can we use upstream pthread instead of this library?

[ian-h-chamberlain]

The new newlib update from devkitPro seems to include an actual libpthread!!!

$ arm-none-eabi-nm   /opt/devkitpro/devkitARM/arm-none-eabi/lib/libsysbase.a | grep pthread | wc -l 
86

I can't quite tell if it works correctly or at all, but there is definitely some new code in the libsysbase library and we might need to readjust how this library works to play nicely with it (link errors etc.)

The implementation looks to be similar to what we have done in this library, but perhaps the syscalls are still stubs (and we could implement them here, but let newlib handle the business logic):

https://github.com/devkitPro/buildscripts/blob/cd5e224d6a14f7d32712ab10cfc08e0c6a2daea3/dkarm-eabi/patches/newlib-4.3.0.20230120.patch#L7741

There might also be some other stuff that relates to shim-3ds, it seems like a lot of new library surface area. Probably will take a while to crawl through all of it and figure out what we can use or need to change.

Looking at libnx (where I am guessing some of these changes came from) it seems they just implement __syscall_* in terms of the device-specific libraries: https://github.com/switchbrew/libnx/blob/master/nx/source/runtime/newlib.c#L172

Perhaps we could do the same in this lib instead.

[Meziu]

Those… don’t look like stubs 😳

Edit: funny how those features always come after our own implementations… clock_gettime is also a fundamental part of these pthread changes, and it came extremely shortly after our impl. Maybe someone is spying on us 👀

[ian-h-chamberlain]

No, they're not stubs exactly, but I see a lot of stuff like

if (!__has_syscall(thread_create))
	return ENOSYS;

I don't think libctru is providing those syscalls currently, so we would need to define them ourselves to get stuff working.

From trying to build a quick example, I think something in Rust runtime startup got broken by these changes (possibly related to thread locals), so we probably have some work to do, lol!

[Meziu]

To me the "hello-world" example works fine (updated ctru-sys to the latest bindings and all libraries). The only change is that those bindings don't include __getreent, which has now been moved to newlib, rather than libctru. Should we add that to ctru-sys or to libc?

Edit: I've tested multi-threading and mutexes as well but they don't seem to give any issues to me. The threading order is correctly run and the data is not corrupted. All this on pthread-3ds of course.

[WinterMute]

Those… don’t look like stubs 😳

Edit: funny how those features always come after our own implementations… clock_gettime is also a fundamental part of these pthread changes, and it came extremely shortly after our impl. Maybe someone is spying on us 👀

Those features were implemented for devkitA64 basically 4 years ago in 2019, see https://devkitpro.org/viewtopic.php?f=13&t=8891&p=16395&hilit=pthreads#p16395. We have simply consolidated a lot of the patching between devkitA64 and devkitARM to ease maintenance and feature parity in future releases.

I would be grateful if people could refrain from making thoughtless comments like this and have some respect for the work we do.

[Meziu]

I would be grateful if people could refrain from making thoughtless comments like this and have some respect for the work we do.

We respect your work immensely. It’s only thanks to your and your collaborators’ work that our project could even exist. I only meant that comment as a joke, of course you aren’t copying from us (if anything, it’s the opposite). I often don’t act serious when responding, and I was pretty excited about the changes.

I’m sorry if I said anything you found disrespectful. 😓