Exception stack misses actual throw position

[Simn]

import haxe.Exception;

function f1() {
	throw new Exception("f1");
}

function f2() {
	f1();
}

function main() {
	try {
		f2();
	} catch(e:Exception) {
		for (item in e.stack.asArray()) {
			trace(item);
		}
	}
}

HL:

source/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,f2),source/Main.hx,8,null)
source/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,main),source/Main.hx,13,null)
source/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,main),source/Main.hx,12,null)

All other targets (that I've tried) have the original FilePos(Method(_Main.Main_Fields_,f1),source/Main.hx,4,null) as part of the stack.

I'm also just now seeing the line 12 entry, which seems to leak some internal exception handling mechanism because that's on the try.

[ncannasse]

Yes I had something similar recently. It's definitely a problem on the Haxe side that does increase some ignore stack counter too much at some places.
We should have proper unit tests with various cases with rethrow etc to make sure everything works as intended everywhere.

[Simn]

I'll be testing that properly over at hxcoro. It's focused on the async stacks but the normal stacks will be part of these tests as well.

One detail that has come up there: it's actually fine on Linux, so it's specific to Windows and Mac. This made me think it can't be only the Haxe side to blame, but I haven't investigated this further yet.

[Simn]

Yuxiao confirmed that this indeed works correctly on Linux, so we all think this is likely something at the Hashlink VM-level.

[ncannasse]

@yuxiaomao usually that means that hl_throw does does not push rbp or rsp before throwing. So we can't find a (rip+rsp) séquence on the stack that marks a call.

[yuxiaomao]

Note: On windows even C exception stack is strange (same result with -O2 and -Od)

src/Main.hx:16: Module(fun$init(init.c:402))
src/Main.hx:16: Module(fun$init(init.c:402))
src/Main.hx:16: Module(fun$init(init.c:402))

[yuxiaomao]

Note2: (copy-paste from my comment on copilot's PR)

I don't know why exactly but, the detection works slightly better on Windows if I use ((void**)&stack)-1

// module.c
static int module_capture_stack( void **stack, int size ) {
	return hl_module_capture_stack_range(hl_get_thread()->stack_top, ((void**)&stack)-1, stack, size);
}

produces

src/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,f1),src/Main.hx,4,null)
src/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,f2),src/Main.hx,8,null)
src/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,main),src/Main.hx,13,null)
src/Main.hx:16: FilePos(Method(_Main.$Main_Fields_,main),src/Main.hx,12,null)