Bootstrapping i686->i686 on x86_64 host fails in stage0

[nagisa]

$ python src/bootstrap/bootstrap.py --step rustc --stage 1 -j8 --target=i686-unknown-linux-gnu
    Finished debug [unoptimized + debuginfo] target(s) in 0.0 secs
Synchronizing submodule url for 'src/compiler-rt'
Synchronizing submodule url for 'src/jemalloc'
Synchronizing submodule url for 'src/liblibc'
Synchronizing submodule url for 'src/llvm'
Synchronizing submodule url for 'src/rt/hoedown'
Synchronizing submodule url for 'src/rust-installer'
HEAD is now at 8598065 Rust: fix MSVC-built lib pulling in other lib files
HEAD is now at e058ca6 Change how the default zone is found
HEAD is now at eb708c0 Auto merge of #402 - tmiasko:getaddrinfo-error-codes, r=alexcrichton
HEAD is now at a3736a0 Merge pull request #6 from intelfx/patch-1
HEAD is now at 755bc3d Merge pull request #48 from Diggsey/master
Building stage0 std artifacts (i686-unknown-linux-gnu -> i686-unknown-linux-gnu)
   Compiling compiler_builtins v0.0.0 (file:///home/nagisa/Documents/rust/rust/worktrees/more-cace/src/libcompiler_builtins)
   Compiling std v0.0.0 (file:///home/nagisa/Documents/rust/rust/worktrees/more-cace/src/libstd)
   Compiling std_shim v0.1.0 (file:///home/nagisa/Documents/rust/rust/worktrees/more-cace/src/rustc/std_shim)
    Finished release [optimized] target(s) in 15.57 secs
Building stage0 test artifacts (i686-unknown-linux-gnu -> i686-unknown-linux-gnu)
   Compiling term v0.0.0 (file:///home/nagisa/Documents/rust/rust/worktrees/more-cace/src/libterm)
   Compiling getopts v0.0.0 (file:///home/nagisa/Documents/rust/rust/worktrees/more-cace/src/libgetopts)
error: linking with `cc` failed: exit code: 1
  |
  = note: "cc" "-Wl,--as-needed" "-Wl,-z,noexecstack" "-m32" "-L" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-sysroot/lib/rustlib/i686-unknown-linux-gnu/lib" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps/getopts-603f5ee4adfd7630.0.o" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps/getopts-603f5ee4adfd7630.1.o" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps/getopts-603f5ee4adfd7630.2.o" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps/getopts-603f5ee4adfd7630.3.o" "-o" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps/libgetopts-603f5ee4adfd7630.so" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps/getopts-603f5ee4adfd7630.metadata.o" "-Wl,-O1" "-nodefaultlibs" "-L" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-test/i686-unknown-linux-gnu/release/deps" "-L" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-sysroot/lib/rustlib/i686-unknown-linux-gnu/lib" "-Wl,-Bstatic" "-Wl,-Bdynamic" "-L" "/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-sysroot/lib/rustlib/i686-unknown-linux-gnu/lib" "-l" "std-015eff4399913833" "-l" "dl" "-l" "rt" "-l" "pthread" "-l" "gcc_s" "-l" "c" "-l" "m" "-l" "rt" "-l" "util" "-shared" "-Wl,-rpath,$ORIGIN/../lib" "-l" "compiler-rt"
  = note: /usr/bin/ld: /home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-sysroot/lib/rustlib/i686-unknown-linux-gnu/lib/libstd-015eff4399913833.so: __moddi3: invalid version 21 (max 0)
/home/nagisa/Documents/rust/rust/worktrees/more-cace/build/i686-unknown-linux-gnu/stage0-sysroot/lib/rustlib/i686-unknown-linux-gnu/lib/libstd-015eff4399913833.so: error adding symbols: Bad value
collect2: error: ld returned 1 exit status

Some manual bisection shows that it stared happening after rust_builtins PR got merged.

Removing a number of assembly specialisations for i386 makes stuff compile.

[alexcrichton]

Interesting! I've never seen this error before:

/usr/bin/ld: /.../libstd-015eff4399913833.so: __moddi3: invalid version 21 (max 0)

I wonder if may be we're compiling compiler-rt incorrectly?

cc @japaric

[japaric]

Removing a number of assembly specialisations for i386 makes stuff compile.

Perhaps gcc-rs doesn't correctly handle assembly when TARGET=i386 and HOST=x86_64. Because this sounds like the assembly may be being turned into a 64-bit ELF or something like that? (I could be talking nonsense here)

Some manual bisection shows that it stared happening after rust_builtins PR got merged.

You mean #35021? We were already using gcc-rs before that landed. Perhaps I mis-copy-pasted some code from bootstrap?

@nagisa This outcome is with rustbuild, right?

[nagisa]

You mean #35021?

Yes

This outcome is with rustbuild, right?

Yes, as can clearly be seen from the command line invocation. Keep in mind that this bootstraps from i686 beta compiler (i.e. i686 -> i686).

turned into a 64-bit ELF or something like that?

Object files with failing functions were elf32-i386 and contain 32bit code, though I can’t say for sure if all that stayed intact after they got packed into some archive somewhere.

[japaric]

in stage0

Ohhh, I think I know what's happening. Because the compiler still passes -lcompiler-rt in stage0 (more precisely, when building stage1 using stage0 compiler) but building std doesn't generate that artifact, we copy stage0 (the snapshot) libcompiler-rt.a into stage1 build artifacts. The logic for that is probably not correct for this case and we are using copying the 64-bit version of libcompiler-rt.a instead of the 32-bit version -- bootstrap should have downloaded a 32-bit version of stage0, right?

Instead of tweaking that logic we could simply create an empty libcompiler-rt.a and place that in stage1 build artifacts when we build stage1 compiler (before stage1 std). That should also work.

[alexcrichton]

Oh that makes sense. This'll also get fixed on master very soon when we update the bootstrap compiler.

[alexcrichton]

Ok #36811 landed so I think this is closed now