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_index::vec::Idx;
8 use rustc_macros::symbols;
9 use rustc_serialize::{Decodable, Decoder, Encodable, Encoder};
10 use rustc_serialize::{UseSpecializedDecodable, UseSpecializedEncodable};
12 use std::cmp::{PartialEq, PartialOrd, Ord};
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.
103 // Symbols that can be referred to with syntax_pos::sym::*. The symbol is
104 // the stringified identifier unless otherwise specified (e.g.
105 // `proc_dash_macro` represents "proc-macro").
107 // As well as the symbols listed, there are symbols for the the strings
108 // "0", "1", ..., "9", which are accessible via `sym::integer`.
110 aarch64_target_feature,
114 abi_msp430_interrupt,
122 advanced_slice_patterns,
134 allow_internal_unsafe,
135 allow_internal_unstable,
136 allow_internal_unstable_backcompat_hack,
140 arbitrary_enum_discriminant,
141 arbitrary_self_types,
148 associated_type_bounds,
149 associated_type_defaults,
156 augmented_assignments,
157 automatically_derived,
158 avx512_target_feature,
163 bind_by_move_pattern_guards,
166 borrowck_graphviz_postflow,
167 borrowck_graphviz_preflow,
170 braced_empty_structs,
178 cfg_target_has_atomic,
179 cfg_target_thread_local,
187 closure_to_fn_coercion,
189 cmpxchg16b_target_feature,
196 conservative_impl_trait,
198 const_compare_raw_pointers,
205 const_in_array_repeat_expressions,
209 const_raw_ptr_to_usize_cast,
223 crate_visibility_modifier,
226 custom_inner_attributes,
227 custom_test_frameworks,
235 default_lib_allocator,
236 default_type_parameter_fallback,
252 document_private_items,
253 dotdoteq_in_patterns,
254 dotdot_in_tuple_patterns,
255 double_braced_crate: "{{crate}}",
256 double_braced_impl: "{{impl}}",
257 double_braced_misc: "{{misc}}",
258 double_braced_closure: "{{closure}}",
259 double_braced_constructor: "{{constructor}}",
260 double_braced_constant: "{{constant}}",
261 double_braced_opaque: "{{opaque}}",
263 dropck_parametricity,
278 exclusive_range_pattern,
279 exhaustive_integer_patterns,
285 extern_absolute_paths,
287 extern_crate_item_prelude,
298 field_init_shorthand,
321 generic_associated_types,
330 hexagon_target_feature,
332 homogeneous_aggregate,
346 if_while_or_patterns,
348 impl_header_lifetime_elision,
350 impl_trait_in_bindings,
358 inclusive_range_syntax,
359 infer_outlives_requirements,
360 infer_static_outlives_requirements,
367 irrefutable_let_patterns,
370 issue_5723_bootstrap,
371 issue_tracker_base_url,
373 item_context: "ItemContext",
392 link_llvm_intrinsics,
402 macro_at_most_once_rep,
405 macro_lifetime_matcher,
406 macro_literal_matcher,
417 match_beginning_vert,
418 match_default_bindings,
431 movbe_target_feature,
464 object_safe_for_dispatch,
466 omit_gdb_pretty_printer_section,
473 optin_builtin_traits,
483 overlapping_marker_traits,
488 panic_implementation,
508 poll_with_tls_context,
509 powerpc_target_feature,
510 precise_pointer_size_matching,
514 proc_dash_macro: "proc-macro",
516 proc_macro_attribute,
522 proc_macro_internals,
524 proc_macro_non_items,
525 proc_macro_path_invoc,
529 quad_precision_float,
543 reexport_test_harness_main,
553 re_rebalance_coherence,
570 rustc_allocator_nounwind,
571 rustc_allow_const_fn_ptr,
572 rustc_args_required_const,
576 rustc_const_unstable,
577 rustc_conversion_suggestion,
580 rustc_diagnostic_item,
581 rustc_diagnostic_macros,
584 rustc_dump_env_program_clauses,
585 rustc_dump_program_clauses,
586 rustc_dump_user_substs,
588 rustc_expected_cgu_reuse,
589 rustc_if_this_changed,
590 rustc_inherit_overflow_checks,
592 rustc_layout_scalar_valid_range_end,
593 rustc_layout_scalar_valid_range_start,
594 rustc_macro_transparency,
596 rustc_nonnull_optimization_guaranteed,
597 rustc_object_lifetime_default,
598 rustc_on_unimplemented,
601 rustc_partition_codegened,
602 rustc_partition_reused,
604 rustc_peek_definite_init,
605 rustc_peek_maybe_init,
606 rustc_peek_maybe_uninit,
607 rustc_peek_indirectly_mutable,
609 rustc_proc_macro_decls,
613 rustc_std_internal_symbol,
616 rustc_reservation_impl,
618 rustc_then_this_would_need,
623 rust_eh_unwind_resume,
625 rvalue_static_promotion,
642 sse4a_target_feature,
655 stmt_expr_attributes,
657 struct_field_attributes,
665 target_has_atomic_load_store,
670 termination_trait_test,
673 test_accepted_feature,
675 test_removed_feature,
696 type_alias_impl_trait,
699 type_alias_enum_variants,
709 underscore_const_names,
711 underscore_lifetimes,
714 universal_impl_trait,
717 unrestricted_attribute_tokens,
720 unsized_tuple_coercion,
735 visible_private_types,
738 warn_directory_ownership,
749 #[derive(Copy, Clone, Eq)]
757 /// Constructs a new identifier from a symbol and a span.
