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,
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_format,
172 borrowck_graphviz_postflow,
173 borrowck_graphviz_preflow,
176 braced_empty_structs,
188 cfg_target_has_atomic,
189 cfg_target_thread_local,
197 closure_to_fn_coercion,
199 cmpxchg16b_target_feature,
206 conservative_impl_trait,
208 const_compare_raw_pointers,
216 const_in_array_repeat_expressions,
222 const_raw_ptr_to_usize_cast,
224 const_trait_bound_opt_out,
238 crate_visibility_modifier,
246 custom_inner_attributes,
247 custom_test_frameworks,
255 default_lib_allocator,
256 default_type_parameter_fallback,
258 delay_span_bug_from_inside_query,
273 document_private_items,
274 dotdoteq_in_patterns,
275 dotdot_in_tuple_patterns,
276 double_braced_crate: "{{crate}}",
277 double_braced_impl: "{{impl}}",
278 double_braced_misc: "{{misc}}",
279 double_braced_closure: "{{closure}}",
280 double_braced_constructor: "{{constructor}}",
281 double_braced_constant: "{{constant}}",
282 double_braced_opaque: "{{opaque}}",
284 dropck_parametricity,
300 exclusive_range_pattern,
301 exhaustive_integer_patterns,
307 extern_absolute_paths,
309 extern_crate_item_prelude,
320 field_init_shorthand,
344 generic_associated_types,
349 half_open_range_patterns,
354 hexagon_target_feature,
356 homogeneous_aggregate,
370 if_while_or_patterns,
372 impl_header_lifetime_elision,
374 impl_trait_in_bindings,
382 inclusive_range_syntax,
383 infer_outlives_requirements,
384 infer_static_outlives_requirements,
391 irrefutable_let_patterns,
394 issue_5723_bootstrap,
395 issue_tracker_base_url,
397 item_context: "ItemContext",
416 link_llvm_intrinsics,
426 macro_at_most_once_rep,
429 macro_lifetime_matcher,
430 macro_literal_matcher,
441 match_beginning_vert,
442 match_default_bindings,
461 movbe_target_feature,
498 object_safe_for_dispatch,
500 omit_gdb_pretty_printer_section,
507 optin_builtin_traits,
517 overlapping_marker_traits,
522 panic_implementation,
542 poll_with_tls_context,
543 powerpc_target_feature,
544 precise_pointer_size_matching,
549 proc_dash_macro: "proc-macro",
551 proc_macro_attribute,
557 proc_macro_internals,
559 proc_macro_non_items,
560 proc_macro_path_invoc,
565 quad_precision_float,
580 reexport_test_harness_main,
592 re_rebalance_coherence,
610 rustc_allocator_nounwind,
611 rustc_allow_const_fn_ptr,
612 rustc_args_required_const,
616 rustc_const_unstable,
618 rustc_conversion_suggestion,
621 rustc_diagnostic_item,
622 rustc_diagnostic_macros,
625 rustc_dump_env_program_clauses,
626 rustc_dump_program_clauses,
627 rustc_dump_user_substs,
629 rustc_expected_cgu_reuse,
630 rustc_if_this_changed,
631 rustc_inherit_overflow_checks,
633 rustc_layout_scalar_valid_range_end,
634 rustc_layout_scalar_valid_range_start,
635 rustc_macro_transparency,
637 rustc_nonnull_optimization_guaranteed,
638 rustc_object_lifetime_default,
639 rustc_on_unimplemented,
642 rustc_partition_codegened,
643 rustc_partition_reused,
645 rustc_peek_definite_init,
646 rustc_peek_maybe_init,
647 rustc_peek_maybe_uninit,
648 rustc_peek_indirectly_mutable,
650 rustc_proc_macro_decls,
654 rustc_std_internal_symbol,
657 rustc_reservation_impl,
659 rustc_then_this_would_need,
663 rust_eh_unwind_resume,
665 rvalue_static_promotion,
689 sse4a_target_feature,
702 stmt_expr_attributes,
704 struct_field_attributes,
714 target_has_atomic_load_store,
719 termination_trait_test,
722 test_accepted_feature,
724 test_removed_feature,
747 type_alias_impl_trait,
752 type_alias_enum_variants,
769 underscore_const_names,
771 underscore_lifetimes,
773 universal_impl_trait,
776 unrestricted_attribute_tokens,
779 unsized_tuple_coercion,
795 visible_private_types,
810 #[derive(Copy, Clone, Eq, HashStable_Generic)]
818 /// Constructs a new identifier from a symbol and a span.
819 pub const fn new(name: Symbol, span: Span) -> Ident {
823 /// Constructs a new identifier with a dummy span.
