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_format,
171 borrowck_graphviz_postflow,
172 borrowck_graphviz_preflow,
175 braced_empty_structs,
187 cfg_target_has_atomic,
188 cfg_target_thread_local,
196 closure_to_fn_coercion,
198 cmpxchg16b_target_feature,
205 conservative_impl_trait,
207 const_compare_raw_pointers,
215 const_in_array_repeat_expressions,
221 const_raw_ptr_to_usize_cast,
223 const_trait_bound_opt_out,
237 crate_visibility_modifier,
245 custom_inner_attributes,
246 custom_test_frameworks,
254 default_lib_allocator,
255 default_type_parameter_fallback,
257 delay_span_bug_from_inside_query,
272 document_private_items,
273 dotdoteq_in_patterns,
274 dotdot_in_tuple_patterns,
275 double_braced_crate: "{{crate}}",
276 double_braced_impl: "{{impl}}",
277 double_braced_misc: "{{misc}}",
278 double_braced_closure: "{{closure}}",
279 double_braced_constructor: "{{constructor}}",
280 double_braced_constant: "{{constant}}",
281 double_braced_opaque: "{{opaque}}",
283 dropck_parametricity,
299 exclusive_range_pattern,
300 exhaustive_integer_patterns,
306 extern_absolute_paths,
308 extern_crate_item_prelude,
319 field_init_shorthand,
343 generic_associated_types,
348 half_open_range_patterns,
353 hexagon_target_feature,
355 homogeneous_aggregate,
369 if_while_or_patterns,
371 impl_header_lifetime_elision,
373 impl_trait_in_bindings,
381 inclusive_range_syntax,
382 infer_outlives_requirements,
383 infer_static_outlives_requirements,
390 irrefutable_let_patterns,
393 issue_5723_bootstrap,
394 issue_tracker_base_url,
396 item_context: "ItemContext",
415 link_llvm_intrinsics,
425 macro_at_most_once_rep,
428 macro_lifetime_matcher,
429 macro_literal_matcher,
440 match_beginning_vert,
441 match_default_bindings,
459 movbe_target_feature,
495 object_safe_for_dispatch,
497 omit_gdb_pretty_printer_section,
504 optin_builtin_traits,
514 overlapping_marker_traits,
519 panic_implementation,
539 poll_with_tls_context,
540 powerpc_target_feature,
541 precise_pointer_size_matching,
546 proc_dash_macro: "proc-macro",
548 proc_macro_attribute,
554 proc_macro_internals,
556 proc_macro_non_items,
557 proc_macro_path_invoc,
562 quad_precision_float,
577 reexport_test_harness_main,
589 re_rebalance_coherence,
607 rustc_allocator_nounwind,
608 rustc_allow_const_fn_ptr,
609 rustc_args_required_const,
613 rustc_const_unstable,
615 rustc_conversion_suggestion,
618 rustc_diagnostic_item,
619 rustc_diagnostic_macros,
622 rustc_dump_env_program_clauses,
623 rustc_dump_program_clauses,
624 rustc_dump_user_substs,
626 rustc_expected_cgu_reuse,
627 rustc_if_this_changed,
628 rustc_inherit_overflow_checks,
630 rustc_layout_scalar_valid_range_end,
631 rustc_layout_scalar_valid_range_start,
632 rustc_macro_transparency,
634 rustc_nonnull_optimization_guaranteed,
635 rustc_object_lifetime_default,
636 rustc_on_unimplemented,
639 rustc_partition_codegened,
640 rustc_partition_reused,
642 rustc_peek_definite_init,
643 rustc_peek_maybe_init,
644 rustc_peek_maybe_uninit,
645 rustc_peek_indirectly_mutable,
647 rustc_proc_macro_decls,
651 rustc_std_internal_symbol,
654 rustc_reservation_impl,
656 rustc_then_this_would_need,
660 rust_eh_unwind_resume,
662 rvalue_static_promotion,
686 sse4a_target_feature,
699 stmt_expr_attributes,
701 struct_field_attributes,
711 target_has_atomic_load_store,
716 termination_trait_test,
719 test_accepted_feature,
721 test_removed_feature,
743 type_alias_impl_trait,
748 type_alias_enum_variants,
760 underscore_const_names,
762 underscore_lifetimes,
764 universal_impl_trait,
767 unrestricted_attribute_tokens,
770 unsized_tuple_coercion,
786 visible_private_types,
801 #[derive(Copy, Clone, Eq, HashStable_Generic)]
809 /// Constructs a new identifier from a symbol and a span.
810 pub const fn new(name: Symbol, span: Span) -> Ident {
814 /// Constructs a new identifier with a dummy span.
816 pub const fn with_dummy_span(name: Symbol) -> Ident {
817 Ident::new(name, DUMMY_SP)
821 pub fn invalid() -> Ident {
822 Ident::with_dummy_span(kw::Invalid)
825 /// Maps a string to an identifier with a dummy span.
826 pub fn from_str(string: &str) -> Ident {
827 Ident::with_dummy_span(Symbol::intern(string))
830 /// Maps a string and a span to an identifier.
831 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
832 Ident::new(Symbol::intern(string), span)
835 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
836 pub fn with_span_pos(self, span: Span) -> Ident {
837 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
840 pub fn without_first_quote(self) -> Ident {
841 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
844 /// "Normalize" ident for use in comparisons using "item hygiene".
845 /// Identifiers with same string value become same if they came from the same "modern" macro
846 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
847 /// different "modern" macros.
848 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
849 pub fn modern(self) -> Ident {
850 Ident::new(self.name, self.span.modern())
853 /// "Normalize" ident for use in comparisons using "local variable hygiene".
854 /// Identifiers with same string value become same if they came from the same non-transparent
855 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
856 /// non-transparent macros.
857 /// Technically, this operation strips all transparent marks from ident's syntactic context.
858 pub fn modern_and_legacy(self) -> Ident {
859 Ident::new(self.name, self.span.modern_and_legacy())
862 /// Convert the name to a `SymbolStr`. This is a slowish operation because
863 /// it requires locking the symbol interner.
864 pub fn as_str(self) -> SymbolStr {
869 impl PartialEq for Ident {
870 fn eq(&self, rhs: &Self) -> bool {
871 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
875 impl Hash for Ident {
876 fn hash<H: Hasher>(&self, state: &mut H) {
877 self.name.hash(state);
878 self.span.ctxt().hash(state);
882 impl fmt::Debug for Ident {
883 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
884 if self.is_raw_guess() {
887 write!(f, "{}{:?}", self.name, self.span.ctxt())
891 impl fmt::Display for Ident {
892 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
893 if self.is_raw_guess() {
896 fmt::Display::fmt(&self.name, f)
900 impl UseSpecializedEncodable for Ident {
901 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
902 s.emit_struct("Ident", 2, |s| {
903 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
904 s.emit_struct_field("span", 1, |s| self.span.encode(s))
909 impl UseSpecializedDecodable for Ident {
910 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
911 d.read_struct("Ident", 2, |d| {
913 name: d.read_struct_field("name", 0, Decodable::decode)?,
914 span: d.read_struct_field("span", 1, Decodable::decode)?,
920 /// An interned string.
922 /// Internally, a `Symbol` is implemented as an index, and all operations
923 /// (including hashing, equality, and ordering) operate on that index. The use
924 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
925 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
927 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
928 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
929 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
930 pub struct Symbol(SymbolIndex);
932 rustc_index::newtype_index! {
933 pub struct SymbolIndex { .. }
937 const fn new(n: u32) -> Self {
938 Symbol(SymbolIndex::from_u32_const(n))
941 /// Maps a string to its interned representation.
942 pub fn intern(string: &str) -> Self {
943 with_interner(|interner| interner.intern(string))
946 /// Access the symbol's chars. This is a slowish operation because it
947 /// requires locking the symbol interner.
948 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
949 with_interner(|interner| f(interner.get(self)))
952 /// Convert to a `SymbolStr`. This is a slowish operation because it
953 /// requires locking the symbol interner.
954 pub fn as_str(self) -> SymbolStr {
955 with_interner(|interner| unsafe {
956 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
960 pub fn as_u32(self) -> u32 {
965 impl fmt::Debug for Symbol {
966 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
967 self.with(|str| fmt::Debug::fmt(&str, f))
971 impl fmt::Display for Symbol {
972 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
973 self.with(|str| fmt::Display::fmt(&str, f))
977 impl Encodable for Symbol {
978 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
979 self.with(|string| s.emit_str(string))
983 impl Decodable for Symbol {
984 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
985 Ok(Symbol::intern(&d.read_str()?))
989 impl<CTX> HashStable<CTX> for Symbol {
991 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
992 self.as_str().hash_stable(hcx, hasher);
996 impl<CTX> ToStableHashKey<CTX> for Symbol {
997 type KeyType = SymbolStr;
1000 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1005 // The `&'static str`s in this type actually point into the arena.
1007 pub struct Interner {
1008 arena: DroplessArena,
1009 names: FxHashMap<&'static str, Symbol>,
1010 strings: Vec<&'static str>,
1014 fn prefill(init: &[&'static str]) -> Self {
1016 strings: init.into(),
1017 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1018 ..Default::default()
1022 pub fn intern(&mut self, string: &str) -> Symbol {
1023 if let Some(&name) = self.names.get(string) {
1027 let name = Symbol::new(self.strings.len() as u32);
1029 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1032 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1033 // It is safe to extend the arena allocation to `'static` because we only access
1034 // these while the arena is still alive.