758 pub const fn new(name: Symbol, span: Span) -> Ident {
762 /// Constructs a new identifier with a dummy span.
764 pub const fn with_dummy_span(name: Symbol) -> Ident {
765 Ident::new(name, DUMMY_SP)
769 pub fn invalid() -> Ident {
770 Ident::with_dummy_span(kw::Invalid)
773 /// Maps a string to an identifier with a dummy span.
774 pub fn from_str(string: &str) -> Ident {
775 Ident::with_dummy_span(Symbol::intern(string))
778 /// Maps a string and a span to an identifier.
779 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
780 Ident::new(Symbol::intern(string), span)
783 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
784 pub fn with_span_pos(self, span: Span) -> Ident {
785 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
788 pub fn without_first_quote(self) -> Ident {
789 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
792 /// "Normalize" ident for use in comparisons using "item hygiene".
793 /// Identifiers with same string value become same if they came from the same "modern" macro
794 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
795 /// different "modern" macros.
796 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
797 pub fn modern(self) -> Ident {
798 Ident::new(self.name, self.span.modern())
801 /// "Normalize" ident for use in comparisons using "local variable hygiene".
802 /// Identifiers with same string value become same if they came from the same non-transparent
803 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
804 /// non-transparent macros.
805 /// Technically, this operation strips all transparent marks from ident's syntactic context.
806 pub fn modern_and_legacy(self) -> Ident {
807 Ident::new(self.name, self.span.modern_and_legacy())
810 /// Convert the name to a `SymbolStr`. This is a slowish operation because
811 /// it requires locking the symbol interner.
812 pub fn as_str(self) -> SymbolStr {
817 impl PartialEq for Ident {
818 fn eq(&self, rhs: &Self) -> bool {
819 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
823 impl Hash for Ident {
824 fn hash<H: Hasher>(&self, state: &mut H) {
825 self.name.hash(state);
826 self.span.ctxt().hash(state);
830 impl fmt::Debug for Ident {
831 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
832 write!(f, "{}{:?}", self.name, self.span.ctxt())
836 impl fmt::Display for Ident {
837 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
838 fmt::Display::fmt(&self.name, f)
842 impl UseSpecializedEncodable for Ident {
843 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
844 s.emit_struct("Ident", 2, |s| {
845 s.emit_struct_field("name", 0, |s| {
848 s.emit_struct_field("span", 1, |s| {
855 impl UseSpecializedDecodable for Ident {
856 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
857 d.read_struct("Ident", 2, |d| {
859 name: d.read_struct_field("name", 0, Decodable::decode)?,
860 span: d.read_struct_field("span", 1, Decodable::decode)?,
866 /// An interned string.
868 /// Internally, a `Symbol` is implemented as an index, and all operations
869 /// (including hashing, equality, and ordering) operate on that index. The use
870 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
871 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
873 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
874 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
875 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
876 pub struct Symbol(SymbolIndex);
878 rustc_index::newtype_index! {
879 pub struct SymbolIndex { .. }
883 const fn new(n: u32) -> Self {
884 Symbol(SymbolIndex::from_u32_const(n))
887 /// Maps a string to its interned representation.
888 pub fn intern(string: &str) -> Self {
889 with_interner(|interner| interner.intern(string))
892 /// Access the symbol's chars. This is a slowish operation because it
893 /// requires locking the symbol interner.
894 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
895 with_interner(|interner| {
896 f(interner.get(self))
900 /// Convert to a `SymbolStr`. This is a slowish operation because it
901 /// requires locking the symbol interner.
902 pub fn as_str(self) -> SymbolStr {
903 with_interner(|interner| unsafe {
905 string: std::mem::transmute::<&str, &str>(interner.get(self))
910 pub fn as_u32(self) -> u32 {
915 impl fmt::Debug for Symbol {
916 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
917 self.with(|str| fmt::Debug::fmt(&str, f))
921 impl fmt::Display for Symbol {
922 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
923 self.with(|str| fmt::Display::fmt(&str, f))
927 impl Encodable for Symbol {
928 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
929 self.with(|string| s.emit_str(string))
933 impl Decodable for Symbol {
934 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
935 Ok(Symbol::intern(&d.read_str()?))
939 // The `&'static str`s in this type actually point into the arena.
