nalgebra NoSolution Broken MIR ICE

[Andlon]

(Please feel free to make the title more precise)

This issue was first reported as part of #65934, as it was first believed to be similar in nature. Please also see the discussion therein. However, it seems to be sufficiently different as to warrant its own issue.

Here is an example build log

I've created a relatively minimal testcase, and @WildCryptoFox helped me minimize it further. The testcase can be found here: https://github.com/Andlon/ice_testcase . The current rust-toolchain file is set to nightly-2019-11-26, but the bug manifests for a number of different version of nightly (but not all!), as well as beta, and before minification it was also present on stable.

The conditions triggering the ICE seem to be very brittle: Changing features of dependencies or removing unused dependencies might make the ICE vanish. See also @WildCryptoFox's observations in the aforementioned issue #65934.

[hellow554]

Backtrace:

warning: Error finalizing incremental compilation session directory `/tmp/tmp.RK5UN5mIuz/ice_testcase/target/debug/incremental/ice_testcase-xfn1k7c1h8y/s-fi7fvzf5cc-4ag7eb-working`: No such file or directory (os error 2)

error: internal compiler error: broken MIR in DefId(0:35 ~ ice_testcase[1b24]::problematic_function[0]) (Terminator { source_info: SourceInfo { span: src/lib.rs:67:26: 67:80, scope: scope[0] }, kind: _2 = const <nalgebra::base::matrix::Matrix<f64, nalgebra::base::dimension::U2, nalgebra::base::dimension::U1, <nalgebra::base::default_allocator::DefaultAllocator as nalgebra::base::allocator::Allocator<f64, nalgebra::base::dimension::U2>>::Buffer> as std::convert::Into<nalgebra::geometry::point::Point<f64, nalgebra::base::dimension::U2>>>::into(move _3) -> [return: bb3, unwind: bb1] }): bad arg #0 (nalgebra::base::matrix::Matrix<f64, nalgebra::base::dimension::U2, nalgebra::base::dimension::U1, nalgebra::base::array_storage::ArrayStorage<f64, nalgebra::base::dimension::U2, nalgebra::base::dimension::U1>> <- nalgebra::base::matrix::Matrix<f64, nalgebra::base::dimension::U2, nalgebra::base::dimension::U1, <nalgebra::base::default_allocator::DefaultAllocator as nalgebra::base::allocator::Allocator<f64, nalgebra::base::dimension::U2>>::Buffer>): NoSolution
  --> src/lib.rs:67:72
   |
67 |     let _: Point2<f64> = material_surface_element.map_reference_coords().into();
   |                                                                        ^

