1 //! An "interner" is a data structure that associates values with usize tags and
2 //! allows bidirectional lookup; i.e., given a value, one can easily find the
3 //! type, and vice versa.
5 use arena::DroplessArena;
6 use rustc_data_structures::fx::FxHashMap;
7 use rustc_data_structures::stable_hasher::{HashStable, StableHasher, ToStableHashKey};
8 use rustc_macros::{symbols, HashStable_Generic};
9 use rustc_serialize::{Decodable, Decoder, Encodable, Encoder};
10 use rustc_serialize::{UseSpecializedDecodable, UseSpecializedEncodable};
12 use std::cmp::{Ord, PartialEq, PartialOrd};
14 use std::hash::{Hash, Hasher};
17 use crate::{Span, DUMMY_SP, GLOBALS};
23 // After modifying this list adjust `is_special`, `is_used_keyword`/`is_unused_keyword`,
24 // this should be rarely necessary though if the keywords are kept in alphabetic order.
26 // Special reserved identifiers used internally for elided lifetimes,
27 // unnamed method parameters, crate root module, error recovery etc.
30 DollarCrate: "$crate",
33 // Keywords that are used in stable Rust.
70 // Keywords that are used in unstable Rust or reserved for future use.
84 // Edition-specific keywords that are used in stable Rust.
85 Async: "async", // >= 2018 Edition only
86 Await: "await", // >= 2018 Edition only
87 Dyn: "dyn", // >= 2018 Edition only
89 // Edition-specific keywords that are used in unstable Rust or reserved for future use.
90 Try: "try", // >= 2018 Edition only
92 // Special lifetime names
93 UnderscoreLifetime: "'_",
94 StaticLifetime: "'static",
96 // Weak keywords, have special meaning only in specific contexts.
104 // Symbols that can be referred to with rustc_span::sym::*. The symbol is
105 // the stringified identifier unless otherwise specified (e.g.
106 // `proc_dash_macro` represents "proc-macro").
108 // As well as the symbols listed, there are symbols for the the strings
109 // "0", "1", ..., "9", which are accessible via `sym::integer`.
111 aarch64_target_feature,
115 abi_msp430_interrupt,
124 advanced_slice_patterns,
136 allow_internal_unsafe,
137 allow_internal_unstable,
138 allow_internal_unstable_backcompat_hack,
142 arbitrary_enum_discriminant,
143 arbitrary_self_types,
150 associated_type_bounds,
151 associated_type_defaults,
159 augmented_assignments,
160 automatically_derived,
161 avx512_target_feature,
166 bind_by_move_pattern_guards,
170 borrowck_graphviz_postflow,
171 borrowck_graphviz_preflow,
174 braced_empty_structs,
186 cfg_target_has_atomic,
187 cfg_target_thread_local,
195 closure_to_fn_coercion,
197 cmpxchg16b_target_feature,
204 conservative_impl_trait,
206 const_compare_raw_pointers,
214 const_in_array_repeat_expressions,
220 const_raw_ptr_to_usize_cast,
222 const_trait_bound_opt_out,
236 crate_visibility_modifier,
244 custom_inner_attributes,
245 custom_test_frameworks,
253 default_lib_allocator,
254 default_type_parameter_fallback,
256 delay_span_bug_from_inside_query,
271 document_private_items,
272 dotdoteq_in_patterns,
273 dotdot_in_tuple_patterns,
274 double_braced_crate: "{{crate}}",
275 double_braced_impl: "{{impl}}",
276 double_braced_misc: "{{misc}}",
277 double_braced_closure: "{{closure}}",
278 double_braced_constructor: "{{constructor}}",
279 double_braced_constant: "{{constant}}",
280 double_braced_opaque: "{{opaque}}",
282 dropck_parametricity,
298 exclusive_range_pattern,
299 exhaustive_integer_patterns,
305 extern_absolute_paths,
307 extern_crate_item_prelude,
318 field_init_shorthand,
341 generic_associated_types,
350 hexagon_target_feature,
352 homogeneous_aggregate,
366 if_while_or_patterns,
368 impl_header_lifetime_elision,
370 impl_trait_in_bindings,
378 inclusive_range_syntax,
379 infer_outlives_requirements,
380 infer_static_outlives_requirements,
387 irrefutable_let_patterns,
390 issue_5723_bootstrap,
391 issue_tracker_base_url,
393 item_context: "ItemContext",
412 link_llvm_intrinsics,
422 macro_at_most_once_rep,
425 macro_lifetime_matcher,
426 macro_literal_matcher,
437 match_beginning_vert,
438 match_default_bindings,
456 movbe_target_feature,
492 object_safe_for_dispatch,
494 omit_gdb_pretty_printer_section,
501 optin_builtin_traits,
511 overlapping_marker_traits,
516 panic_implementation,
536 poll_with_tls_context,
537 powerpc_target_feature,
538 precise_pointer_size_matching,
543 proc_dash_macro: "proc-macro",
545 proc_macro_attribute,
551 proc_macro_internals,
553 proc_macro_non_items,
554 proc_macro_path_invoc,
559 quad_precision_float,
574 reexport_test_harness_main,
586 re_rebalance_coherence,
604 rustc_allocator_nounwind,
605 rustc_allow_const_fn_ptr,
606 rustc_args_required_const,
610 rustc_const_unstable,
612 rustc_conversion_suggestion,
615 rustc_diagnostic_item,
616 rustc_diagnostic_macros,
619 rustc_dump_env_program_clauses,
620 rustc_dump_program_clauses,
621 rustc_dump_user_substs,
623 rustc_expected_cgu_reuse,
624 rustc_if_this_changed,
625 rustc_inherit_overflow_checks,
627 rustc_layout_scalar_valid_range_end,
628 rustc_layout_scalar_valid_range_start,
629 rustc_macro_transparency,
631 rustc_nonnull_optimization_guaranteed,
632 rustc_object_lifetime_default,
633 rustc_on_unimplemented,
636 rustc_partition_codegened,
637 rustc_partition_reused,
639 rustc_peek_definite_init,
640 rustc_peek_maybe_init,
641 rustc_peek_maybe_uninit,
642 rustc_peek_indirectly_mutable,
644 rustc_proc_macro_decls,
648 rustc_std_internal_symbol,
651 rustc_reservation_impl,
653 rustc_then_this_would_need,
657 rust_eh_unwind_resume,
659 rvalue_static_promotion,
683 sse4a_target_feature,
696 stmt_expr_attributes,
698 struct_field_attributes,
708 target_has_atomic_load_store,
713 termination_trait_test,
716 test_accepted_feature,
718 test_removed_feature,
739 type_alias_impl_trait,
744 type_alias_enum_variants,
756 underscore_const_names,
758 underscore_lifetimes,
760 universal_impl_trait,
763 unrestricted_attribute_tokens,
766 unsized_tuple_coercion,
782 visible_private_types,
797 #[derive(Copy, Clone, Eq, HashStable_Generic)]
805 /// Constructs a new identifier from a symbol and a span.
806 pub const fn new(name: Symbol, span: Span) -> Ident {
810 /// Constructs a new identifier with a dummy span.
812 pub const fn with_dummy_span(name: Symbol) -> Ident {
813 Ident::new(name, DUMMY_SP)
817 pub fn invalid() -> Ident {
818 Ident::with_dummy_span(kw::Invalid)
821 /// Maps a string to an identifier with a dummy span.
822 pub fn from_str(string: &str) -> Ident {
823 Ident::with_dummy_span(Symbol::intern(string))
826 /// Maps a string and a span to an identifier.
827 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
828 Ident::new(Symbol::intern(string), span)
831 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
832 pub fn with_span_pos(self, span: Span) -> Ident {
833 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
836 pub fn without_first_quote(self) -> Ident {
837 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
840 /// "Normalize" ident for use in comparisons using "item hygiene".
841 /// Identifiers with same string value become same if they came from the same "modern" macro
842 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
843 /// different "modern" macros.
844 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
845 pub fn modern(self) -> Ident {
846 Ident::new(self.name, self.span.modern())
849 /// "Normalize" ident for use in comparisons using "local variable hygiene".
