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_index::vec::Idx;
9 use rustc_macros::{symbols, HashStable_Generic};
10 use rustc_serialize::{Decodable, Decoder, Encodable, Encoder};
11 use rustc_serialize::{UseSpecializedDecodable, UseSpecializedEncodable};
13 use std::cmp::{Ord, PartialEq, PartialOrd};
15 use std::hash::{Hash, Hasher};
18 use crate::{Span, DUMMY_SP, GLOBALS};
24 // After modifying this list adjust `is_special`, `is_used_keyword`/`is_unused_keyword`,
25 // this should be rarely necessary though if the keywords are kept in alphabetic order.
27 // Special reserved identifiers used internally for elided lifetimes,
28 // unnamed method parameters, crate root module, error recovery etc.
31 DollarCrate: "$crate",
34 // Keywords that are used in stable Rust.
71 // Keywords that are used in unstable Rust or reserved for future use.
85 // Edition-specific keywords that are used in stable Rust.
86 Async: "async", // >= 2018 Edition only
87 Await: "await", // >= 2018 Edition only
88 Dyn: "dyn", // >= 2018 Edition only
90 // Edition-specific keywords that are used in unstable Rust or reserved for future use.
91 Try: "try", // >= 2018 Edition only
93 // Special lifetime names
94 UnderscoreLifetime: "'_",
95 StaticLifetime: "'static",
97 // Weak keywords, have special meaning only in specific contexts.
105 // Symbols that can be referred to with rustc_span::sym::*. The symbol is
106 // the stringified identifier unless otherwise specified (e.g.
107 // `proc_dash_macro` represents "proc-macro").
109 // As well as the symbols listed, there are symbols for the the strings
110 // "0", "1", ..., "9", which are accessible via `sym::integer`.
112 aarch64_target_feature,
116 abi_msp430_interrupt,
125 advanced_slice_patterns,
137 allow_internal_unsafe,
138 allow_internal_unstable,
139 allow_internal_unstable_backcompat_hack,
143 arbitrary_enum_discriminant,
144 arbitrary_self_types,
151 associated_type_bounds,
152 associated_type_defaults,
160 augmented_assignments,
161 automatically_derived,
162 avx512_target_feature,
167 bind_by_move_pattern_guards,
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,
235 crate_visibility_modifier,
243 custom_inner_attributes,
244 custom_test_frameworks,
252 default_lib_allocator,
253 default_type_parameter_fallback,
255 delay_span_bug_from_inside_query,
270 document_private_items,
271 dotdoteq_in_patterns,
272 dotdot_in_tuple_patterns,
273 double_braced_crate: "{{crate}}",
274 double_braced_impl: "{{impl}}",
275 double_braced_misc: "{{misc}}",
276 double_braced_closure: "{{closure}}",
277 double_braced_constructor: "{{constructor}}",
278 double_braced_constant: "{{constant}}",
279 double_braced_opaque: "{{opaque}}",
281 dropck_parametricity,
297 exclusive_range_pattern,
298 exhaustive_integer_patterns,
304 extern_absolute_paths,
306 extern_crate_item_prelude,
317 field_init_shorthand,
340 generic_associated_types,
349 hexagon_target_feature,
351 homogeneous_aggregate,
365 if_while_or_patterns,
367 impl_header_lifetime_elision,
369 impl_trait_in_bindings,
377 inclusive_range_syntax,
378 infer_outlives_requirements,
379 infer_static_outlives_requirements,
386 irrefutable_let_patterns,
389 issue_5723_bootstrap,
390 issue_tracker_base_url,
392 item_context: "ItemContext",
411 link_llvm_intrinsics,
421 macro_at_most_once_rep,
424 macro_lifetime_matcher,
425 macro_literal_matcher,
436 match_beginning_vert,
437 match_default_bindings,
455 movbe_target_feature,
491 object_safe_for_dispatch,
493 omit_gdb_pretty_printer_section,
500 optin_builtin_traits,
510 overlapping_marker_traits,
515 panic_implementation,
535 poll_with_tls_context,
