RUST_BACKTRACE truncated with -Cpanic=abort

[Swatinem]

When using -Cpanic=abort, the resulting panic backtrace is being truncated.

Interestingly, when wrapping with a custom panic hook, this problem is even worse.

fn main() {
    let next = std::panic::take_hook();
    std::panic::set_hook(Box::new(move |info| next(info)));

    panic!("oh no! anyway...");
}

Tested on nightly and stable. Only seems to be a problem on mac and linux, windows does not seem to be effected.

RUST_BACKTRACE=full with -Cpanic=unwind

thread 'main' panicked at 'oh no! anyway...', panic-backtrace.rs:5:5
stack backtrace:
   0:     0x5641e75c2937 - std::backtrace_rs::backtrace::libunwind::trace::h8bc78f7bc1f75bdb
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/../../backtrace/src/backtrace/libunwind.rs:90:5
   1:     0x5641e75c2937 - std::backtrace_rs::backtrace::trace_unsynchronized::ha4c961c9576f95ba
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/../../backtrace/src/backtrace/mod.rs:66:5
   2:     0x5641e75c2937 - std::sys_common::backtrace::_print_fmt::hc20cfdc233d6eb02
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:67:5
   3:     0x5641e75c2937 - <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt::h04bea14549780a0f
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:46:22
   4:     0x5641e75d94dc - core::fmt::write::h3868db8542c90941
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/core/src/fmt/mod.rs:1096:17
   5:     0x5641e75c0b22 - std::io::Write::write_fmt::h55dd75636eef6a5c
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/io/mod.rs:1568:15
   6:     0x5641e75c47f5 - std::sys_common::backtrace::_print::h6439e8525077c7da
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:49:5
   7:     0x5641e75c47f5 - std::sys_common::backtrace::print::h02b685f728ce3931
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:36:9
   8:     0x5641e75c47f5 - std::panicking::default_hook::{{closure}}::h7c29625065fafae1
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:208:50
   9:     0x5641e75c4353 - std::panicking::default_hook::hba441710bb71f4d5
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:225:9
  10:     0x5641e75ac41f - <alloc::boxed::Box<F,A> as core::ops::function::Fn<Args>>::call::h96c2c6718f794857
  11:     0x5641e75abc34 - panic_backtrace::main::{{closure}}::hce20eff12a892edc
  12:     0x5641e75c4fb6 - std::panicking::rust_panic_with_hook::ha936eb00e51bc4d9
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:595:17
  13:     0x5641e75ac0da - std::panicking::begin_panic::{{closure}}::h7a9154edebf9ece3
  14:     0x5641e75abc85 - std::sys_common::backtrace::__rust_end_short_backtrace::h1e2014397c41a685
  15:     0x5641e75ac026 - std::panicking::begin_panic::h53c35f27b097d4f4
  16:     0x5641e75abc16 - panic_backtrace::main::h5ef5bf96a2375ff6
  17:     0x5641e75abeb6 - core::ops::function::FnOnce::call_once::hc2eccc5827063ab0
  18:     0x5641e75abcb9 - std::sys_common::backtrace::__rust_begin_short_backtrace::h2e49d68cd7dda505
  19:     0x5641e75ac91c - std::rt::lang_start::{{closure}}::he479198adea7d32d
  20:     0x5641e75c53b7 - core::ops::function::impls::<impl core::ops::function::FnOnce<A> for &F>::call_once::h7c92107c3c98a949
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/core/src/ops/function.rs:259:13
  21:     0x5641e75c53b7 - std::panicking::try::do_call::h7454c57e8fbf0cc1
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:379:40
  22:     0x5641e75c53b7 - std::panicking::try::he71cdc15ed83b5ec
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:343:19
  23:     0x5641e75c53b7 - std::panic::catch_unwind::hfc08ddb043e6e18c
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panic.rs:410:14
  24:     0x5641e75c53b7 - std::rt::lang_start_internal::he4958117ec6901fd
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/rt.rs:51:25
  25:     0x5641e75ac8f8 - std::rt::lang_start::hdc838a40e951ed3a
  26:     0x5641e75abc5b - main
  27:     0x7f1653f01152 - __libc_start_main
  28:     0x5641e75ab09e - _start
  29:                0x0 - <unknown>

RUST_BACKTRACE=full with -Cpanic=abort

thread 'main' panicked at 'oh no! anyway...', panic-backtrace.rs:5:5
stack backtrace:
   0:     0x55bf3d3dc5a7 - std::backtrace_rs::backtrace::libunwind::trace::h8bc78f7bc1f75bdb
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/../../backtrace/src/backtrace/libunwind.rs:90:5
   1:     0x55bf3d3dc5a7 - std::backtrace_rs::backtrace::trace_unsynchronized::ha4c961c9576f95ba
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/../../backtrace/src/backtrace/mod.rs:66:5
   2:     0x55bf3d3dc5a7 - std::sys_common::backtrace::_print_fmt::hc20cfdc233d6eb02
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:67:5
   3:     0x55bf3d3dc5a7 - <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt::h04bea14549780a0f
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:46:22
   4:     0x55bf3d3f2a3c - core::fmt::write::h3868db8542c90941
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/core/src/fmt/mod.rs:1096:17
   5:     0x55bf3d3da792 - std::io::Write::write_fmt::h55dd75636eef6a5c
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/io/mod.rs:1568:15
   6:     0x55bf3d3de3c5 - std::sys_common::backtrace::_print::h6439e8525077c7da
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:49:5
   7:     0x55bf3d3de3c5 - std::sys_common::backtrace::print::h02b685f728ce3931
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/sys_common/backtrace.rs:36:9
   8:     0x55bf3d3de3c5 - std::panicking::default_hook::{{closure}}::h7c29625065fafae1
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:208:50
   9:     0x55bf3d3ddf23 - std::panicking::default_hook::hba441710bb71f4d5
                               at /rustc/04caa632dd10c2bf64b69524c7f9c4c30a436877/library/std/src/panicking.rs:225:9
  10:     0x55bf3d3c610f - <alloc::boxed::Box<F,A> as core::ops::function::Fn<Args>>::call::h96c2c6718f794857
[1]    488156 abort (core dumped)  RUST_BACKTRACE=full ./panic-backtrace

Using only RUST_BACKTRACE=1 in that case will not print any backtrace at all.

[bjorn3]

This happens because -Cpanic=abort omits the unwinding tables necessary for unwinding. You can use -Cforce-unwind-tables=yes to add them back at the cost of an increased binary. The fact that you have part of a backtrace at all is because libstd is compiled with -Cpanic=unwind and as such does have unwinding tables available.

See rust-lang/backtrace-rs#397 for more information.

[m-ou-se]

Maybe we should find a way to remove the run with `RUST_BACKTRACE=1` suggestion from the panic message when unwind tables are not available.

[bjorn3]

How would that work? Plain checking for .eh_frame would almost always result in unwind tables for linked in native dependencies. And when one rust dependency doesn't have unwind tables, another may have. A better solution would be I think to handle this case more gracefully during unwinding by printing something like "unwind info unavailable for the next function; aborting".

[Swatinem]

Thinking a bit more about this, I think linking the creation of unwind tables to the panic method is not a sensible thing.
"catching panics" and "creating stack traces" are two different concerns.

[martin-t]

I second this, I started using panic = "abort" after reading this article because my code is not written to handle random threads unwinding. I, however, still do want backtraces.

Having to add rustflags = ["-Cforce-unwind-tables=y"] to .cargo/config.toml is a bit unwieldy, especially when trying to be careful not to override other rustflags set somewhere else. Is there a better place?

Is it still possible to change the behavior of panic = "abort" to keep unwind tables? If not, then there should at least be a way to set -Cforce-unwind-tables from Cargo.toml.

I also think the behavior and the solution should be documented here.

[Enselic]

This is actually a regression-from-stable-to-stable, namely 1.22 to 1.23. More specifically from nightly-2017-10-13 to nightly-2017-10-14 (very likely #45031)

for toolchain in 1.22 \
                 1.23 \
                 nightly-2017-10-13 \
                 nightly-2017-10-14; do
    echo -e "\n$toolchain:"
    echo 'fn main() { part_of_backtrace(); } fn part_of_backtrace() { panic!("boom"); }' |
        rustc +$toolchain -Cpanic=abort - && \
        RUST_BACKTRACE=full ./rust_out 2>&1 | grep -q part_of_backtrace && echo OK backtrace || echo BROKEN backtrace
done

1.22:
OK backtrace

1.23:
BROKEN backtrace

nightly-2017-10-13:
OK backtrace

nightly-2017-10-14:
BROKEN backtrace

Note that -C panic=abort was introduced already in Rust 1.10.

IMHO this lowers the barrier of changing the default of -C force-unwind-tables for Linux (#94815). And among all issues regarding this I have browsed I have yet to see anyone being against changing that default.

Edit: See #143613 for a PR to change the default.