Add a build container for thumbv7neon-unknown-linux-gnueabihf

[hsivonen]

The thumbv7neon-unknown-linux-gnueabihf target was added recently. However, rust-std for the target isn't available via rustup, because there's no infra automation for building it.

A container configuration for actually building release artifacts for the target is needed.

[hsivonen]

It unclear to me what the right fix is. It seems that the only Docker config that talks about armv7_unknown_linux_gnueabihf (the closest existing config) at all is src/ci/docker/dist-armv7-linux/Dockerfile. However, its format seems to differ from the format of containers that build multiple targets.

[hsivonen]

Changing the src/ci/docker/dist-armv7-linux/Dockerfile like this:

ENV CC_armv7_unknown_linux_gnueabihf=armv7-unknown-linux-gnueabihf-gcc \
    AR_armv7_unknown_linux_gnueabihf=armv7-unknown-linux-gnueabihf-ar \
    CXX_armv7_unknown_linux_gnueabihf=armv7-unknown-linux-gnueabihf-g++

ENV CC_thumbv7neon-unknown-linux-gnueabihf=armv7-unknown-linux-gnueabihf-gcc \
    AR_thumbv7neon-unknown-linux-gnueabihf=armv7-unknown-linux-gnueabihf-ar \
    CXX_thumbv7neon-unknown-linux-gnueabihf=armv7-unknown-linux-gnueabihf-g++

ENV HOSTS=armv7-unknown-linux-gnueabihf,thumbv7neon-unknown-linux-gnueabihf

Fails like this:

+ python2.7 ../x.py dist --host armv7-unknown-linux-gnueabihf,thumbv7neon-unknown-linux-gnueabihf --target armv7-unknown-linux-gnueabihf,thumbv7neon-unknown-linux-gnueabihf
    Finished dev [unoptimized] target(s) in 0.19s
thread 'main' panicked at '

couldn't find required command: "arm-linux-gnueabihf-gcc"

', src/bootstrap/sanity.rs:65:13
note: Run with `RUST_BACKTRACE=1` for a backtrace.
failed to run: /checkout/obj/build/bootstrap/debug/bootstrap dist --host armv7-unknown-linux-gnueabihf,thumbv7neon-unknown-linux-gnueabihf --target armv7-unknown-linux-gnueabihf,thumbv7neon-unknown-linux-gnueabihf
Build completed unsuccessfully in 0:00:00

[hsivonen]

ENV CC_armv7_unknown_linux_gnueabihf=armv7-unknown-linux-gnueabihf-gcc \
    AR_armv7_unknown_linux_gnueabihf=armv7-unknown-linux-gnueabihf-ar \
    CXX_armv7_unknown_linux_gnueabihf=armv7-unknown-linux-gnueabihf-g++

ENV CC_thumbv7neon-unknown-linux-gnueabihf=armv7-unknown-linux-gnueabihf-gcc \
    AR_thumbv7neon-unknown-linux-gnueabihf=armv7-unknown-linux-gnueabihf-ar \
    CXX_thumbv7neon-unknown-linux-gnueabihf=armv7-unknown-linux-gnueabihf-g++

ENV HOSTS=armv7-unknown-linux-gnueabihf

ENV TARGETS=armv7-unknown-linux-gnueabihf,thumbv7neon-unknown-linux-gnueabihf

At least doesn't fail immediately...

[hsivonen]

Trying to use comma-separated targets seems to break even the target that was unchanged by making backtrace-sys build with a mix of ARM and x86_64 inputs.