536 powerpc_target_feature,
537 precise_pointer_size_matching,
542 proc_dash_macro: "proc-macro",
544 proc_macro_attribute,
550 proc_macro_internals,
552 proc_macro_non_items,
553 proc_macro_path_invoc,
558 quad_precision_float,
573 reexport_test_harness_main,
585 re_rebalance_coherence,
603 rustc_allocator_nounwind,
604 rustc_allow_const_fn_ptr,
605 rustc_args_required_const,
609 rustc_const_unstable,
611 rustc_conversion_suggestion,
614 rustc_diagnostic_item,
615 rustc_diagnostic_macros,
618 rustc_dump_env_program_clauses,
619 rustc_dump_program_clauses,
620 rustc_dump_user_substs,
622 rustc_expected_cgu_reuse,
623 rustc_if_this_changed,
624 rustc_inherit_overflow_checks,
626 rustc_layout_scalar_valid_range_end,
627 rustc_layout_scalar_valid_range_start,
628 rustc_macro_transparency,
630 rustc_nonnull_optimization_guaranteed,
631 rustc_object_lifetime_default,
632 rustc_on_unimplemented,
635 rustc_partition_codegened,
636 rustc_partition_reused,
638 rustc_peek_definite_init,
639 rustc_peek_maybe_init,
640 rustc_peek_maybe_uninit,
641 rustc_peek_indirectly_mutable,
643 rustc_proc_macro_decls,
647 rustc_std_internal_symbol,
650 rustc_reservation_impl,
652 rustc_then_this_would_need,
656 rust_eh_unwind_resume,
658 rvalue_static_promotion,
682 sse4a_target_feature,
695 stmt_expr_attributes,
697 struct_field_attributes,
707 target_has_atomic_load_store,
712 termination_trait_test,
715 test_accepted_feature,
717 test_removed_feature,
738 type_alias_impl_trait,
743 type_alias_enum_variants,
755 underscore_const_names,
757 underscore_lifetimes,
759 universal_impl_trait,
762 unrestricted_attribute_tokens,
765 unsized_tuple_coercion,
781 visible_private_types,
796 #[derive(Copy, Clone, Eq, HashStable_Generic)]
804 /// Constructs a new identifier from a symbol and a span.
805 pub const fn new(name: Symbol, span: Span) -> Ident {
809 /// Constructs a new identifier with a dummy span.
811 pub const fn with_dummy_span(name: Symbol) -> Ident {
812 Ident::new(name, DUMMY_SP)
816 pub fn invalid() -> Ident {
817 Ident::with_dummy_span(kw::Invalid)
820 /// Maps a string to an identifier with a dummy span.
821 pub fn from_str(string: &str) -> Ident {
822 Ident::with_dummy_span(Symbol::intern(string))
825 /// Maps a string and a span to an identifier.
826 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
827 Ident::new(Symbol::intern(string), span)
830 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
831 pub fn with_span_pos(self, span: Span) -> Ident {
832 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
835 pub fn without_first_quote(self) -> Ident {
836 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
839 /// "Normalize" ident for use in comparisons using "item hygiene".
840 /// Identifiers with same string value become same if they came from the same "modern" macro
841 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
842 /// different "modern" macros.
843 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
844 pub fn modern(self) -> Ident {
845 Ident::new(self.name, self.span.modern())
848 /// "Normalize" ident for use in comparisons using "local variable hygiene".
849 /// Identifiers with same string value become same if they came from the same non-transparent
850 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
851 /// non-transparent macros.
852 /// Technically, this operation strips all transparent marks from ident's syntactic context.
853 pub fn modern_and_legacy(self) -> Ident {
854 Ident::new(self.name, self.span.modern_and_legacy())
857 /// Convert the name to a `SymbolStr`. This is a slowish operation because
858 /// it requires locking the symbol interner.
859 pub fn as_str(self) -> SymbolStr {
864 impl PartialEq for Ident {
865 fn eq(&self, rhs: &Self) -> bool {
866 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
870 impl Hash for Ident {
871 fn hash<H: Hasher>(&self, state: &mut H) {
872 self.name.hash(state);
873 self.span.ctxt().hash(state);
877 impl fmt::Debug for Ident {
878 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
879 if self.is_raw_guess() {
882 write!(f, "{}{:?}", self.name, self.span.ctxt())
886 impl fmt::Display for Ident {
887 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
888 if self.is_raw_guess() {
891 fmt::Display::fmt(&self.name, f)
895 impl UseSpecializedEncodable for Ident {
896 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
897 s.emit_struct("Ident", 2, |s| {
898 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
899 s.emit_struct_field("span", 1, |s| self.span.encode(s))
904 impl UseSpecializedDecodable for Ident {
905 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
906 d.read_struct("Ident", 2, |d| {
908 name: d.read_struct_field("name", 0, Decodable::decode)?,
909 span: d.read_struct_field("span", 1, Decodable::decode)?,
915 /// An interned string.
917 /// Internally, a `Symbol` is implemented as an index, and all operations
918 /// (including hashing, equality, and ordering) operate on that index. The use
919 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
920 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
922 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
923 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
924 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
925 pub struct Symbol(SymbolIndex);
927 rustc_index::newtype_index! {
928 pub struct SymbolIndex { .. }
932 const fn new(n: u32) -> Self {
933 Symbol(SymbolIndex::from_u32_const(n))
936 /// Maps a string to its interned representation.
937 pub fn intern(string: &str) -> Self {
938 with_interner(|interner| interner.intern(string))
941 /// Access the symbol's chars. This is a slowish operation because it
942 /// requires locking the symbol interner.
943 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
944 with_interner(|interner| f(interner.get(self)))
947 /// Convert to a `SymbolStr`. This is a slowish operation because it
948 /// requires locking the symbol interner.
949 pub fn as_str(self) -> SymbolStr {
950 with_interner(|interner| unsafe {
951 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
955 pub fn as_u32(self) -> u32 {
960 impl fmt::Debug for Symbol {
961 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
962 self.with(|str| fmt::Debug::fmt(&str, f))
966 impl fmt::Display for Symbol {
967 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
968 self.with(|str| fmt::Display::fmt(&str, f))
972 impl Encodable for Symbol {
973 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
974 self.with(|string| s.emit_str(string))
978 impl Decodable for Symbol {
979 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
980 Ok(Symbol::intern(&d.read_str()?))
984 impl<CTX> HashStable<CTX> for Symbol {
986 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
987 self.as_str().hash_stable(hcx, hasher);
991 impl<CTX> ToStableHashKey<CTX> for Symbol {
992 type KeyType = SymbolStr;
995 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1000 // The `&'static str`s in this type actually point into the arena.
1002 pub struct Interner {
1003 arena: DroplessArena,
1004 names: FxHashMap<&'static str, Symbol>,
1005 strings: Vec<&'static str>,
1009 fn prefill(init: &[&'static str]) -> Self {
1011 strings: init.into(),
1012 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1013 ..Default::default()
1017 pub fn intern(&mut self, string: &str) -> Symbol {
1018 if let Some(&name) = self.names.get(string) {
1022 let name = Symbol::new(self.strings.len() as u32);
1024 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1027 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1028 // It is safe to extend the arena allocation to `'static` because we only access
1029 // these while the arena is still alive.
1030 let string: &'static str = unsafe { &*(string as *const str) };
1031 self.strings.push(string);
1032 self.names.insert(string, name);
1036 // Get the symbol as a string. `Symbol::as_str()` should be used in
1037 // preference to this function.
1038 pub fn get(&self, symbol: Symbol) -> &str {
1039 self.strings[symbol.0.as_usize()]
1043 // This module has a very short name because it's used a lot.
1049 // This module has a very short name because it's used a lot.
1052 use std::convert::TryInto;
1056 // Get the symbol for an integer. The first few non-negative integers each
1057 // have a static symbol and therefore are fast.
1058 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1059 if let Result::Ok(idx) = n.try_into() {
1060 if let Option::Some(&sym) = digits_array.get(idx) {
1064 Symbol::intern(&n.to_string())
1069 fn is_used_keyword_2018(self) -> bool {
1070 self >= kw::Async && self <= kw::Dyn
1073 fn is_unused_keyword_2018(self) -> bool {
1077 /// Used for sanity checking rustdoc keyword sections.
1078 pub fn is_doc_keyword(self) -> bool {
1082 /// A keyword or reserved identifier that can be used as a path segment.
1083 pub fn is_path_segment_keyword(self) -> bool {
1085 || self == kw::SelfLower
1086 || self == kw::SelfUpper
1087 || self == kw::Crate
1088 || self == kw::PathRoot
1089 || self == kw::DollarCrate
1092 /// Returns `true` if the symbol is `true` or `false`.
1093 pub fn is_bool_lit(self) -> bool {
1094 self == kw::True || self == kw::False
1097 /// This symbol can be a raw identifier.
1098 pub fn can_be_raw(self) -> bool {
1099 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1104 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1105 // unnamed method parameters, crate root module, error recovery etc.
1106 pub fn is_special(self) -> bool {
1107 self.name <= kw::Underscore
1110 /// Returns `true` if the token is a keyword used in the language.
1111 pub fn is_used_keyword(self) -> bool {
1112 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1113 self.name >= kw::As && self.name <= kw::While
1114 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1117 /// Returns `true` if the token is a keyword reserved for possible future use.
1118 pub fn is_unused_keyword(self) -> bool {
1119 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1120 self.name >= kw::Abstract && self.name <= kw::Yield
1121 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1124 /// Returns `true` if the token is either a special identifier or a keyword.
1125 pub fn is_reserved(self) -> bool {
1126 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1129 /// A keyword or reserved identifier that can be used as a path segment.
1130 pub fn is_path_segment_keyword(self) -> bool {
1131 self.name.is_path_segment_keyword()
1134 /// We see this identifier in a normal identifier position, like variable name or a type.
1135 /// How was it written originally? Did it use the raw form? Let's try to guess.
1136 pub fn is_raw_guess(self) -> bool {
1137 self.name.can_be_raw() && self.is_reserved()
1142 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1143 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1146 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1147 /// be accessed. It deliberately has limited functionality and should only be
1148 /// used for temporary values.
1150 /// Because the interner outlives any thread which uses this type, we can
1151 /// safely treat `string` which points to interner data, as an immortal string,
1152 /// as long as this type never crosses between threads.
1154 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1155 // by creating a new thread right after constructing the interner.
1156 #[derive(Clone, Eq, PartialOrd, Ord)]
1157 pub struct SymbolStr {
1158 string: &'static str,
1161 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1163 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1164 fn eq(&self, other: &T) -> bool {
1165 self.string == other.deref()
1169 impl !Send for SymbolStr {}
1170 impl !Sync for SymbolStr {}
1172 /// This impl means that if `ss` is a `SymbolStr`:
1173 /// - `*ss` is a `str`;
1174 /// - `&*ss` is a `&str`;
1175 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1176 /// function expecting a `&str`.
1177 impl std::ops::Deref for SymbolStr {
1180 fn deref(&self) -> &str {
1185 impl fmt::Debug for SymbolStr {
1186 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1187 fmt::Debug::fmt(self.string, f)
1191 impl fmt::Display for SymbolStr {
1192 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1193 fmt::Display::fmt(self.string, f)
1197 impl<CTX> HashStable<CTX> for SymbolStr {
1199 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1200 self.string.hash_stable(hcx, hasher)
1204 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1205 type KeyType = SymbolStr;
1208 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {