Parse error with WASI errno types

[juj]

Updating our codebases to new Wasm backend... Here is a silly issue.

On some platforms, the C errno defines EAGAIN and EWOULDBLOCK can have the same value. Hence, portable code in our codebase does

switch(errno) {
   case EAGAIN:
#if EAGAIN != EWOULDBLOCK
   case EWOULDBLOCK:
#endif
      // handle EAGAIN/EWOULDBLOCK here
}

But attempting to compile such code with new Emscripten code gives an error

os\Posix\Error.cpp:30:9: error: token is not a valid binary operator in a preprocessor subexpression
    #if EAGAIN != EWOULDBLOCK
        ^~~~~~
C:\code\emsdk\emscripten\incoming\system\lib\libc\musl\arch\emscripten\bits\errno.h:13:28: note: expanded from macro 'EAGAIN'
#define EAGAIN             __WASI_ERRNO_AGAIN
                           ^~~~~~~~~~~~~~~~~~
C:\code\emsdk\emscripten\incoming\system\include\wasi\api.h:129:45: note: expanded from macro '__WASI_ERRNO_AGAIN'
#define __WASI_ERRNO_AGAIN ((__wasi_errno_t)6)
                            ~~~~~~~~~~~~~~~~^

Changing

diff --git a/system/include/wasi/api.h b/system/include/wasi/api.h
index 137fc264d..7a928aa9f 100644
--- a/system/include/wasi/api.h
+++ b/system/include/wasi/api.h
@@ -126,7 +126,7 @@ typedef uint16_t __wasi_errno_t;
 /**
  * Resource unavailable, or operation would block.
  */
-#define __WASI_ERRNO_AGAIN ((__wasi_errno_t)6)
+#define __WASI_ERRNO_AGAIN 6

 /**
  * Connection already in progress.

resolves the build error, but not sure if that makes sense.

I don't know if spec mandates that testing #if EAGAIN != EWOULDBLOCK should be safe or not, but that line has not caused issues on any other platform (and Unity builds to a ton of platforms), so I'm wondering if Emscripten can change here? Is the (__wasi_errno_t) cast somehow important?

[kripken]

This looks like #9996, and overall I am starting to think we should just modify the wasi headers locally in emscripten... what do you think @sbc100 @Brion ?

[bvibber]

A local change should be an easy fix, but will be a maintenance burden later. I'd recommend seeing if wasi would be ok with changing the #defines in their upstream headers to improve portability.

[sbc100]

Closing this as dup of #9996.