Cheng Shao

https://gitlab.haskell.org/ghc/ghc/-/merge_requests/11697 should hopefully mitigate the need for `poll_oneoff` in the ghc wasm backend when it comes to stdin/stdout/stderr handles

It's still failing, but with a different error message: ``` Downloading backprop-0.2.6.4 Downloaded backprop-0.2.6.4 Configuring backprop-0.2.6.4... Preprocessing library for backprop-0.2.6.4.. Building library for backprop-0.2.6.4.. [ 1 of 10] Compiling Data.Type.Util...

I've reran the build, the log is available at https://gist.githubusercontent.com/TerrorJack/7f0e88c51b441a78b7e9805cc84fa874/raw/361a0917ffdb6237f13c5f370517aef880a66c10/codeworld.log

Another build error: https://gist.githubusercontent.com/TerrorJack/531c3be43d0e100d45b2bfad6633b18f/raw/4af713d1ac8d57e6847393f49f1e3b5de0ce46b7/gistfile1.txt

It's still failing, not ghcjs-built libraries this time, but in `codeworld-auth`: https://gist.githubusercontent.com/TerrorJack/f9618fbc752cffa7f529fb9b070fe34d/raw/514ec7bcd524bd08d68a1c8b1442d9187ef6bb05/codeworld.log

Even after installing `g++`, same error occurs. But using `g++` to compile it yields no error.

This will also hurt users that do compile their own entire sysroot with experimental mv abi and expect compiler-rt to not have in-memory overhead with i128s.

As long as `-mmultivalue -Xclang -target-abi -Xclang experimental-mv` does not regress from current behavior of always passing i128 via multi value (even for `compiler-rt` when compiled from source) then it's...

Ah sorry, I missed the `dlmalloc.c` file! Yes, `USE_LOCKS` does get defined. Though it seems that currently spin locks is used in dlmalloc, which looks not ideal compared to futex...

I think there's an even bigger problem there; IIUIC `pthread_mutex_lock` in `wasi-libc` currently uses spinlock. We do have a futex-based lock at `libc-top-half/musl/src/thread/__lock.c` which gets used in some places, but...