825 pub const fn with_dummy_span(name: Symbol) -> Ident {
826 Ident::new(name, DUMMY_SP)
830 pub fn invalid() -> Ident {
831 Ident::with_dummy_span(kw::Invalid)
834 /// Maps a string to an identifier with a dummy span.
835 pub fn from_str(string: &str) -> Ident {
836 Ident::with_dummy_span(Symbol::intern(string))
839 /// Maps a string and a span to an identifier.
840 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
841 Ident::new(Symbol::intern(string), span)
844 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
845 pub fn with_span_pos(self, span: Span) -> Ident {
846 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
849 pub fn without_first_quote(self) -> Ident {
850 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
853 /// "Normalize" ident for use in comparisons using "item hygiene".
854 /// Identifiers with same string value become same if they came from the same "modern" macro
855 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
856 /// different "modern" macros.
857 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
858 pub fn modern(self) -> Ident {
859 Ident::new(self.name, self.span.modern())
862 /// "Normalize" ident for use in comparisons using "local variable hygiene".
863 /// Identifiers with same string value become same if they came from the same non-transparent
864 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
865 /// non-transparent macros.
866 /// Technically, this operation strips all transparent marks from ident's syntactic context.
867 pub fn modern_and_legacy(self) -> Ident {
868 Ident::new(self.name, self.span.modern_and_legacy())
871 /// Convert the name to a `SymbolStr`. This is a slowish operation because
872 /// it requires locking the symbol interner.
873 pub fn as_str(self) -> SymbolStr {
878 impl PartialEq for Ident {
879 fn eq(&self, rhs: &Self) -> bool {
880 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
884 impl Hash for Ident {
885 fn hash<H: Hasher>(&self, state: &mut H) {
886 self.name.hash(state);
887 self.span.ctxt().hash(state);
891 impl fmt::Debug for Ident {
892 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
893 if self.is_raw_guess() {
896 write!(f, "{}{:?}", self.name, self.span.ctxt())
900 impl fmt::Display for Ident {
901 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
902 if self.is_raw_guess() {
905 fmt::Display::fmt(&self.name, f)
909 impl UseSpecializedEncodable for Ident {
910 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
911 s.emit_struct("Ident", 2, |s| {
912 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
913 s.emit_struct_field("span", 1, |s| self.span.encode(s))
918 impl UseSpecializedDecodable for Ident {
919 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
920 d.read_struct("Ident", 2, |d| {
922 name: d.read_struct_field("name", 0, Decodable::decode)?,
923 span: d.read_struct_field("span", 1, Decodable::decode)?,
929 /// An interned string.
931 /// Internally, a `Symbol` is implemented as an index, and all operations
932 /// (including hashing, equality, and ordering) operate on that index. The use
933 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
934 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
936 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
937 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
938 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
939 pub struct Symbol(SymbolIndex);
941 rustc_index::newtype_index! {
942 pub struct SymbolIndex { .. }
946 const fn new(n: u32) -> Self {
947 Symbol(SymbolIndex::from_u32_const(n))
950 /// Maps a string to its interned representation.
951 pub fn intern(string: &str) -> Self {
952 with_interner(|interner| interner.intern(string))
955 /// Access the symbol's chars. This is a slowish operation because it
956 /// requires locking the symbol interner.
957 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
958 with_interner(|interner| f(interner.get(self)))
961 /// Convert to a `SymbolStr`. This is a slowish operation because it
962 /// requires locking the symbol interner.
963 pub fn as_str(self) -> SymbolStr {
964 with_interner(|interner| unsafe {
965 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
969 pub fn as_u32(self) -> u32 {
974 impl fmt::Debug for Symbol {
975 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
976 self.with(|str| fmt::Debug::fmt(&str, f))
980 impl fmt::Display for Symbol {
981 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
982 self.with(|str| fmt::Display::fmt(&str, f))
986 impl Encodable for Symbol {
987 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
988 self.with(|string| s.emit_str(string))
992 impl Decodable for Symbol {
993 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
994 Ok(Symbol::intern(&d.read_str()?))
998 impl<CTX> HashStable<CTX> for Symbol {
1000 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1001 self.as_str().hash_stable(hcx, hasher);
1005 impl<CTX> ToStableHashKey<CTX> for Symbol {
1006 type KeyType = SymbolStr;
1009 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1014 // The `&'static str`s in this type actually point into the arena.
1016 pub struct Interner {
1017 arena: DroplessArena,
1018 names: FxHashMap<&'static str, Symbol>,
1019 strings: Vec<&'static str>,
1023 fn prefill(init: &[&'static str]) -> Self {
1025 strings: init.into(),
1026 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1027 ..Default::default()
1031 pub fn intern(&mut self, string: &str) -> Symbol {
1032 if let Some(&name) = self.names.get(string) {
1036 let name = Symbol::new(self.strings.len() as u32);
1038 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1041 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1042 // It is safe to extend the arena allocation to `'static` because we only access
1043 // these while the arena is still alive.