thread 'rustc' panicked at 'no errors encountered even though `delay_span_bug` issued', src/librustc_errors/lib.rs:347:17
stack backtrace:
   0:     0x7f812a9f15b4 - backtrace::backtrace::libunwind::trace::h0fbcca877b5cc7be
                               at /cargo/registry/src/github.com-1ecc6299db9ec823/backtrace-0.3.40/src/backtrace/libunwind.rs:88
   1:     0x7f812a9f15b4 - backtrace::backtrace::trace_unsynchronized::heee54a145e5e242e
                               at /cargo/registry/src/github.com-1ecc6299db9ec823/backtrace-0.3.40/src/backtrace/mod.rs:66
   2:     0x7f812a9f15b4 - std::sys_common::backtrace::_print_fmt::hcd55bcb24d288827
                               at src/libstd/sys_common/backtrace.rs:84
   3:     0x7f812a9f15b4 - <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt::hcf782761d696ebc5
                               at src/libstd/sys_common/backtrace.rs:61
   4:     0x7f812aa29cec - core::fmt::write::h6d12a02d697eb964
                               at src/libcore/fmt/mod.rs:1030
   5:     0x7f812a9e5947 - std::io::Write::write_fmt::h92b068839d5f88f6
                               at src/libstd/io/mod.rs:1412
   6:     0x7f812a9f5a5e - std::sys_common::backtrace::_print::hd7ae75a7027ca13e
                               at src/libstd/sys_common/backtrace.rs:65
   7:     0x7f812a9f5a5e - std::sys_common::backtrace::print::he36b145a0f885346
                               at src/libstd/sys_common/backtrace.rs:50
   8:     0x7f812a9f5a5e - std::panicking::default_hook::{{closure}}::h78532abf43437641
                               at src/libstd/panicking.rs:190
   9:     0x7f812a9f5751 - std::panicking::default_hook::hb1a613bdd06f1622
                               at src/libstd/panicking.rs:207
  10:     0x7f812af06ac3 - rustc_driver::report_ice::h114705861dab380f
  11:     0x7f812a9f6230 - std::panicking::rust_panic_with_hook::ha051cd69a44ecdee
                               at src/libstd/panicking.rs:470
  12:     0x7f812cd918e5 - std::panicking::begin_panic::h81d44c05963fa168
  13:     0x7f812cdc8cdc - <rustc_errors::HandlerInner as core::ops::drop::Drop>::drop::h25de78a9e6dff3f2
  14:     0x7f812aed34e6 - core::ptr::real_drop_in_place::h9c6508b592c1d638
  15:     0x7f812aed80a3 - <alloc::rc::Rc<T> as core::ops::drop::Drop>::drop::h071d376be54d060e
  16:     0x7f812af1f11c - core::ptr::real_drop_in_place::hbbcadcaa3930f1de
  17:     0x7f812af1a809 - rustc_interface::interface::run_compiler_in_existing_thread_pool::hc0c5fb91eaefa48a
  18:     0x7f812aec95a1 - std::thread::local::LocalKey<T>::with::h47f871d4948616e8
  19:     0x7f812aec24de - scoped_tls::ScopedKey<T>::set::h35e3aee7969efb0c
  20:     0x7f812aee8862 - syntax::with_globals::h9899c44fce69c29e
  21:     0x7f812aec382b - std::sys_common::backtrace::__rust_begin_short_backtrace::h7c08dd05be391ff7
  22:     0x7f812aa06d1a - __rust_maybe_catch_panic
                               at src/libpanic_unwind/lib.rs:81
  23:     0x7f812aede219 - core::ops::function::FnOnce::call_once{{vtable.shim}}::hc69113382c8c4814
  24:     0x7f812a9d787f - <alloc::boxed::Box<F> as core::ops::function::FnOnce<A>>::call_once::h21cb5edd552bea91
                               at /rustc/a44774c3a9739b2eea8923e09d67b14312c78ef3/src/liballoc/boxed.rs:942
  25:     0x7f812aa05740 - <alloc::boxed::Box<F> as core::ops::function::FnOnce<A>>::call_once::hb548d9c171babe6b
                               at /rustc/a44774c3a9739b2eea8923e09d67b14312c78ef3/src/liballoc/boxed.rs:942
  26:     0x7f812aa05740 - std::sys_common::thread::start_thread::hb5b09c960584ac18
                               at src/libstd/sys_common/thread.rs:13
  27:     0x7f812aa05740 - std::sys::unix::thread::Thread::new::thread_start::h341c42eadd8ca9e6
                               at src/libstd/sys/unix/thread.rs:79
  28:     0x7f812a944669 - start_thread
  29:     0x7f812a859323 - clone
  30:                0x0 - <unknown>

error: internal compiler error: unexpected panic

note: the compiler unexpectedly panicked. this is a bug.

note: we would appreciate a bug report: https://github.com/rust-lang/rust/blob/master/CONTRIBUTING.md#bug-reports

note: rustc 1.41.0-nightly (a44774c3a 2019-11-25) running on x86_64-unknown-linux-gnu

note: compiler flags: -C debuginfo=2 -C incremental --crate-type lib

note: some of the compiler flags provided by cargo are hidden

query stack during panic:
end of query stack
error: could not compile `ice_testcase`.

[hellow554]

@rustbot modify labels: E-needs-mcve E-needs-bisection

[jonas-schievink]

The types in the ICE message are:

Matrix<f64, U2, U1, ArrayStorage<f64, U2, U1>>

Matrix<f64, U2, U1, <DefaultAllocator as Allocator<f64, U2>>::Buffer>

That does look like a missing normalization, but those are not usually so hard to reproduce.

[hellow554]

What about the warning beforehand (sorry, I overseen it):

warning: Error finalizing incremental compilation session directory `/tmp/tmp.RK5UN5mIuz/ice_testcase/target/debug/incremental/ice_testcase-xfn1k7c1h8y/s-fi7fvzf5cc-4ag7eb-working`: No such file or directory (os error 2)

maybe an incremental problem which might be the reason for the "hard to reproduce" problem?

[Andlon]

@hellow554: I turned off incremental compilation for all my tests. The bug is completely reproducible without incremental compilation.

EDIT: See clarification below.

[Andlon]

@hellow554: sorry, my initial statement was unclear and might be misunderstood. What I meant is that incremental compilation has no influence on the "hard to reproduce" property of the bug. I explicitly turned off incremental compilation for all my debugging, and the weird behavior with e.g. removing dependencies or features still persisted.

[WildCryptoFox]

Especially bizarre is when you substitute "default" for "std" when default only implies "std" in the Cargo.toml for [dependencies.nalgebra]. Non-default doesn't ICE despite not influencing the generated code at all AFAICT.

[hellow554]

This issue reminds of #66353 which has been fixed by #66388.

Bisection shows regression somewhere between b9de4ef...c6e9c76 but I can't find a commit that would make sense... :/