850 /// Identifiers with same string value become same if they came from the same non-transparent
851 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
852 /// non-transparent macros.
853 /// Technically, this operation strips all transparent marks from ident's syntactic context.
854 pub fn modern_and_legacy(self) -> Ident {
855 Ident::new(self.name, self.span.modern_and_legacy())
858 /// Convert the name to a `SymbolStr`. This is a slowish operation because
859 /// it requires locking the symbol interner.
860 pub fn as_str(self) -> SymbolStr {
865 impl PartialEq for Ident {
866 fn eq(&self, rhs: &Self) -> bool {
867 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
871 impl Hash for Ident {
872 fn hash<H: Hasher>(&self, state: &mut H) {
873 self.name.hash(state);
874 self.span.ctxt().hash(state);
878 impl fmt::Debug for Ident {
879 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
880 if self.is_raw_guess() {
883 write!(f, "{}{:?}", self.name, self.span.ctxt())
887 impl fmt::Display for Ident {
888 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
889 if self.is_raw_guess() {
892 fmt::Display::fmt(&self.name, f)
896 impl UseSpecializedEncodable for Ident {
897 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
898 s.emit_struct("Ident", 2, |s| {
899 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
900 s.emit_struct_field("span", 1, |s| self.span.encode(s))
905 impl UseSpecializedDecodable for Ident {
906 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
907 d.read_struct("Ident", 2, |d| {
909 name: d.read_struct_field("name", 0, Decodable::decode)?,
910 span: d.read_struct_field("span", 1, Decodable::decode)?,
916 /// An interned string.
918 /// Internally, a `Symbol` is implemented as an index, and all operations
919 /// (including hashing, equality, and ordering) operate on that index. The use
920 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
921 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
923 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
924 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
925 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
926 pub struct Symbol(SymbolIndex);
928 rustc_index::newtype_index! {
929 pub struct SymbolIndex { .. }
933 const fn new(n: u32) -> Self {
934 Symbol(SymbolIndex::from_u32_const(n))
937 /// Maps a string to its interned representation.
938 pub fn intern(string: &str) -> Self {
939 with_interner(|interner| interner.intern(string))
942 /// Access the symbol's chars. This is a slowish operation because it
943 /// requires locking the symbol interner.
944 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
945 with_interner(|interner| f(interner.get(self)))
948 /// Convert to a `SymbolStr`. This is a slowish operation because it
949 /// requires locking the symbol interner.
950 pub fn as_str(self) -> SymbolStr {
951 with_interner(|interner| unsafe {
952 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
956 pub fn as_u32(self) -> u32 {
961 impl fmt::Debug for Symbol {
962 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
963 self.with(|str| fmt::Debug::fmt(&str, f))
967 impl fmt::Display for Symbol {
968 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
969 self.with(|str| fmt::Display::fmt(&str, f))
973 impl Encodable for Symbol {
974 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
975 self.with(|string| s.emit_str(string))
979 impl Decodable for Symbol {
980 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
981 Ok(Symbol::intern(&d.read_str()?))
985 impl<CTX> HashStable<CTX> for Symbol {
987 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
988 self.as_str().hash_stable(hcx, hasher);
992 impl<CTX> ToStableHashKey<CTX> for Symbol {
993 type KeyType = SymbolStr;
996 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1001 // The `&'static str`s in this type actually point into the arena.
1003 pub struct Interner {
1004 arena: DroplessArena,
1005 names: FxHashMap<&'static str, Symbol>,
1006 strings: Vec<&'static str>,
1010 fn prefill(init: &[&'static str]) -> Self {
1012 strings: init.into(),
1013 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1014 ..Default::default()
1018 pub fn intern(&mut self, string: &str) -> Symbol {
1019 if let Some(&name) = self.names.get(string) {
1023 let name = Symbol::new(self.strings.len() as u32);
1025 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1028 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1029 // It is safe to extend the arena allocation to `'static` because we only access
1030 // these while the arena is still alive.
1031 let string: &'static str = unsafe { &*(string as *const str) };
1032 self.strings.push(string);
1033 self.names.insert(string, name);
1037 // Get the symbol as a string. `Symbol::as_str()` should be used in
1038 // preference to this function.
1039 pub fn get(&self, symbol: Symbol) -> &str {
1040 self.strings[symbol.0.as_usize()]
1044 // This module has a very short name because it's used a lot.
1050 // This module has a very short name because it's used a lot.
1053 use std::convert::TryInto;
1057 // Get the symbol for an integer. The first few non-negative integers each
1058 // have a static symbol and therefore are fast.
1059 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1060 if let Result::Ok(idx) = n.try_into() {
1061 if let Option::Some(&sym) = digits_array.get(idx) {
1065 Symbol::intern(&n.to_string())
1070 fn is_used_keyword_2018(self) -> bool {
1071 self >= kw::Async && self <= kw::Dyn
1074 fn is_unused_keyword_2018(self) -> bool {
1078 /// Used for sanity checking rustdoc keyword sections.
1079 pub fn is_doc_keyword(self) -> bool {
1083 /// A keyword or reserved identifier that can be used as a path segment.
1084 pub fn is_path_segment_keyword(self) -> bool {
1086 || self == kw::SelfLower
1087 || self == kw::SelfUpper
1088 || self == kw::Crate
1089 || self == kw::PathRoot
1090 || self == kw::DollarCrate
1093 /// Returns `true` if the symbol is `true` or `false`.
1094 pub fn is_bool_lit(self) -> bool {
1095 self == kw::True || self == kw::False
1098 /// This symbol can be a raw identifier.
1099 pub fn can_be_raw(self) -> bool {
1100 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1105 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1106 // unnamed method parameters, crate root module, error recovery etc.
1107 pub fn is_special(self) -> bool {
1108 self.name <= kw::Underscore
1111 /// Returns `true` if the token is a keyword used in the language.
1112 pub fn is_used_keyword(self) -> bool {
1113 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1114 self.name >= kw::As && self.name <= kw::While
1115 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1118 /// Returns `true` if the token is a keyword reserved for possible future use.
1119 pub fn is_unused_keyword(self) -> bool {
1120 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1121 self.name >= kw::Abstract && self.name <= kw::Yield
1122 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1125 /// Returns `true` if the token is either a special identifier or a keyword.
1126 pub fn is_reserved(self) -> bool {
1127 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1130 /// A keyword or reserved identifier that can be used as a path segment.
1131 pub fn is_path_segment_keyword(self) -> bool {
1132 self.name.is_path_segment_keyword()
1135 /// We see this identifier in a normal identifier position, like variable name or a type.
1136 /// How was it written originally? Did it use the raw form? Let's try to guess.
1137 pub fn is_raw_guess(self) -> bool {
1138 self.name.can_be_raw() && self.is_reserved()
1143 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1144 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1147 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1148 /// be accessed. It deliberately has limited functionality and should only be
1149 /// used for temporary values.
1151 /// Because the interner outlives any thread which uses this type, we can
1152 /// safely treat `string` which points to interner data, as an immortal string,
1153 /// as long as this type never crosses between threads.
1155 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1156 // by creating a new thread right after constructing the interner.
1157 #[derive(Clone, Eq, PartialOrd, Ord)]
1158 pub struct SymbolStr {
1159 string: &'static str,
1162 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1164 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1165 fn eq(&self, other: &T) -> bool {
1166 self.string == other.deref()
1170 impl !Send for SymbolStr {}
1171 impl !Sync for SymbolStr {}
1173 /// This impl means that if `ss` is a `SymbolStr`:
1174 /// - `*ss` is a `str`;
1175 /// - `&*ss` is a `&str`;
1176 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1177 /// function expecting a `&str`.
1178 impl std::ops::Deref for SymbolStr {
1181 fn deref(&self) -> &str {
1186 impl fmt::Debug for SymbolStr {
1187 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1188 fmt::Debug::fmt(self.string, f)
1192 impl fmt::Display for SymbolStr {
1193 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1194 fmt::Display::fmt(self.string, f)
1198 impl<CTX> HashStable<CTX> for SymbolStr {
1200 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1201 self.string.hash_stable(hcx, hasher)
1205 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1206 type KeyType = SymbolStr;
1209 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {