`rustfmt` no longer builds after rust-lang/rust#77306

[rust-highfive]

Hello, this is your friendly neighborhood mergebot.
After merging PR #77306, I observed that the tool rustfmt no longer builds.
A follow-up PR to the repository https://github.com/rust-lang/rustfmt is needed to fix the fallout.

cc @lcnr, do you think you would have time to do the follow-up work?
If so, that would be great!

[matthewjasper]

 thread 'rustc' panicked at 'assertion failed: `(left == right)`
  left: `Binder(<smallvec::SmallVec<[u8; _]> as std::ops::Index<std::ops::RangeFull>>)`,
 right: `Binder(<smallvec::SmallVec<[u8; 36]> as std::ops::Index<std::ops::RangeFull>>)`', compiler/rustc_trait_selection/src/traits/codegen/mod.rs:30:5
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

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/issues/new?labels=C-bug%2C+I-ICE%2C+T-compiler&template=ice.md

note: rustc 1.49.0-nightly (4d247ad7d 2020-10-18) running on x86_64-unknown-linux-gnu

note: compiler flags: -Z macro-backtrace -Z binary-dep-depinfo -C opt-level=3 -C embed-bitcode=no -C debuginfo=0 -C debug-assertions=on -C overflow-checks=off -C link-args=-Wl,-rpath,$ORIGIN/../lib --crate-type lib

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

query stack during panic:
#0 [codegen_fulfill_obligation] checking if `std::ops::Index` fulfills its obligations
#1 [resolve_instance] resolving instance `<smallvec::SmallVec<[u8; _]> as std::ops::Index<std::ops::RangeFull>>::index`
end of query stack
[RUSTC-TIMING] rustc_ap_rustc_data_structures test:false 1.603
error: could not compile **`rustc-ap-rustc_data_structures`**

[bishtpawan]

Hi @matthewjasper, can you please the command to replicate this bug? Thanks

[matthewjasper]

.\x.py build .\src\tools\rustfmt\ with debug-assertions = true set in config.toml

[lcnr]

won't be able to get to this myself in the next few days. To fix this we probably have to compile it with RUST_BACKTRACE=1, find the place where we call Instance::resolve and put a tcx.normalize_erasing_regions before that

[bishtpawan]

Hi @lcnr, the error occurred where we just erasing regions by simply doing this (tcx.erase_regions(&trait_ref)) and provided assertion with (tcx.normalize_erasing_regions(param_env, trait_ref)). Is this scenario correct? Just want to confirm as you already mentioned to put a tcx.normalize_erasing_regions. Thanks

[lcnr]

yes, so we do not want to use tcx.normalize_erasing_regions in codegen_fulfill_obligations as that can be a bit costly but instead normalize eagerly before then.

To detect cases where we are still using unnormalized values #77306 added a debug assert there.
If we look at the backtrace we can find out where the trait_ref is coming from and normalize it there

[bishtpawan]

okay, thanks for the update.

[bishtpawan]

Hi @lcnr as you mentioned [find the place where we call Instance::resolve and put a tcx.normalize_erasing_regions before that] can you please elaborate it please? As I can see the Instace::resolve in [compiler/rustc_middle/src/ty/instance.rs] but don't get you point to put tcx.normalize_erasing_regions. Thanks

[lcnr]

Can you post the resulting backtrace using RUST_BACKTRACE=1?

[bishtpawan]

@lcnr, here it is:

0: rust_begin_unwind
   1: std::panicking::begin_panic_fmt
   2: rustc_trait_selection::traits::codegen::codegen_fulfill_obligation
   3: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::codegen_fulfill_obligation>::compute
   4: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
   5: rustc_data_structures::stack::ensure_sufficient_stack
   6: rustc_query_system::query::plumbing::get_query_impl
   7: rustc_ty::instance::inner_resolve_instance
   8: rustc_ty::instance::resolve_instance
   9: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::resolve_instance>::compute
  10: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  11: rustc_data_structures::stack::ensure_sufficient_stack
  12: rustc_query_system::query::plumbing::get_query_impl
  13: rustc_middle::ty::instance::Instance::resolve_opt_const_arg
  14: rustc_middle::ty::instance::Instance::resolve
  15: rustc_mir_build::lints::Search::is_recursive_call
  16: rustc_data_structures::graph::iterate::TriColorDepthFirstSearch<G>::run_from_start
  17: rustc_mir_build::lints::check
  18: rustc_infer::infer::InferCtxtBuilder::enter
  19: rustc_mir_build::build::mir_built
  20: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::mir_built>::compute
  21: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  22: rustc_data_structures::stack::ensure_sufficient_stack
  23: rustc_query_system::query::plumbing::get_query_impl
  24: rustc_mir::transform::check_unsafety::unsafety_check_result
  25: core::ops::function::FnOnce::call_once
  26: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::unsafety_check_result>::compute
  27: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  28: rustc_data_structures::stack::ensure_sufficient_stack
  29: rustc_query_system::query::plumbing::get_query_impl
  30: rustc_query_system::query::plumbing::ensure_query_impl
  31: rustc_mir::transform::mir_const
  32: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::mir_const>::compute
  33: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  34: rustc_data_structures::stack::ensure_sufficient_stack
  35: rustc_query_system::query::plumbing::get_query_impl
  36: rustc_mir::transform::mir_promoted
  37: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::mir_promoted>::compute
  38: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  39: rustc_data_structures::stack::ensure_sufficient_stack
  40: rustc_query_system::query::plumbing::get_query_impl
  41: rustc_mir::borrow_check::mir_borrowck
  42: core::ops::function::FnOnce::call_once
  43: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::mir_borrowck>::compute
  44: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  45: rustc_data_structures::stack::ensure_sufficient_stack
  46: rustc_query_system::query::plumbing::get_query_impl
  47: rustc_query_system::query::plumbing::ensure_query_impl
  48: rustc_middle::ty::<impl rustc_middle::ty::context::TyCtxt>::par_body_owners
  49: rustc_interface::passes::analysis
  50: rustc_middle::ty::query::<impl rustc_query_system::query::config::QueryAccessors<rustc_middle::ty::context::TyCtxt> for rustc_middle::ty::query::queries::analysis>::compute
  51: rustc_query_system::dep_graph::graph::DepGraph<K>::with_task_impl
  52: rustc_data_structures::stack::ensure_sufficient_stack
  53: rustc_query_system::query::plumbing::get_query_impl
  54: rustc_interface::queries::<impl rustc_interface::interface::Compiler>::enter
  55: rustc_span::with_source_map
  56: rustc_interface::interface::create_compiler_and_run
  57: scoped_tls::ScopedKey<T>::set

[lcnr]

So I think that the easiest fix is to add tcx.normalize_erasing_regions(param_env, substs) before this call.

rust/compiler/rustc_mir_build/src/lints.rs

Lines 74 to 80 in 9832374

	let (callee, call_substs) =
	if let Ok(Some(instance)) = Instance::resolve(tcx, param_env, callee, substs) {
	(instance.def_id(), instance.substs)
	} else {
	(callee, substs)
	};

I am however surprised that func.ty(body, tcx) isn't actually normalized here. cc @eddyb @nikomatsakis do we want to instead normalize this in mir building?