Triples: Environment vs Object format

[mintsuki]

Reproducer:

#include <cstdio>
#include <llvm/ADT/StringRef.h>
#include <llvm/ADT/Twine.h>
#include <llvm/TargetParser/Triple.h>
#include <llvm/TargetParser/TargetParser.h>

#define props(X) \
    X(hasEnvironment) \
    X(isArch64Bit) \
    X(isArch32Bit) \
    X(isArch16Bit) \
    X(isMacOSX) \
    X(isiOS) \
    X(isTvOS) \
    X(isWatchOS) \
    X(isWatchABI) \
    X(isXROS) \
    X(isDriverKit) \
    X(isOSzOS) \
    X(isOSDarwin) \
    X(isSimulatorEnvironment) \
    X(isMacCatalystEnvironment) \
    X(isTargetMachineMac) \
    X(isOSNetBSD) \
    X(isOSOpenBSD) \
    X(isOSFreeBSD) \
    X(isOSFuchsia) \
    X(isOSDragonFly) \
    X(isOSSolaris) \
    X(isOSIAMCU) \
    X(isOSUnknown) \
    X(isGNUEnvironment) \
    X(isOSHaiku) \
    X(isUEFI) \
    X(isOSWindows) \
    X(isKnownWindowsMSVCEnvironment) \
    X(isWindowsMSVCEnvironment) \
    X(isWindowsArm64EC) \
    X(isWindowsCoreCLREnvironment) \
    X(isWindowsItaniumEnvironment) \
    X(isWindowsCygwinEnvironment) \
    X(isWindowsGNUEnvironment) \
    X(isOSCygMing) \
    X(isOSMSVCRT) \
    X(isOSNaCl) \
    X(isOSLinux) \
    X(isOSKFreeBSD) \
    X(isOSHurd) \
    X(isOSWASI) \
    X(isOSEmscripten) \
    X(isOSGlibc) \
    X(isOSAIX) \
    X(isOSSerenity) \
    X(isOSBinFormatELF) \
    X(isOSBinFormatCOFF) \
    X(isOSBinFormatGOFF) \
    X(isOSBinFormatMachO) \
    X(isOSBinFormatWasm) \
    X(isOSBinFormatXCOFF) \
    X(isOSBinFormatDXContainer) \
    X(isPS4) \
    X(isPS5) \
    X(isPS) \
    X(isAndroid) \
    X(isMusl) \
    X(isOHOSFamily) \
    X(isOpenHOS) \
    X(isOSLiteOS) \
    X(isDXIL) \
    X(isShaderModelOS) \
    X(isVulkanOS) \
    X(isShaderStageEnvironment) \
    X(isSPIR) \
    X(isSPIRV) \
    X(isSPIRVLogical) \
    X(isNVPTX) \
    X(isAMDGCN) \
    X(isAMDGPU) \
    X(isThumb) \
    X(isARM) \
    X(isTargetEHABICompatible) \
    X(isArmT32) \
    X(isArmMClass) \
    X(isAArch64) \
    X(isLoongArch32) \
    X(isLoongArch64) \
    X(isLoongArch) \
    X(isMIPS32) \
    X(isMIPS64) \
    X(isMIPS) \
    X(isPPC) \
    X(isPPC32) \
    X(isPPC64) \
    X(isPPC64ELFv2ABI) \
    X(isPPC32SecurePlt) \
    X(isRISCV32) \
    X(isRISCV64) \
    X(isRISCV) \
    X(isSPARC32) \
    X(isSPARC64) \
    X(isSPARC) \
    X(isSystemZ) \
    X(isX86) \
    X(isVE) \
    X(isWasm) \
    X(isCSKY) \
    X(isArm64e) \
    X(isX32) \
    X(isBPF) \
    X(supportsCOMDAT) \
    X(hasDefaultEmulatedTLS) \
    X(hasDefaultTLSDESC) \
    X(hasDefaultDataSections) \
    X(hasDLLImportExport) \
    X(isLittleEndian)

#define handleProp(P) if (triple.P()) printf(" %s", #P);

const char* rstr(llvm::StringRef r) {
    if (r.size() == 0) return "(empty)";
    char* dst = (char*)malloc(r.size() + 1);
    memcpy(dst, r.begin(), r.size());
    dst[r.size()] = 0;
    return dst;
}

int main(int argc, char** argv) {
    if (argc != 2) {
        printf("Usage: %s <triple>\n", argv[0]);
        return 1;
    }
    std::string normal = llvm::Triple::normalize(argv[1]);
    printf("normalize %s -> %s\n", argv[1], normal.c_str());
    llvm::Triple triple{normal};
    printf("arch: %s\n", rstr(triple.getArchTypeName(triple.getArch())));
    printf("vendor: %s\n", rstr(triple.getVendorTypeName(triple.getVendor())));
    printf("OS: %s\n", rstr(triple.getOSTypeName(triple.getOS())));
    printf("env: %s\n", rstr(triple.getEnvironmentTypeName(triple.getEnvironment())));
    printf("object format: %s\n", rstr(triple.getObjectFormatTypeName(triple.getObjectFormat())));
    props(handleProp);
    printf("\n");
}

Hello,

Consider a target triple such as x86_64-unknown-unknown-elf.

elf will be considered the object format for the final, normalised Triple object, but hasEnvironment() will return true, despite there not being a clear environment set.

-elf does not set a valid environment, as one can see from triple.getEnvironmentTypeName(triple.getEnvironment()) returning unknown. I understand that the environment and object format fields go in the same component of the triple, but in this case there is only an object format, not an environment.

Is this a misunderstanding of mine, or is this a bug with the normalisation/parsing of triples? In my opinion, hasEnvironment() should return false as -elf does not contain an environment at all, but rather just an object format.

[mintsuki]

If the environment is considered to be the triple component as a whole, so, including the object format, if any, and if that is the intended behaviour, then I'd say the loongarch64 function computeTargetABI() should be changed to not rely on hasEnvironment(), but, rather, to check if there is a non-unknown environment set, for example.

Without this change, using a (ideally valid) target of loongarch64-unknown-none-elf, with a manually specified ABI of lp64s, will result in a completely superfluous warning:

warning: triple-implied ABI conflicts with provided target-abi 'lp64s', using target-abi

[to268]

CC: @MaskRay