Build failure: `hostinfo.h: error: unknown type name 'X86Features'; did you mean 'X86Feature'?`

[barracuda156]

Build fails on powerpc:

[ 59%] Building C object CMakeFiles/c3c.dir/src/compiler/sema_asm.c.o
/opt/local/bin/ccache /opt/local/bin/gcc-mp-14 -DCURL_FOUND=1 -DLLVM_AVAILABLE=0 -DTB_AVAILABLE=0 -I/opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/src -I/opt/local/var/macports/build/c3c-7b939bf2/work/build -I/opt/local/include -I/opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/dependencies/miniz -pipe -Os -DNDEBUG -I/opt/local/include -std=gnu11 -arch ppc -mmacosx-version-min=10.6 -gdwarf-3 -fno-exceptions -pthread -Wall -Werror -Wno-unknown-pragmas -Wno-unused-result -Wno-unused-function -Wno-unused-variable -Wno-unused-parameter -MD -MT CMakeFiles/c3c.dir/src/compiler/sema_asm.c.o -MF CMakeFiles/c3c.dir/src/compiler/sema_asm.c.o.d -o CMakeFiles/c3c.dir/src/compiler/sema_asm.c.o -c /opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/src/compiler/sema_asm.c
In file included from /opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/src/compiler/target.c:10:
/opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/src/utils/hostinfo.h:4:28: error: unknown type name 'X86Features'; did you mean 'X86Feature'?
    4 | void hostinfo_x86_features(X86Features *cpu_features);
      |                            ^~~~~~~~~~~
      |                            X86Feature
/opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/src/compiler/target.c: In function 'x86_features_from_host':
/opt/local/var/macports/build/c3c-7b939bf2/work/c3c-0.7.8/src/compiler/target.c:969:9: error: implicit declaration of function 'hostinfo_x86_features' [-Wimplicit-function-declaration]
  969 |         hostinfo_x86_features(cpu_features);
      |         ^~~~~~~~~~~~~~~~~~~~~
make[2]: *** [CMakeFiles/c3c.dir/src/compiler/target.c.o] Error 1

[barracuda156]

X86Features is just not defined anywhere in the code.

[ManuLinares]

Can you confirm if changing this line: https://github.com/c3lang/c3c/blob/master/src/utils/hostinfo.h#L4
to void hostinfo_x86_features(CpuFeatures *cpu_features); works?

new hostinfo.h

#if !defined(HOSTINFO_H) && !LLVM_AVAILABLE
#define HOSTINFO_H 

void hostinfo_x86_features(CpuFeatures *cpu_features);
ArchType hostinfo_arch_type(void);
EnvironmentType hostinfo_env_type(void);
OsType hostinfo_os_type(void);
VendorType hostinfo_vendor_type(void);
const char * hostinfo_default_triple(void);
const char * hostinfo_x86_cpu_name(void); // for example: "x86-64", "x86-64-v4"

#endif 

[ManuLinares]

Out of topic, can I ask what is your exact hardware? and is it used broadly in your region?
Thanks.

[barracuda156]

Thank you, I will try.

P. S. I am looking through the sources, and we will need quite a bit of additions to support powerpc-darwin. Almost everywhere it is assumed that macOS is x86_64/arm64, so either the build gonna fail or codegen will be broken.

[barracuda156]

Out of topic, can I ask what is your exact hardware?

This is on powerpc (G5 specifically, but that is irrelevant). The codepath was triggered because there is no working and recent enough LLVM for the target.

[lerno]

PowerPC is unlikely to be supported except in the future for cross compilation, not as a compiler target.

[lerno]

There is so much work in different areas to possibly support PowerPC, and it such old hardware at this point that it's likely better to hand this off to a C backend rather than trying to use LLVM's cross compilation even.

[barracuda156]

it's likely better to hand this off to a C backend rather than trying to use LLVM's cross compilation even.

@lerno Oh, absolutely, I was not planning to add any support for LLVM for it. Could you say what is needed to use C backend for it? Does c3c need to handle the ABI on its own, or is it passed on to gcc?

[lerno]

Note that the C backend is pushed rather far into the future, because I would like to simplify the work needed to implement a backend before the C backend gets done to reduce the effort. For the C backend, ABI would be handled by the C compiler.

[barracuda156]

Note that the C backend is pushed rather far into the future

@lerno Ah, so at the moment it is just pointless without LLVM – it cannot possibly work? Until there is some codegen, adding support for an arch makes little sense, I won’t even be able to test anything. And re-writing LLVM support from the modern one to LLVM 7 (the only version currently working on ppc) is not something I intend to do: even if feasible, it will likely be too time-consuming.