1044 let string: &'static str = unsafe { &*(string as *const str) };
1045 self.strings.push(string);
1046 self.names.insert(string, name);
1050 // Get the symbol as a string. `Symbol::as_str()` should be used in
1051 // preference to this function.
1052 pub fn get(&self, symbol: Symbol) -> &str {
1053 self.strings[symbol.0.as_usize()]
1057 // This module has a very short name because it's used a lot.
1063 // This module has a very short name because it's used a lot.
1064 #[allow(rustc::default_hash_types)]
1067 use std::convert::TryInto;
1071 // Get the symbol for an integer. The first few non-negative integers each
1072 // have a static symbol and therefore are fast.
1073 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1074 if let Result::Ok(idx) = n.try_into() {
1075 if let Option::Some(&sym) = digits_array.get(idx) {
1079 Symbol::intern(&n.to_string())
1084 fn is_used_keyword_2018(self) -> bool {
1085 self >= kw::Async && self <= kw::Dyn
1088 fn is_unused_keyword_2018(self) -> bool {
1092 /// Used for sanity checking rustdoc keyword sections.
1093 pub fn is_doc_keyword(self) -> bool {
1097 /// A keyword or reserved identifier that can be used as a path segment.
1098 pub fn is_path_segment_keyword(self) -> bool {
1100 || self == kw::SelfLower
1101 || self == kw::SelfUpper
1102 || self == kw::Crate
1103 || self == kw::PathRoot
1104 || self == kw::DollarCrate
1107 /// Returns `true` if the symbol is `true` or `false`.
1108 pub fn is_bool_lit(self) -> bool {
1109 self == kw::True || self == kw::False
1112 /// This symbol can be a raw identifier.
1113 pub fn can_be_raw(self) -> bool {
1114 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1119 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1120 // unnamed method parameters, crate root module, error recovery etc.
1121 pub fn is_special(self) -> bool {
1122 self.name <= kw::Underscore
1125 /// Returns `true` if the token is a keyword used in the language.
1126 pub fn is_used_keyword(self) -> bool {
1127 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1128 self.name >= kw::As && self.name <= kw::While
1129 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1132 /// Returns `true` if the token is a keyword reserved for possible future use.
1133 pub fn is_unused_keyword(self) -> bool {
1134 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1135 self.name >= kw::Abstract && self.name <= kw::Yield
1136 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1139 /// Returns `true` if the token is either a special identifier or a keyword.
1140 pub fn is_reserved(self) -> bool {
1141 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1144 /// A keyword or reserved identifier that can be used as a path segment.
1145 pub fn is_path_segment_keyword(self) -> bool {
1146 self.name.is_path_segment_keyword()
1149 /// We see this identifier in a normal identifier position, like variable name or a type.
1150 /// How was it written originally? Did it use the raw form? Let's try to guess.
1151 pub fn is_raw_guess(self) -> bool {
1152 self.name.can_be_raw() && self.is_reserved()
1157 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1158 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1161 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1162 /// be accessed. It deliberately has limited functionality and should only be
1163 /// used for temporary values.
1165 /// Because the interner outlives any thread which uses this type, we can
1166 /// safely treat `string` which points to interner data, as an immortal string,
1167 /// as long as this type never crosses between threads.
1169 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1170 // by creating a new thread right after constructing the interner.
1171 #[derive(Clone, Eq, PartialOrd, Ord)]
1172 pub struct SymbolStr {
1173 string: &'static str,
1176 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1178 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1179 fn eq(&self, other: &T) -> bool {
1180 self.string == other.deref()
1184 impl !Send for SymbolStr {}
1185 impl !Sync for SymbolStr {}
1187 /// This impl means that if `ss` is a `SymbolStr`:
1188 /// - `*ss` is a `str`;
1189 /// - `&*ss` is a `&str`;
1190 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1191 /// function expecting a `&str`.
1192 impl std::ops::Deref for SymbolStr {
1195 fn deref(&self) -> &str {
1200 impl fmt::Debug for SymbolStr {
1201 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1202 fmt::Debug::fmt(self.string, f)
1206 impl fmt::Display for SymbolStr {
1207 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1208 fmt::Display::fmt(self.string, f)
1212 impl<CTX> HashStable<CTX> for SymbolStr {
1214 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1215 self.string.hash_stable(hcx, hasher)
1219 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1220 type KeyType = SymbolStr;
1223 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {