[PATCH] D22943: [Driver] Add FIXME's where we can't use effective triples (NFC)

[Joerg Sonnenberger via cfe-commits]

On Mon, Aug 01, 2016 at 08:50:59AM -0700, Vedant Kumar wrote:
> 
> > On Jul 30, 2016, at 12:59 PM, Joerg Sonnenberger via cfe-commits <cfe-commits at lists.llvm.org> wrote:
> > 
> > On Thu, Jul 28, 2016 at 10:11:05PM +0000, Vedant Kumar wrote:
> >>  - test/Driver/netbsd.c
> >> 
> >>    We see -mcpu=arm1022e instead of arm926ej-s in a single run.
> > 
> > Which one exactly?
> 
> Here is the test (test/Driver/netbsd.c @ Line 175):
> 
> > // ARM: clang{{.*}}" "-cc1" "-triple" "armv5e--netbsd-eabi"
> > // ARM: as{{.*}}" "-mcpu=arm926ej-s" "-o"
> > // ARM: ld{{.*}}" "--eh-frame-hdr" "-dynamic-linker" "/libexec/ld.elf_so"
> > // ARM: "-m" "armelf_nbsd_eabi"
> > // ARM: "-o" "a.out" "{{.*}}/usr/lib{{/|\\\\}}crt0.o"
> > // ARM: "{{.*}}/usr/lib{{/|\\\\}}eabi{{/|\\\\}}crti.o"
> > // ARM: "{{.*}}/usr/lib{{/|\\\\}}crtbegin.o" "{{.*}}.o" "-lc"
> > // ARM: "{{.*}}/usr/lib{{/|\\\\}}crtend.o" "{{.*}}/usr/lib{{/|\\\\}}crtn.o"
> 
> And here is the test output:
> 
> > /Users/vk/Desktop/llvm/tools/clang/test/Driver/netbsd.c:175:9: error: expected string not found in input
> > // ARM: as{{.*}}" "-mcpu=arm926ej-s" "-o"
> >         ^
> > <stdin>:5:81: note: scanning from here
> >  "/Users/vk/Desktop/llvm/DA/./bin/clang" "-cc1" "-triple" "armv5e--netbsd-eabi" "-S" "-disable-free" "-main-file-name" "netbsd.c" "-mrelocation-model" "static" "-mthread-model" "posix" "-mdisable-fp-elim" "-masm-verbose" "-no-integrated-as" "-mconstructor-aliases" "-munwind-tables" "-target-cpu" "arm1022e" "-target-feature" "+soft-float-abi" "-target-feature" "+strict-align" "-target-abi" "aapcs" "-mfloat-abi" "soft" "-target-linker-version" "272" "-dwarf-column-info" "-debugger-tuning=gdb" "-resource-dir" "/Users/vk/Desktop/llvm/DA/./bin/../lib/clang/4.0.0" "-isysroot" "/Users/vk/Desktop/llvm/tools/clang/test/Driver/Inputs/basic_netbsd_tree" "-fno-dwarf-directory-asm" "-fdebug-compilation-dir" "/Users/vk/Desktop/llvm/build/tools/clang/test/Driver" "-ferror-limit" "19" "-fmessage-length" "0" "-fallow-half-arguments-and-returns" "-fno-signed-char" "-fobjc-runtime=gnustep" "-fdiagnostics-show-option" "-o" "/var/folders/j5/4k_dcg5j5kl7z4fd3gp1pjp40000gn/T/netbsd-0d4ff3.s" "-x" "c" "/Users/vk/Desktop/llvm/tools/clang/test/Driver/netbsd.c"
> >                                                                                 ^
> > <stdin>:6:6: note: possible intended match here
> >  "/usr/bin/as" "-mcpu=arm1022e" "-o" "/var/folders/j5/4k_dcg5j5kl7z4fd3gp1pjp40000gn/T/netbsd-82d94c.o" "/var/folders/j5/4k_dcg5j5kl7z4fd3gp1pjp40000gn/T/netbsd-0d4ff3.s"
> 
> For context, the inconsistency arises when the driver uses
> ToolChain::ComputeLLVMTriple() instead of the default toolchain triple.

Hm. So we are hitting the no-architecture fall-through in
getARMCPUForArch for this test case before, but not after? That looks
suspicious. I'm not saying the new result is wrong, but that we should
investigate where the difference comes from. Sadly, I'm currently on
vacation, so I don't have time for that.

Joerg