(), data, f);
}

bool uiprivSysCheckThread(void)
{
	return pthread_equal(pthread_self(), mainThread);
}

// Debugger() was deprecated in macOS 10.8 (as part of the larger CarbonCore deprecation), but they did not provide a replacement.
// Though some people say inline asm, I'd rather make this work automatically everywhere.
// Others say raise(SIGTRAP) and others still say __builtin_trap(), but I can't confirm these do what I want (some sources, including documentation, claim they either cause a core dump or send a SIGILL instead).
// I've also heard of some new clang intrinsics, __builtin_debugtrap(), but this is totally undocumented and the original patch for this suggested it would be identical to __builtin_trap(), so...
// Also I cannot figure out how to manually raise EXC_BREAKPOINT.
// So that leaves us with one option: just use Debugger(), turning off the deprecation warnings.
// Also, while we could turn off the deprecation warning temporarily in gcc >= 4.6 and any clang, I'm not sure what minimum version of gcc we need, and I'm not sure if "any clang" is even accurate.
// So instead of juggling version macros, just turn off deprecation warnings at the bottom of this file.
// References:
// - https://stackoverflow.com/questions/37299/xcode-equivalent-of-asm-int-3-debugbreak-halt
// - https://stackoverflow.com/questions/2622017/suppressing-deprecated-warnings-in-xcode
// - https://stackoverflow.com/questions/28166565/detect-gcc-as-opposed-to-msvc-clang-with-macro
// - https://stackoverflow.com/questions/16555585/why-pragma-gcc-diagnostic-push-pop-warning-in-gcc-c
// - possibly others, all on stackoverflow.com (and maybe once on Apple's own forums?); I forget now
static void debugBreak(void);

void uiprivReportError(const char *prefix, const char *msg, const char *suffix, bool internal)
{
	NSExceptionName exceptionName;

	NSLog(@"*** %s: %s. %s", prefix, msg, suffix);
	exceptionName = NSInternalInconsistencyException;
	if (!internal)
		// TODO either find an appropriate exception for each possible message or use a custom exception name
		exceptionName = NSInvalidArgumentException;
	[NSException raise:exceptionName
		format:@"%s: %s", prefix, msg];
	debugBreak();
	abort();		// we shouldn't reach here
}

#ifdef __clang__
#pragma clang diagnostic ignored "-Wdeprecated-declarations"
#else
#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
#endif
static void debugBreak(void)
{
	Debugger();
}

// DO NOT ADD NEW CODE HERE. IT WILL NOT BE SUBJECT TO DEPRECATION WARNINGS.
// I am making an exception here with Debugger(); see the large comment above.