Exponential compile time with broken huge nested enum
jruderman opened this issue · comments
Long ago, in #42747, an exponential compile time issue was fixed for huge nested enums.
Fuzzing discovered that deleting lines from src/test/ui/enum/issue-42747.rs, and thus introducing a resolve error, brings back exponential compile time.
Code
macro_rules! fooN {
($cur:ident $prev:ty) => {
#[allow(dead_code)]
enum $cur {
Empty,
First($prev),
Second($prev),
Third($prev),
Fourth($prev),
}
}
}
fooN!(Foo16 Foo15); //~ ERROR cannot find type `Foo15` in this scope
fooN!(Foo17 Foo16);
fooN!(Foo18 Foo17);
fooN!(Foo19 Foo18);
fooN!(Foo20 Foo19);
fooN!(Foo21 Foo20);
fooN!(Foo22 Foo21);
fooN!(Foo23 Foo22);
fooN!(Foo24 Foo23);
fooN!(Foo25 Foo24);
fooN!(Foo26 Foo25);
fooN!(Foo27 Foo26);
fn main() {
let _foo = Foo27::Empty;
}Where it's slow
The slow passes are wf_checking, item_bodies_checking, and type_check_crate. Memory usage is also much higher than with the non-erroneous, original test.
Note that the error is emitted much earlier, in resolve_report_errors.
Output with -Z time-passes
time: 0.010; rss: 21MB -> 22MB ( +1MB) parse_crate
time: 0.000; rss: 22MB -> 22MB ( +0MB) attributes_injection
time: 0.000; rss: 22MB -> 22MB ( +0MB) plugin_loading
time: 0.002; rss: 22MB -> 22MB ( +0MB) crate_injection
time: 0.035; rss: 22MB -> 26MB ( +3MB) expand_crate
time: 0.036; rss: 22MB -> 26MB ( +4MB) macro_expand_crate
time: 0.000; rss: 26MB -> 26MB ( +0MB) maybe_building_test_harness
time: 0.004; rss: 26MB -> 26MB ( +0MB) AST_validation
time: 0.000; rss: 26MB -> 26MB ( +0MB) maybe_create_a_macro_crate
time: 0.000; rss: 26MB -> 26MB ( +0MB) finalize_imports
time: 0.001; rss: 26MB -> 26MB ( +0MB) compute_effective_visibilities
time: 0.000; rss: 26MB -> 27MB ( +0MB) finalize_macro_resolutions
time: 0.029; rss: 27MB -> 38MB ( +11MB) late_resolve_crate
time: 0.000; rss: 38MB -> 38MB ( +0MB) resolve_check_unused
error[E0412]: cannot find type `Foo15` in this scope
--> *.rs:14:13
|
4 | enum $cur {
| --------- similarly named enum `Foo16` defined here
...
14 | fooN!(Foo16 Foo15); //~ ERROR cannot find type `Foo15` in this scope
| ^^^^^ help: an enum with a similar name exists: `Foo16`
time: 0.003; rss: 38MB -> 38MB ( +0MB) resolve_report_errors
time: 0.001; rss: 38MB -> 38MB ( +0MB) resolve_postprocess
time: 0.034; rss: 26MB -> 38MB ( +12MB) resolve_crate
time: 0.000; rss: 38MB -> 38MB ( +0MB) complete_gated_feature_checking
time: 0.003; rss: 38MB -> 39MB ( +0MB) early_lint_checks
time: 0.089; rss: 22MB -> 39MB ( +17MB) configure_and_expand
time: 0.001; rss: 39MB -> 39MB ( +0MB) prepare_outputs
time: 0.003; rss: 39MB -> 39MB ( +0MB) setup_global_ctxt
time: 0.001; rss: 40MB -> 40MB ( +0MB) drop_ast
time: 0.051; rss: 39MB -> 40MB ( +1MB) looking_for_entry_point
time: 0.001; rss: 40MB -> 40MB ( +0MB) looking_for_derive_registrar
time: 0.000; rss: 41MB -> 41MB ( +0MB) unused_lib_feature_checking
time: 0.068; rss: 39MB -> 41MB ( +2MB) misc_checking_1
time: 0.016; rss: 41MB -> 42MB ( +1MB) type_collecting
time: 0.000; rss: 42MB -> 42MB ( +0MB) impl_wf_inference
time: 0.004; rss: 42MB -> 42MB ( +0MB) coherence_checking
time: 11.404; rss: 42MB -> 93MB ( +51MB) wf_checking
time: 0.115; rss: 93MB -> 94MB ( +1MB) item_types_checking
time: 32.556; rss: 94MB -> 166MB ( +72MB) item_bodies_checking
time: 44.123; rss: 41MB -> 167MB ( +126MB) type_check_crate
time: 0.001; rss: 167MB -> 167MB ( +0MB) serialize_dep_graph
time: 0.045; rss: 167MB -> 24MB ( -143MB) free_global_ctxt
error: aborting due to previous error
For more information about this error, try `rustc --explain E0412`.
time: 44.474; rss: 13MB -> 23MB ( +10MB) total
Version
rustc 1.67.0-nightly (7eef946fc 2022-11-06)
binary: rustc
commit-hash: 7eef946fc0e0eff40e588eab77b09b287accbec3
commit-date: 2022-11-06
host: x86_64-apple-darwin
release: 1.67.0-nightly
LLVM version: 15.0.4
@rustbot label +I-compiletime
Root-cause is likely that we're doing some expensive computation during typeck, but then suppressing it due to the fact that name resolution fails.