1035 let string: &'static str = unsafe { &*(string as *const str) };
1036 self.strings.push(string);
1037 self.names.insert(string, name);
1041 // Get the symbol as a string. `Symbol::as_str()` should be used in
1042 // preference to this function.
1043 pub fn get(&self, symbol: Symbol) -> &str {
1044 self.strings[symbol.0.as_usize()]
1048 // This module has a very short name because it's used a lot.
1054 // This module has a very short name because it's used a lot.
1055 #[allow(rustc::default_hash_types)]
1058 use std::convert::TryInto;
1062 // Get the symbol for an integer. The first few non-negative integers each
1063 // have a static symbol and therefore are fast.
1064 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1065 if let Result::Ok(idx) = n.try_into() {
1066 if let Option::Some(&sym) = digits_array.get(idx) {
1070 Symbol::intern(&n.to_string())
1075 fn is_used_keyword_2018(self) -> bool {
1076 self >= kw::Async && self <= kw::Dyn
1079 fn is_unused_keyword_2018(self) -> bool {
1083 /// Used for sanity checking rustdoc keyword sections.
1084 pub fn is_doc_keyword(self) -> bool {
1088 /// A keyword or reserved identifier that can be used as a path segment.
1089 pub fn is_path_segment_keyword(self) -> bool {
1091 || self == kw::SelfLower
1092 || self == kw::SelfUpper
1093 || self == kw::Crate
1094 || self == kw::PathRoot
1095 || self == kw::DollarCrate
1098 /// Returns `true` if the symbol is `true` or `false`.
1099 pub fn is_bool_lit(self) -> bool {
1100 self == kw::True || self == kw::False
1103 /// This symbol can be a raw identifier.
1104 pub fn can_be_raw(self) -> bool {
1105 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1110 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1111 // unnamed method parameters, crate root module, error recovery etc.
1112 pub fn is_special(self) -> bool {
1113 self.name <= kw::Underscore
1116 /// Returns `true` if the token is a keyword used in the language.
1117 pub fn is_used_keyword(self) -> bool {
1118 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1119 self.name >= kw::As && self.name <= kw::While
1120 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1123 /// Returns `true` if the token is a keyword reserved for possible future use.
1124 pub fn is_unused_keyword(self) -> bool {
1125 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1126 self.name >= kw::Abstract && self.name <= kw::Yield
1127 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1130 /// Returns `true` if the token is either a special identifier or a keyword.
1131 pub fn is_reserved(self) -> bool {
1132 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1135 /// A keyword or reserved identifier that can be used as a path segment.
1136 pub fn is_path_segment_keyword(self) -> bool {
1137 self.name.is_path_segment_keyword()
1140 /// We see this identifier in a normal identifier position, like variable name or a type.
1141 /// How was it written originally? Did it use the raw form? Let's try to guess.
1142 pub fn is_raw_guess(self) -> bool {
1143 self.name.can_be_raw() && self.is_reserved()
1148 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1149 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1152 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1153 /// be accessed. It deliberately has limited functionality and should only be
1154 /// used for temporary values.
1156 /// Because the interner outlives any thread which uses this type, we can
1157 /// safely treat `string` which points to interner data, as an immortal string,
1158 /// as long as this type never crosses between threads.
1160 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1161 // by creating a new thread right after constructing the interner.
1162 #[derive(Clone, Eq, PartialOrd, Ord)]
1163 pub struct SymbolStr {
1164 string: &'static str,
1167 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1169 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1170 fn eq(&self, other: &T) -> bool {
1171 self.string == other.deref()
1175 impl !Send for SymbolStr {}
1176 impl !Sync for SymbolStr {}
1178 /// This impl means that if `ss` is a `SymbolStr`:
1179 /// - `*ss` is a `str`;
1180 /// - `&*ss` is a `&str`;
1181 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1182 /// function expecting a `&str`.
1183 impl std::ops::Deref for SymbolStr {
1186 fn deref(&self) -> &str {
1191 impl fmt::Debug for SymbolStr {
1192 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1193 fmt::Debug::fmt(self.string, f)
1197 impl fmt::Display for SymbolStr {
1198 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1199 fmt::Display::fmt(self.string, f)
1203 impl<CTX> HashStable<CTX> for SymbolStr {
1205 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1206 self.string.hash_stable(hcx, hasher)
1210 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1211 type KeyType = SymbolStr;
1214 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {