ICE: const eval error with an empty slice.

[rodrimati1992]

This code:

pub struct Slice<'a, T>(&'a [T]);

impl<'a, T: 'a> Slice<'a, T> {
    pub const EMPTY: Self = Slice ({
        let v: &[T] = &[];
        v
    });
}

Errors on rustc 1.40.0-nightly (22bc9e1d9 2019-09-30) running on x86_64-unknown-linux-gnu with this message:

   Compiling playground v0.0.1 (/playground)
[ERROR rustc_mir::transform::qualify_consts] old validator: []
[ERROR rustc_mir::transform::qualify_consts] new validator: [(src/lib.rs:5:24: 5:26, "LiveDrop")]
error: internal compiler error: src/librustc_mir/transform/qualify_consts.rs:1038: Disagreement between legacy and dataflow-based const validators.
    After filing an issue, use `-Zsuppress-const-validation-back-compat-ice` to compile your code.
 --> src/lib.rs:4:5
  |
4 | /     pub const EMPTY: Self = Slice ({
5 | |         let v: &[T] = &[];
6 | |         v
7 | |     });
  | |_______^

thread 'rustc' panicked at 'Box<Any>', src/librustc_errors/lib.rs:871:9
stack backtrace:
   0: backtrace::backtrace::libunwind::trace
             at /cargo/registry/src/git.luolix.top-1ecc6299db9ec823/backtrace-0.3.37/src/backtrace/libunwind.rs:88
   1: backtrace::backtrace::trace_unsynchronized
             at /cargo/registry/src/git.luolix.top-1ecc6299db9ec823/backtrace-0.3.37/src/backtrace/mod.rs:66
   2: std::sys_common::backtrace::_print_fmt
             at src/libstd/sys_common/backtrace.rs:76
   3: <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt
             at src/libstd/sys_common/backtrace.rs:60
   4: core::fmt::write
             at src/libcore/fmt/mod.rs:1028
   5: std::io::Write::write_fmt
             at src/libstd/io/mod.rs:1412
   6: std::sys_common::backtrace::_print
             at src/libstd/sys_common/backtrace.rs:64
   7: std::sys_common::backtrace::print
             at src/libstd/sys_common/backtrace.rs:49
   8: std::panicking::default_hook::{{closure}}
             at src/libstd/panicking.rs:196
   9: std::panicking::default_hook
             at src/libstd/panicking.rs:210
  10: rustc_driver::report_ice
  11: std::panicking::rust_panic_with_hook
             at src/libstd/panicking.rs:477
  12: std::panicking::begin_panic
  13: rustc_errors::HandlerInner::span_bug
  14: rustc_errors::Handler::span_bug
  15: rustc::util::bug::opt_span_bug_fmt::{{closure}}
  16: rustc::ty::context::tls::with_opt::{{closure}}
  17: rustc::ty::context::tls::with_context_opt
  18: rustc::ty::context::tls::with_opt
  19: rustc::util::bug::opt_span_bug_fmt
  20: rustc::util::bug::span_bug_fmt
  21: rustc_mir::transform::qualify_consts::Checker::check_const
  22: rustc_mir::transform::qualify_consts::mir_const_qualif
  23: rustc::ty::query::__query_compute::mir_const_qualif
  24: rustc::ty::query::<impl rustc::ty::query::config::QueryAccessors for rustc::ty::query::queries::mir_const_qualif>::compute
  25: rustc::dep_graph::graph::DepGraph::with_task_impl
  26: rustc::ty::query::plumbing::<impl rustc::ty::context::TyCtxt>::get_query
  27: rustc_mir::transform::mir_validated
  28: rustc::ty::query::__query_compute::mir_validated
  29: rustc::ty::query::<impl rustc::ty::query::config::QueryAccessors for rustc::ty::query::queries::mir_validated>::compute
  30: rustc::dep_graph::graph::DepGraph::with_task_impl
  31: rustc::ty::query::plumbing::<impl rustc::ty::context::TyCtxt>::get_query
  32: rustc_mir::borrow_check::mir_borrowck
  33: rustc::ty::query::__query_compute::mir_borrowck
  34: rustc::ty::query::<impl rustc::ty::query::config::QueryAccessors for rustc::ty::query::queries::mir_borrowck>::compute
  35: rustc::dep_graph::graph::DepGraph::with_task_impl
  36: rustc::ty::query::plumbing::<impl rustc::ty::context::TyCtxt>::get_query
  37: rustc::ty::<impl rustc::ty::context::TyCtxt>::par_body_owners
  38: rustc::util::common::time
  39: rustc_interface::passes::analysis
  40: rustc::ty::query::__query_compute::analysis
  41: rustc::dep_graph::graph::DepGraph::with_task_impl
  42: rustc::ty::query::plumbing::<impl rustc::ty::context::TyCtxt>::get_query
  43: rustc_interface::passes::BoxedGlobalCtxt::access::{{closure}}
  44: rustc_interface::passes::create_global_ctxt::{{closure}}
  45: rustc_interface::interface::run_compiler_in_existing_thread_pool
  46: std::thread::local::LocalKey<T>::with
  47: scoped_tls::ScopedKey<T>::set
  48: syntax::with_globals
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

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 1.40.0-nightly (22bc9e1d9 2019-09-30) running on x86_64-unknown-linux-gnu

note: compiler flags: -C codegen-units=1 -C debuginfo=2 --crate-type lib

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

query stack during panic:
#0 [mir_const_qualif] processing `Slice::<'a, T>::EMPTY`
#1 [mir_validated] processing `Slice::<'a, T>::EMPTY`
#2 [mir_borrowck] processing `Slice::<'a, T>::EMPTY`
#3 [analysis] running analysis passes on this crate
end of query stack
error: aborting due to previous error

error: could not compile `playground`.

To learn more, run the command again with --verbose.

[Centril]

cc @ecstatic-morse @oli-obk

[hellow554]

regression in 0bbab7d

[ecstatic-morse]

I'm still building a compiler with graphviz debugging enabled so I can have a closer look, but I have a theory.

I believe the problem is that we are too conservative when marking locals as indirectly mutable. Because we pessimistically assume that type T is !Freeze, we also assume that type [T; 0] can be mutated through a shared reference. Once a shared reference to it is taken, we assume that all qualifs that could apply to [T; 0] are applied to that local. We should special case zero-sized arrays here, so when the borrow of [T; 0] occurs, it is not marked as indirectly mutable. Perhaps we could assume that zero-sized arrays never actually Drop as well?

While this change would solve this particular issue, I believe that this won't solve the problem entirely. What if we take two references to the zero-sized type (&&[])? While the local containing the ZST itself will not be marked as indirectly mutable, the local containing the first borrow will be, which diverges from the old analysis. We may need to base whether a Local is indirectly mutable on the HasMutInterior analysis.