941 pub struct Interner {
942 arena: DroplessArena,
943 names: FxHashMap<&'static str, Symbol>,
944 strings: Vec<&'static str>,
948 fn prefill(init: &[&'static str]) -> Self {
950 strings: init.into(),
951 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
956 pub fn intern(&mut self, string: &str) -> Symbol {
957 if let Some(&name) = self.names.get(string) {
961 let name = Symbol::new(self.strings.len() as u32);
963 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
965 let string: &str = unsafe {
966 str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes()))
968 // It is safe to extend the arena allocation to `'static` because we only access
969 // these while the arena is still alive.
970 let string: &'static str = unsafe {
971 &*(string as *const str)
973 self.strings.push(string);
974 self.names.insert(string, name);
978 // Get the symbol as a string. `Symbol::as_str()` should be used in
979 // preference to this function.
980 pub fn get(&self, symbol: Symbol) -> &str {
981 self.strings[symbol.0.as_usize()]
985 // This module has a very short name because it's used a lot.
991 // This module has a very short name because it's used a lot.
993 use std::convert::TryInto;
998 // Get the symbol for an integer. The first few non-negative integers each
999 // have a static symbol and therefore are fast.
1000 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1001 if let Result::Ok(idx) = n.try_into() {
1002 if let Option::Some(&sym) = digits_array.get(idx) {
1006 Symbol::intern(&n.to_string())
1011 fn is_used_keyword_2018(self) -> bool {
1012 self >= kw::Async && self <= kw::Dyn
1015 fn is_unused_keyword_2018(self) -> bool {
1019 /// Used for sanity checking rustdoc keyword sections.
1020 pub fn is_doc_keyword(self) -> bool {
1024 /// A keyword or reserved identifier that can be used as a path segment.
1025 pub fn is_path_segment_keyword(self) -> bool {
1026 self == kw::Super ||
1027 self == kw::SelfLower ||
1028 self == kw::SelfUpper ||
1029 self == kw::Crate ||
1030 self == kw::PathRoot ||
1031 self == kw::DollarCrate
1034 /// Returns `true` if the symbol is `true` or `false`.
1035 pub fn is_bool_lit(self) -> bool {
1036 self == kw::True || self == kw::False
1039 /// This symbol can be a raw identifier.
1040 pub fn can_be_raw(self) -> bool {
1041 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1046 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1047 // unnamed method parameters, crate root module, error recovery etc.
1048 pub fn is_special(self) -> bool {
1049 self.name <= kw::Underscore
1052 /// Returns `true` if the token is a keyword used in the language.
1053 pub fn is_used_keyword(self) -> bool {
1054 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1055 self.name >= kw::As && self.name <= kw::While ||
1056 self.name.is_used_keyword_2018() && self.span.rust_2018()
1059 /// Returns `true` if the token is a keyword reserved for possible future use.
1060 pub fn is_unused_keyword(self) -> bool {
1061 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1062 self.name >= kw::Abstract && self.name <= kw::Yield ||
1063 self.name.is_unused_keyword_2018() && self.span.rust_2018()
1066 /// Returns `true` if the token is either a special identifier or a keyword.
1067 pub fn is_reserved(self) -> bool {
1068 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1071 /// A keyword or reserved identifier that can be used as a path segment.
1072 pub fn is_path_segment_keyword(self) -> bool {
1073 self.name.is_path_segment_keyword()
1076 /// We see this identifier in a normal identifier position, like variable name or a type.
1077 /// How was it written originally? Did it use the raw form? Let's try to guess.
1078 pub fn is_raw_guess(self) -> bool {
1079 self.name.can_be_raw() && self.is_reserved()
1084 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1085 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1088 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1089 /// be accessed. It deliberately has limited functionality and should only be
1090 /// used for temporary values.
1092 /// Because the interner outlives any thread which uses this type, we can
1093 /// safely treat `string` which points to interner data, as an immortal string,
1094 /// as long as this type never crosses between threads.
1096 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1097 // by creating a new thread right after constructing the interner.
1098 #[derive(Clone, Eq, PartialOrd, Ord)]
1099 pub struct SymbolStr {
1100 string: &'static str,
1103 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1105 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1106 fn eq(&self, other: &T) -> bool {
1107 self.string == other.deref()
1111 impl !Send for SymbolStr {}
1112 impl !Sync for SymbolStr {}
1114 /// This impl means that if `ss` is a `SymbolStr`:
1115 /// - `*ss` is a `str`;
1116 /// - `&*ss` is a `&str`;
1117 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1118 /// function expecting a `&str`.
1119 impl std::ops::Deref for SymbolStr {
1122 fn deref(&self) -> &str { self.string }
1125 impl fmt::Debug for SymbolStr {
1126 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1127 fmt::Debug::fmt(self.string, f)
1131 impl fmt::Display for SymbolStr {
1132 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1133 fmt::Display::fmt(self.string, f)