Building stage2 tool clippy-driver (armv7-unknown-linux-gnueabihf)
   Compiling backtrace-sys v0.1.24
   Compiling memchr v2.1.1
   Compiling ryu v0.2.7
   Compiling idna v0.1.5
   Compiling smallvec v0.6.7
   Compiling pulldown-cmark v0.2.0
   Compiling clippy v0.0.212 (/checkout/src/tools/clippy)
   Compiling toml v0.4.10
error: failed to run custom build command for `backtrace-sys v0.1.24`
process didn't exit successfully: `/checkout/obj/build/x86_64-unknown-linux-gnu/stage2-tools/release/build/backtrace-sys-0483c43875de1fe5/build-script-build` (exit code: 101)
--- stdout
TARGET = Some("x86_64-unknown-linux-gnu")
OPT_LEVEL = Some("2")
HOST = Some("x86_64-unknown-linux-gnu")
CC_x86_64-unknown-linux-gnu = None
CC_x86_64_unknown_linux_gnu = None
HOST_CC = None
CC = Some("sccache armv7-unknown-linux-gnueabihf-gcc")
CFLAGS_x86_64-unknown-linux-gnu = None
CFLAGS_x86_64_unknown_linux_gnu = None
HOST_CFLAGS = None
CFLAGS = Some("-ffunction-sections -fdata-sections -fPIC -march=armv7-a")
DEBUG = Some("false")
running: "sccache" "armv7-unknown-linux-gnueabihf-gcc" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-ffunction-sections" "-fdata-sections" "-fPIC" "-march=armv7-a" "-m64" "-I" "src/libbacktrace" "-I" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage2-tools/release/build/backtrace-sys-b7b7de2d05114cff/out" "-fvisibility=hidden" "-DBACKTRACE_ELF_SIZE=64" "-DBACKTRACE_SUPPORTED=1" "-DBACKTRACE_USES_MALLOC=1" "-DBACKTRACE_SUPPORTS_THREADS=0" "-DBACKTRACE_SUPPORTS_DATA=0" "-DHAVE_DL_ITERATE_PHDR=1" "-D_GNU_SOURCE=1" "-D_LARGE_FILES=1" "-Dbacktrace_full=__rbt_backtrace_full" "-Dbacktrace_dwarf_add=__rbt_backtrace_dwarf_add" "-Dbacktrace_initialize=__rbt_backtrace_initialize" "-Dbacktrace_pcinfo=__rbt_backtrace_pcinfo" "-Dbacktrace_syminfo=__rbt_backtrace_syminfo" "-Dbacktrace_get_view=__rbt_backtrace_get_view" "-Dbacktrace_release_view=__rbt_backtrace_release_view" "-Dbacktrace_alloc=__rbt_backtrace_alloc" "-Dbacktrace_free=__rbt_backtrace_free" "-Dbacktrace_vector_finish=__rbt_backtrace_vector_finish" "-Dbacktrace_vector_grow=__rbt_backtrace_vector_grow" "-Dbacktrace_vector_release=__rbt_backtrace_vector_release" "-Dbacktrace_close=__rbt_backtrace_close" "-Dbacktrace_open=__rbt_backtrace_open" "-Dbacktrace_print=__rbt_backtrace_print" "-Dbacktrace_simple=__rbt_backtrace_simple" "-Dbacktrace_qsort=__rbt_backtrace_qsort" "-Dbacktrace_create_state=__rbt_backtrace_create_state" "-Dbacktrace_uncompress_zdebug=__rbt_backtrace_uncompress_zdebug" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage2-tools/release/build/backtrace-sys-b7b7de2d05114cff/out/src/libbacktrace/alloc.o" "-c" "src/libbacktrace/alloc.c"
cargo:warning=armv7-unknown-linux-gnueabihf-gcc: error: unrecognized command line option '-m64'
exit code: 1

[hsivonen]

I'd appreciate input from someone who actually understands the system. (I'll be away from GitHub until January 7, though.)

[Centril]

cc @nagisa @japaric @alexcrichton

[alexcrichton]

@hsivonen this probably wants to change dist-various-{1,2} (one or the other) rather than --hosts, because we just want a standard library for this target rather than a full compiler

[hsivonen]

The removal of compiler support for the simd crate made this issue a lot more urgent.

[hsivonen]

Link to the removal PR.

[hsivonen]

dist-various-1 without modifications (i.e. baseline situation) fails for me with

CMake Error: The source directory "/checkout/src/llvm-emscripten" does not appear to contain CMakeLists.txt.

[hsivonen]

Indeed src/llvm-emscripten is empty.

[hsivonen]

I guess this is one of those "start with a new clone to get submodules in a non-broken state" moments.

[mati865]

This submodule should be checked out by running ./x.py. If it doesn't work try git submodule update --init --recursive --progress.

[hsivonen]

This submodule should be checked out by running ./x.py.

The llvm-emscripten submodule isn't checked out by ./x.py or by a fresh clone of the rust repo.

If it doesn't work try git submodule update --init --recursive --progress.

This worked. Thank you!

[hsivonen]

this probably wants to change dist-various-{1,2} (one or the other) rather than --hosts, because we just want a standard library for this target rather than a full compiler

@alexcrichton This target is the closest to armv7-unknown-linux-gnueabihf. That target is built by the dist-armv7-linux container, which builds a custom GCC toolchain. Presumably, thumbv7neon-unknown-linux-gnueabihf should use the same toolchain, which isn't available in dist-various-1, which seems the closest match for a container used for building stdlib-only Linux targets.

On surface, it seems it should be easier to make dist-armv7-linux build another stdlib than to make dist-various-1 have a custom toolchain that'd need to stay in sync with dist-armv7-linux.

[hsivonen]

On surface, it seems it should be easier to make dist-armv7-linux build another stdlib than to make dist-various-1 have a custom toolchain that'd need to stay in sync with dist-armv7-linux.

This approach seems to work locally for me. Created a PR.