stack overflow when compiling jemallocator

[gnzlbg]

See https://github.com/alexcrichton/jemallocator/issues/61

It might be a cargo issue, since the crate doesn't even compile.

git clone git@github.com:alexcrichton/jemallocator.git
# We are working around this issue on master, the following commit
# that used to work reproduces the issue:
git checkout de8bcc693242e7ad3deb2b4a083e6b2cb107379e
cd jemallocator
cargo test -vv  -p jemalloc-sys

fails with:

thread 'main' has overflowed its stack
fatal runtime error: stack overflow

Running lldb reports:

    frame #30674: 0x00000001003b3406 cargo`cargo::core::compiler::context::compilation_files::metadata_of::h1377c51c44490a0e (.llvm.10155370706564831419) + 1046
    frame #30675: 0x00000001002c1a73 cargo`_$LT$alloc..vec..Vec$LT$T$GT$$u20$as$u20$alloc..vec..SpecExtend$LT$T$C$$u20$I$GT$$GT$::from_iter::h8d627d05abe98e62 + 163
    frame #30676: 0x00000001003b3406 cargo`cargo::core::compiler::context::compilation_files::metadata_of::h1377c51c44490a0e (.llvm.10155370706564831419) + 1046
    frame #30677: 0x00000001002c1a73 cargo`_$LT$alloc..vec..Vec$LT$T$GT$$u20$as$u20$alloc..vec..SpecExtend$LT$T$C$$u20$I$GT$$GT$::from_iter::h8d627d05abe98e62 + 163
    frame #30678: 0x00000001003b3406 cargo`cargo::core::compiler::context::compilation_files::metadata_of::h1377c51c44490a0e (.llvm.10155370706564831419) + 1046
    frame #30679: 0x00000001002c1a73 cargo`_$LT$alloc..vec..Vec$LT$T$GT$$u20$as$u20$alloc..vec..SpecExtend$LT$T$C$$u20$I$GT$$GT$::from_iter::h8d627d05abe98e62 + 163
    frame #30680: 0x00000001003b3406 cargo`cargo::core::compiler::context::compilation_files::metadata_of::h1377c51c44490a0e (.llvm.10155370706564831419) + 1046
    frame #30681: 0x00000001002c1a73 cargo`_$LT$alloc..vec..Vec$LT$T$GT$$u20$as$u20$alloc..vec..SpecExtend$LT$T$C$$u20$I$GT$$GT$::from_iter::h8d627d05abe98e62 + 163
    frame #30682: 0x00000001003b3406 cargo`cargo::core::compiler::context::compilation_files::metadata_of::h1377c51c44490a0e (.llvm.10155370706564831419) + 1046
    frame #30683: 0x00000001003b071c cargo`cargo::core::compiler::context::compilation_files::CompilationFiles::new::h66fee72ceee36cd6 + 108
    frame #30684: 0x00000001001fc5de cargo`cargo::core::compiler::context::Context::prepare_units::ha547edde6c67aa45 + 750
    frame #30685: 0x00000001001fa46e cargo`cargo::core::compiler::context::Context::compile::h12022378937f4dce + 238
    frame #30686: 0x00000001001250c1 cargo`cargo::ops::cargo_compile::compile_ws::h729cd70799cd6467 + 11105
    frame #30687: 0x00000001001220db cargo`cargo::ops::cargo_compile::compile_with_exec::h9eb79a516b8ba936 + 395
    frame #30688: 0x00000001001a14dc cargo`cargo::ops::cargo_test::compile_tests::hf96f120cf7301ba3 + 92
    frame #30689: 0x00000001001a098c cargo`cargo::ops::cargo_test::run_tests::h32ed0913c2b1837e + 44
    frame #30690: 0x000000010004fe9b cargo`cargo::commands::test::exec::h5212fd998f544836 + 1147
    frame #30691: 0x0000000100012f59 cargo`cargo::cli::main::h5a4c5fdf41da489e + 4345
    frame #30692: 0x000000010000ded6 cargo`cargo::main::hd16f3801c6514bd6 + 134
    frame #30693: 0x0000000100034ac6 cargo`std::rt::lang_start::_$u7b$$u7b$closure$u7d$$u7d$::h132926336c109f4c + 6
    frame #30694: 0x00000001007323e8 cargo`std::panicking::try::do_call::hef9dda5d08757d6c (.llvm.7045045766936433836) + 24
    frame #30695: 0x000000010073ec1f cargo`__rust_maybe_catch_panic + 31
    frame #30696: 0x0000000100721832 cargo`std::rt::lang_start_internal::hccd3760e054b55b8 + 242
    frame #30697: 0x000000010001029c cargo`main + 44
    frame #30698: 0x0000000100001434 cargo`start + 52

[alexcrichton]

@ehuss a bisection points to #5651 with a minimization of:

[package]
name = "foo"
version = "0.1.0"
authors = []

[lib]
test = false

[alexcrichton]

Er actual minimization:

[package]
name = "jemalloc-sys"
version = "0.1.7"
authors = []
links = 'foo' # important
build = 'build.rs'

[lib]
test = false
path = 'lib.rs'

You shouldn't even need any other files on the filesystem, cargo +nightly test is enough to stack overflow

[alexcrichton]

Hm ok I think this was just a regression uncovered by that PR (although not sure why), this looks like it may be a longstanding bug, still investigating.

[alexcrichton]

Er wait no, it's directly related to these changes, I accidentally thought that'd had been there before.

[alexcrichton]

A fix should be in #5711