save-analysis fails with proc macro and nested generic type

[kdy1]

rls shows it as warning, but all codes after the field are not analyzed (including sibling modules). And rls ICEs on dependent crate.

Warning from rls-vscode:

[rustc] librustc_save_analysis\span_utils.rs:156: Mis-counted brackets when breaking path? Parsing 'Box>' in ecmascript\ast\src\class.rs, line 10

Backtrace (from wsl, because my windows rustc doesn't produce any stacktrace)

$ CARGO_INCREMENTAL=0  RUST_BACKTRACE=1 rls --cli
Initializing (look for `diagnosticsEnd` message)...
> 1: InitializeResult {
    capabilities: ServerCapabilities {
        text_document_sync: Some(
            Kind(
                Incremental
            )
        ),
        hover_provider: Some(
            true
        ),
        completion_provider: Some(
            CompletionOptions {
                resolve_provider: Some(
                    true
                ),
                trigger_characters: [
                    ".",
                    ":"
                ]
            }
        ),
        signature_help_provider: None,
        definition_provider: Some(
            true
        ),
        references_provider: Some(
            true
        ),
        document_highlight_provider: Some(
            true
        ),
        document_symbol_provider: Some(
            true
        ),
        workspace_symbol_provider: Some(
            true
        ),
        code_action_provider: Some(
            true
        ),
        code_lens_provider: None,
        document_formatting_provider: Some(
            true
        ),
        document_range_formatting_provider: Some(
            false
        ),
        document_on_type_formatting_provider: None,
        rename_provider: Some(
            true
        ),
        execute_command_provider: Some(
            ExecuteCommandOptions {
                commands: [
                    "rls.applySuggestion",
                    "rls.deglobImports"
                ]
            }
        )
    }
}
{"jsonrpc":"2.0","method":"rustDocument/beginBuild"}
thread 'rustc' panicked at 'Box<Any>', librustc_errors/lib.rs:482:9
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
stack backtrace:
   0: std::sys::unix::backtrace::tracing::imp::unwind_backtrace
             at libstd/sys/unix/backtrace/tracing/gcc_s.rs:49
   1: std::sys_common::backtrace::print
             at libstd/sys_common/backtrace.rs:71
             at libstd/sys_common/backtrace.rs:59
   2: std::panicking::default_hook::{{closure}}
             at libstd/panicking.rs:380
   3: std::panicking::default_hook
             at libstd/panicking.rs:396
   4: std::panicking::rust_panic_with_hook
             at libstd/panicking.rs:576
   5: std::panicking::begin_panic
   6: rustc_errors::Handler::span_bug
   7: <std::thread::local::LocalKey<T>>::with
   8: rustc::ty::context::tls::with_opt
   9: rustc::session::span_bug_fmt
  10: rustc_save_analysis::span_utils::SpanUtils::sub_span_for_type_name
  11: <rustc_save_analysis::dump_visitor::DumpVisitor<'l, 'tcx, 'll, O> as syntax::visit::Visitor<'l>>::visit_ty
  12: syntax::visit::walk_path_parameters
  13: <rustc_save_analysis::dump_visitor::DumpVisitor<'l, 'tcx, 'll, O> as syntax::visit::Visitor<'l>>::visit_ty
  14: <rustc_save_analysis::dump_visitor::DumpVisitor<'l, 'tcx, 'll, O> as syntax::visit::Visitor<'l>>::visit_item
  15: <rustc_save_analysis::dump_visitor::DumpVisitor<'l, 'tcx, 'll, O> as syntax::visit::Visitor<'l>>::visit_item
  16: <rustc_save_analysis::dump_visitor::DumpVisitor<'l, 'tcx, 'll, O> as syntax::visit::Visitor<'l>>::visit_mod
  17: <rustc_save_analysis::CallbackHandler<'b> as rustc_save_analysis::SaveHandler>::save
  18: rustc_save_analysis::process_crate
  19: <rls::build::rustc::RlsRustcCalls as rustc_driver::CompilerCalls<'a>>::build_controller::{{closure}}
  20: rustc::dep_graph::graph::DepGraph::with_ignore
  21: rustc_driver::driver::compile_input::{{closure}}
  22: <std::thread::local::LocalKey<T>>::with
  23: <std::thread::local::LocalKey<T>>::with
  24: rustc::ty::context::TyCtxt::create_and_enter
  25: rustc_driver::driver::compile_input
  26: rustc_driver::run_compiler

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 1.25.0-nightly (56733bc9f 2018-02-01) running on x86_64-unknown-linux-gnu

{"jsonrpc":"2.0","method":"rustDocument/diagnosticsBegin"}
{"jsonrpc":"2.0","method":"rustDocument/diagnosticsEnd"}

$ rustup -V && rustc -vV  && cargo -vV
rustup 1.10.0 (c6a8f7c60 2018-01-25)
rustc 1.25.0-nightly (616b66dca 2018-02-02)
binary: rustc
commit-hash: 616b66dca25a67321b1654e5a65acc6337d63cf4
commit-date: 2018-02-02
host: x86_64-pc-windows-gnu
release: 1.25.0-nightly
LLVM version: 4.0
cargo 0.26.0-nightly (1d6dfea44 2018-01-26)
release: 0.26.0
commit-hash: 1d6dfea44f97199d5d5c177c7dadcde393eaff9a
commit-date: 2018-01-26

[kdy1]

rls works well if you use Option<(Box<Expr>)> instead of Option<Box<Expr>>

[nrc]

Do you have the exact source code which caused this error please?

[kdy1]

PR which changed Option<Box<Expr>> into Option<(Box<Expr>)>: swc-project/swc#30

This part is relevant

https://github.com/swc-project/swc/blob/fa40c8ddf3079d1bc624efd1d2059277bc72dcf2/ecmascript/ast/src/class.rs#L5-L11

#[ast_node] is an alias for #[derive(AstNode, Fold, Clone, Debug, PartialEq)] and inlining it does not have any effect.

---

Oh.. I think I found a clue..

// This is broken
#[derive(AstNode, Fold, Clone, Debug, PartialEq)]
pub struct Class {
    pub span: Span,

    pub body: Vec<ClassMethod>,
    pub super_class: Option<Box<Expr>>,
}

// This works
#[derive(Clone, Debug, PartialEq)]
pub struct Class {
    pub span: Span,

    pub body: Vec<ClassMethod>,
    pub super_class: Option<Box<Expr>>,
}

So the issue is about derive macro, and I'm sure that problematic
part is ::swc_common::Folder::<Option<Box<Expr>>>::fold(_folder, _super_class)

---

Reduced to

#![feature(specialization)]

#[macro_use]
extern crate ast_node;
extern crate swc_common;

#[derive(Fold)]
pub struct Expr {
    pub super_class: Option<Box<Expr>>,
}

---

But when I copy generated code to other file, it works. (and warns about unused macro import)

cargo expand:

pub struct Expr {
    pub super_class: Option<Box<Expr>>,
}
#[allow(dead_code, non_upper_case_globals)]
const DERIVE_FOLD_FOR_Expr: () =
    {
        extern crate swc_common;
        impl <__Folder> swc_common::FoldWith<__Folder> for Expr {
            fn fold_children(self, __folder: &mut __Folder) -> Self {
                match self {
                    Expr { super_class: _super_class } => {
                        return Expr{super_class:
                                        swc_common::Folder::<Option<Box<Expr>>>::fold(__folder,
                                                                                      _super_class),};
                    }
                }
            }
        }
    };

---

So I guess it is related to span generated by proc macro..

https://github.com/swc-project/swc/blob/ca663aa33d7e313daf8e9ef138da052738a73ed2/macros/ast_node/src/fold.rs#L59-L65

In this case,

Span of ::swc_common::Folder::<$>::fold(_folder, $) is call_site (= #[derive(Fold)]).

FieldType is Option<Box<Expr>>, and has same span with field's type.

[kdy1]

I think title should be updated, but I'm not sure if ufcs for generic trait is necessary.

[nrc]

I've been trying to repro, but it is kind of involved to setup the exact conditions to cause the crash. I currently can't do it outside of the entire project.

[nrc]

I can repro, and I can see the assertion which is causing the problem and verify that there is a span problem. But there are a lot of places that the error could be introduce - could be in the proc-macro bits of the compiler or in the libs or (quite likely) in Syn. I think rather than trying to track down the error it is better to just make save-analysis more robust.