Regression in cross-compiling from 64 -> 32 bit

[alexcrichton]

In today's most recent nightly:

rustc 1.13.0-nightly (d0623cf7b 2016-09-26)
binary: rustc
commit-hash: d0623cf7bda44849ab5df78a06b22f9108cf821a
commit-date: 2016-09-26
host: x86_64-unknown-linux-gnu
release: 1.13.0-nightly

Compiling a simple program:

fn main() {}

yields

$ rustc +nightly foo.rs --target i686-unknown-linux-gnu
error: linking with `cc` failed: exit code: 1
  |
  = note: "cc" "-Wl,--as-needed" "-Wl,-z,noexecstack" "-m32" "-L" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib" "foo.0.o" "-o" "foo" "-Wl,--gc-sections" "-pie" "-nodefaultlibs" "-L" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib" "-Wl,-Bstatic" "-Wl,-Bdynamic" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libstd-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libpanic_unwind-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libunwind-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/librand-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libcollections-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/librustc_unicode-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/liballoc-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/liballoc_jemalloc-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/liblibc-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libcore-411f48d3.rlib" "/home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libcompiler_builtins-411f48d3.rlib" "-l" "dl" "-l" "pthread" "-l" "gcc_s" "-l" "pthread" "-l" "c" "-l" "m" "-l" "rt" "-l" "util"
  = note: foo.0.o: In function `main':
foo.cgu-0.rs:(.text.main+0x2d): undefined reference to `std::rt::lang_start::h95854775e4ac8b95'
collect2: error: ld returned 1 exit status


error: aborting due to previous error

Seems odd to have an undefined symbol in the standard library! Taking a look at what's actually inside:

$ nm -g /home/alex/.multirust/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/i686-unknown-linux-gnu/lib/libstd-411f48d3.rlib | grep lang_start | c++filt
00000000 T std::rt::lang_start::hb5404024d2e88016

Looks like the hashes are different!

Did we perhaps leak a usize into metadata/symbol calculation?

cc @eddyb
cc @michaelwoerister

[eddyb]

So I guess tuple lengths were always broken w/ cross-compilation, but only through TypeId.
You can fix it by adding as u64 (or u32 but that'd require an extra check) here and here.

[BurntSushi]

Is this related? https://travis-ci.org/BurntSushi/ripgrep/jobs/163366626 (Cross compiling i686-apple-darwin from x86_64-apple-darwin.)

[TimNN]

@BurntSushi I'd think that you're experiencing exactly this issue, although I'm not sure what to make of these warnings (and if they are even important):

  = note: ld: warning: could not create compact unwind for __ZN4core3num7flt2dec8strategy6dragon15format_shortest17hbdcbfeff9d90c839E: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN4core3fmt23float_to_decimal_common17h07a52791ce6b485eE: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN4core3fmt23float_to_decimal_common17h62a680cf01e490e7E: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN4core3fmt27float_to_exponential_common17hb2f23958bdef5ec0E: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN4core3fmt27float_to_exponential_common17hedda57d97c9f5d15E: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN3std3sys2fs4copy17ha1b8cf4f7b2525caE: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN3std4rand6StdRng3new17h2d301b41144d5cfcE: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN99_$LT$std..rand..ThreadRngReseeder$u20$as$u20$rand..reseeding..Reseeder$LT$std..rand..StdRng$GT$$GT$6reseed17h37d92a6ecfebac86E: stack subl instruction is too different from dwarf stack size
ld: warning: could not create compact unwind for __ZN3std4rand10thread_rng14THREAD_RNG_KEY6__init17hfd5fb3b8ece6f6d7E: stack subl instruction is too different from dwarf stack size

[japaric]

This bug is affecting the current beta: 1.13.0.beta.1. Can the fix be backported? This bug may affect bootstrapping the compiler, I think; that beta is (or will soon be) the current stage0 (?).