John Ericson (Mar 21 2022 at 21:35): John Ericson (Mar 21 2022 at 21:35):Anyone have a moment to take a look at the CI failure for https://github.com/WebAssembly/wasi-libc/pull/270 ? I was getting there is with my (admitting highly nonstandard) toolchain before my change. So I am a confused what's up.
John Ericson (Mar 21 2022 at 21:36):I think I am using -isysroot wrong, but I do not understand how the --sysroot prior worked either the more I read the sparse GCC and Clang documentation on this. I would like to replace both with a plain old -I, but then I get another spooky error.
John Ericson (Mar 21 2022 at 21:37):In particular __strong_reference is only used in _Exit.c and nanosleep.c, and those files don't seem to (before or after) get the include needed for __strong_reference's CPP definition, and yet it worked before.
Dan Gohman (Mar 21 2022 at 21:38):"-isysroot "/home/runner/work/wasi-libc/wasi-libc/sysroot/include"
Dan Gohman (Mar 21 2022 at 21:39):should this be " -isysroot "/home/runner/work/wasi-libc/wasi-libc/sysroot", without the trailing include? the docs for -isysroot say it works like -sysroot, which suggests it inteprrets the path as the root of the sysroot
John Ericson (Mar 21 2022 at 21:39):So I think "-isysroot "/home/runner/work/wasi-libc/wasi-libc/sysroot" would work for CI.
John Ericson (Mar 21 2022 at 21:40):but the docs also talk about things like $sysroot/usr/include. And neither works in my local build! I think sysroot is very magic in that the paths that it looks for within the sysroot are somehow configurable.
Dan Gohman (Mar 21 2022 at 21:41):hrm, I've not encountered things looking for "usr" inside sysroot. That's... unfortunate.
John Ericson (Mar 21 2022 at 21:42):So yeah in conclusion it would be nice to just use plain old -I
John Ericson (Mar 21 2022 at 21:42):an avoid all that spooky ambient configuration that effects what exactly "sysroot" things do.
Dan Gohman (Mar 21 2022 at 21:43):Yeah, or -isystem so that it does the right thing wrt <....h> includes
John Ericson (Mar 21 2022 at 21:43):But then, I get the __strong_reference error. In fact, -I ... -isysroot together also causes the __strong_reference error! madness!
Dan Gohman (Mar 21 2022 at 21:43):what is the __strong_reference error?
John Ericson (Mar 21 2022 at 21:43):Dan Gohman said:
Yeah, or
-isystemso that it does the right thing wrt<....h>includes
What exactly do you mean? <...> vs ".."?
Dan Gohman (Mar 21 2022 at 21:44):Yeah. -isystem is like -I but marks the directory as a "system" include directory.
John Ericson (Mar 21 2022 at 21:44):John Ericson said:
In particular
__strong_referenceis only used in_Exit.candnanosleep.c, and those files don't seem to (before or after) get the include needed for__strong_reference's CPP definition, and yet it worked before.
It doesn't get the macro def and at least one of those two files fail to build.
John Ericson (Mar 21 2022 at 21:44):Dan Gohman said:
Yeah.
-isystemis like-Ibut marks the directory as a "system" include directory.
there is also -Iquote and -isystem.
John Ericson (Mar 21 2022 at 21:45):I will try some of those
Dan Gohman (Mar 21 2022 at 21:45):Include directories provided by wasi-libc should be treated as "system" directories.
John Ericson (Mar 21 2022 at 21:46):OK -isystem + -isysroot worked!
John Ericson (Mar 21 2022 at 21:46):Now trying just -isystem.
John Ericson (Mar 21 2022 at 21:46):-I does affect <...> lookup too. I think Postel's law muddled the distinctions long ago :) but I will check the docs.
Dan Gohman (Mar 21 2022 at 21:47):Yeah I forget the exact rules because srsly
John Ericson (Mar 21 2022 at 21:49):OK! -isystem alone works, and -idirafter also works.
John Ericson (Mar 21 2022 at 21:50):So the issue wasn't "systemness" but order --- that the other -I must override this. Yuck!!!
John Ericson (Mar 21 2022 at 22:03):Wrote a big fat comment and pushed :).
Last updated: Dec 06 2025 at 06:05 UTC