Zeta

Requiring a method specifically called `close` seems unflexible. What about Go-style defer: ```hx final foo = new Bar(); defer foo.destroy(); mutex.lock(); defer mutex.unlock(); ``` Pros: not restricted to close method,...

> Close for a file stream should be for closing file descriptors and other such cleanup which shouldn't fail Actually closing a file descriptor can return an error (though on...

Here's a suggestion: ```hx class Foo { public function new() {} @:cleanup public function myCustomDestructor() {} } class Bar { public function new() {} public function closeWithoutMeta() {} } @:cleanup...

Making `Lambda` work with both would be relatively simple, for example using an abstract with implicit conversions from both Iterable and Iterator: https://try.haxe.org/#8e6b3e3d

Hmm. If I try `haxe.Http.requestUrl("https://raw.githubusercontent.com/HaxeFoundation/haxe/development/tests/unit/res1.txt")` it works, if I try with `https://build.haxe.org/builds/haxe/linux64/haxe_latest.tar.gz` I get a segfault. As for that EBADF issue, the cause would be this function being called with...

That `param0` thing is needed for that function to be able to declare `value` locals, but since it doesn't take any `value` parameters it doesn't need to use `CAMLparam3`. https://ocaml.org/manual/5.1/intfc.html#sec503

Thinking about it some more, the real problem is probably that `ctx` is stored somewhere the GC can't see.

Works fine on my machine, with or without the workaround. (Arch Linux, compiler linked with mbedtls 3)

Yes, I enabled https tests. Does the core dump give a useful backtrace?

Regarding the HL+macOS failure, I don't think it's caused by GC issues or stuff like that. I did a little bit of debugging using CI (https://github.com/Apprentice-Alchemist/hashlink/actions/runs/8907771743/job/24462210850) The error is caused...