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,
157 augmented_assignments,
158 automatically_derived,
159 avx512_target_feature,
164 bind_by_move_pattern_guards,
167 borrowck_graphviz_postflow,
168 borrowck_graphviz_preflow,
171 braced_empty_structs,
179 cfg_target_has_atomic,
180 cfg_target_thread_local,
188 closure_to_fn_coercion,
190 cmpxchg16b_target_feature,
197 conservative_impl_trait,
199 const_compare_raw_pointers,
206 const_in_array_repeat_expressions,
210 const_raw_ptr_to_usize_cast,
224 crate_visibility_modifier,
227 custom_inner_attributes,
228 custom_test_frameworks,
236 default_lib_allocator,
237 default_type_parameter_fallback,
253 document_private_items,
254 dotdoteq_in_patterns,
255 dotdot_in_tuple_patterns,
256 double_braced_crate: "{{crate}}",
257 double_braced_impl: "{{impl}}",
258 double_braced_misc: "{{misc}}",
259 double_braced_closure: "{{closure}}",
260 double_braced_constructor: "{{constructor}}",
261 double_braced_constant: "{{constant}}",
262 double_braced_opaque: "{{opaque}}",
264 dropck_parametricity,
279 exclusive_range_pattern,
280 exhaustive_integer_patterns,
286 extern_absolute_paths,
288 extern_crate_item_prelude,
299 field_init_shorthand,
322 generic_associated_types,
331 hexagon_target_feature,
333 homogeneous_aggregate,
347 if_while_or_patterns,
349 impl_header_lifetime_elision,
351 impl_trait_in_bindings,
359 inclusive_range_syntax,
360 infer_outlives_requirements,
361 infer_static_outlives_requirements,
368 irrefutable_let_patterns,
371 issue_5723_bootstrap,
372 issue_tracker_base_url,
374 item_context: "ItemContext",
393 link_llvm_intrinsics,
403 macro_at_most_once_rep,
406 macro_lifetime_matcher,
407 macro_literal_matcher,
418 match_beginning_vert,
419 match_default_bindings,
434 movbe_target_feature,
467 object_safe_for_dispatch,
469 omit_gdb_pretty_printer_section,
476 optin_builtin_traits,
486 overlapping_marker_traits,
491 panic_implementation,
511 poll_with_tls_context,
512 powerpc_target_feature,
513 precise_pointer_size_matching,
517 proc_dash_macro: "proc-macro",
519 proc_macro_attribute,
525 proc_macro_internals,
527 proc_macro_non_items,
528 proc_macro_path_invoc,
532 quad_precision_float,
546 reexport_test_harness_main,
556 re_rebalance_coherence,
573 rustc_allocator_nounwind,
574 rustc_allow_const_fn_ptr,
575 rustc_args_required_const,
579 rustc_const_unstable,
580 rustc_conversion_suggestion,
583 rustc_diagnostic_item,
584 rustc_diagnostic_macros,
587 rustc_dump_env_program_clauses,
588 rustc_dump_program_clauses,
589 rustc_dump_user_substs,
591 rustc_expected_cgu_reuse,
592 rustc_if_this_changed,
593 rustc_inherit_overflow_checks,
595 rustc_layout_scalar_valid_range_end,
596 rustc_layout_scalar_valid_range_start,
597 rustc_macro_transparency,
599 rustc_nonnull_optimization_guaranteed,
600 rustc_object_lifetime_default,
601 rustc_on_unimplemented,
604 rustc_partition_codegened,
605 rustc_partition_reused,
607 rustc_peek_definite_init,
608 rustc_peek_maybe_init,
609 rustc_peek_maybe_uninit,
610 rustc_peek_indirectly_mutable,
612 rustc_proc_macro_decls,
616 rustc_std_internal_symbol,
619 rustc_reservation_impl,
621 rustc_then_this_would_need,
626 rust_eh_unwind_resume,
628 rvalue_static_promotion,
645 sse4a_target_feature,
658 stmt_expr_attributes,
660 struct_field_attributes,
668 target_has_atomic_load_store,
673 termination_trait_test,
676 test_accepted_feature,
678 test_removed_feature,
699 type_alias_impl_trait,
702 type_alias_enum_variants,
712 underscore_const_names,
714 underscore_lifetimes,
718 universal_impl_trait,
721 unrestricted_attribute_tokens,
724 unsized_tuple_coercion,
739 visible_private_types,
752 #[derive(Copy, Clone, Eq)]
760 /// Constructs a new identifier from a symbol and a span.
761 pub const fn new(name: Symbol, span: Span) -> Ident {
765 /// Constructs a new identifier with a dummy span.
767 pub const fn with_dummy_span(name: Symbol) -> Ident {
768 Ident::new(name, DUMMY_SP)
772 pub fn invalid() -> Ident {
773 Ident::with_dummy_span(kw::Invalid)
776 /// Maps a string to an identifier with a dummy span.
777 pub fn from_str(string: &str) -> Ident {
778 Ident::with_dummy_span(Symbol::intern(string))
781 /// Maps a string and a span to an identifier.
782 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
783 Ident::new(Symbol::intern(string), span)
786 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
787 pub fn with_span_pos(self, span: Span) -> Ident {
788 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
791 pub fn without_first_quote(self) -> Ident {
792 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
795 /// "Normalize" ident for use in comparisons using "item hygiene".
796 /// Identifiers with same string value become same if they came from the same "modern" macro
797 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
798 /// different "modern" macros.
799 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
800 pub fn modern(self) -> Ident {
801 Ident::new(self.name, self.span.modern())
804 /// "Normalize" ident for use in comparisons using "local variable hygiene".
805 /// Identifiers with same string value become same if they came from the same non-transparent
806 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
807 /// non-transparent macros.
808 /// Technically, this operation strips all transparent marks from ident's syntactic context.
809 pub fn modern_and_legacy(self) -> Ident {
810 Ident::new(self.name, self.span.modern_and_legacy())
813 /// Convert the name to a `SymbolStr`. This is a slowish operation because
814 /// it requires locking the symbol interner.
815 pub fn as_str(self) -> SymbolStr {
820 impl PartialEq for Ident {
821 fn eq(&self, rhs: &Self) -> bool {
822 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
826 impl Hash for Ident {
827 fn hash<H: Hasher>(&self, state: &mut H) {
828 self.name.hash(state);
829 self.span.ctxt().hash(state);
833 impl fmt::Debug for Ident {
834 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
835 write!(f, "{}{:?}", self.name, self.span.ctxt())
839 impl fmt::Display for Ident {
840 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
841 fmt::Display::fmt(&self.name, f)
845 impl UseSpecializedEncodable for Ident {
846 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
847 s.emit_struct("Ident", 2, |s| {
848 s.emit_struct_field("name", 0, |s| {
851 s.emit_struct_field("span", 1, |s| {
858 impl UseSpecializedDecodable for Ident {
859 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
860 d.read_struct("Ident", 2, |d| {
862 name: d.read_struct_field("name", 0, Decodable::decode)?,
863 span: d.read_struct_field("span", 1, Decodable::decode)?,
869 /// An interned string.
871 /// Internally, a `Symbol` is implemented as an index, and all operations
872 /// (including hashing, equality, and ordering) operate on that index. The use
873 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
874 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
876 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
877 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
878 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
879 pub struct Symbol(SymbolIndex);
881 rustc_index::newtype_index! {
882 pub struct SymbolIndex { .. }
886 const fn new(n: u32) -> Self {
887 Symbol(SymbolIndex::from_u32_const(n))
890 /// Maps a string to its interned representation.
891 pub fn intern(string: &str) -> Self {
892 with_interner(|interner| interner.intern(string))
895 /// Access the symbol's chars. This is a slowish operation because it
896 /// requires locking the symbol interner.
897 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
898 with_interner(|interner| {
899 f(interner.get(self))
903 /// Convert to a `SymbolStr`. This is a slowish operation because it
904 /// requires locking the symbol interner.
905 pub fn as_str(self) -> SymbolStr {
906 with_interner(|interner| unsafe {
908 string: std::mem::transmute::<&str, &str>(interner.get(self))
913 pub fn as_u32(self) -> u32 {
918 impl fmt::Debug for Symbol {
919 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
920 self.with(|str| fmt::Debug::fmt(&str, f))
924 impl fmt::Display for Symbol {
925 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
926 self.with(|str| fmt::Display::fmt(&str, f))
930 impl Encodable for Symbol {
931 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
932 self.with(|string| s.emit_str(string))
936 impl Decodable for Symbol {
937 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
938 Ok(Symbol::intern(&d.read_str()?))
942 // The `&'static str`s in this type actually point into the arena.
944 pub struct Interner {
945 arena: DroplessArena,
946 names: FxHashMap<&'static str, Symbol>,
947 strings: Vec<&'static str>,
951 fn prefill(init: &[&'static str]) -> Self {
953 strings: init.into(),
954 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
959 pub fn intern(&mut self, string: &str) -> Symbol {
960 if let Some(&name) = self.names.get(string) {
964 let name = Symbol::new(self.strings.len() as u32);
966 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
968 let string: &str = unsafe {
969 str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes()))
971 // It is safe to extend the arena allocation to `'static` because we only access
972 // these while the arena is still alive.
973 let string: &'static str = unsafe {
974 &*(string as *const str)
976 self.strings.push(string);
977 self.names.insert(string, name);
981 // Get the symbol as a string. `Symbol::as_str()` should be used in
982 // preference to this function.
983 pub fn get(&self, symbol: Symbol) -> &str {
984 self.strings[symbol.0.as_usize()]
988 // This module has a very short name because it's used a lot.
994 // This module has a very short name because it's used a lot.
996 use std::convert::TryInto;
1001 // Get the symbol for an integer. The first few non-negative integers each
1002 // have a static symbol and therefore are fast.
1003 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1004 if let Result::Ok(idx) = n.try_into() {
1005 if let Option::Some(&sym) = digits_array.get(idx) {
1009 Symbol::intern(&n.to_string())
1014 fn is_used_keyword_2018(self) -> bool {
1015 self >= kw::Async && self <= kw::Dyn
1018 fn is_unused_keyword_2018(self) -> bool {
1022 /// Used for sanity checking rustdoc keyword sections.
1023 pub fn is_doc_keyword(self) -> bool {
1027 /// A keyword or reserved identifier that can be used as a path segment.
1028 pub fn is_path_segment_keyword(self) -> bool {
1029 self == kw::Super ||
1030 self == kw::SelfLower ||
1031 self == kw::SelfUpper ||
1032 self == kw::Crate ||
1033 self == kw::PathRoot ||
1034 self == kw::DollarCrate
1037 /// Returns `true` if the symbol is `true` or `false`.
1038 pub fn is_bool_lit(self) -> bool {
1039 self == kw::True || self == kw::False
1042 /// This symbol can be a raw identifier.
1043 pub fn can_be_raw(self) -> bool {
1044 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1049 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1050 // unnamed method parameters, crate root module, error recovery etc.
1051 pub fn is_special(self) -> bool {
1052 self.name <= kw::Underscore
1055 /// Returns `true` if the token is a keyword used in the language.
1056 pub fn is_used_keyword(self) -> bool {
1057 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1058 self.name >= kw::As && self.name <= kw::While ||
1059 self.name.is_used_keyword_2018() && self.span.rust_2018()
1062 /// Returns `true` if the token is a keyword reserved for possible future use.
1063 pub fn is_unused_keyword(self) -> bool {
1064 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1065 self.name >= kw::Abstract && self.name <= kw::Yield ||
1066 self.name.is_unused_keyword_2018() && self.span.rust_2018()
1069 /// Returns `true` if the token is either a special identifier or a keyword.
1070 pub fn is_reserved(self) -> bool {
1071 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1074 /// A keyword or reserved identifier that can be used as a path segment.
1075 pub fn is_path_segment_keyword(self) -> bool {
1076 self.name.is_path_segment_keyword()
1079 /// We see this identifier in a normal identifier position, like variable name or a type.
1080 /// How was it written originally? Did it use the raw form? Let's try to guess.
1081 pub fn is_raw_guess(self) -> bool {
1082 self.name.can_be_raw() && self.is_reserved()
1087 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1088 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1091 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1092 /// be accessed. It deliberately has limited functionality and should only be
1093 /// used for temporary values.
1095 /// Because the interner outlives any thread which uses this type, we can
1096 /// safely treat `string` which points to interner data, as an immortal string,
1097 /// as long as this type never crosses between threads.
1099 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1100 // by creating a new thread right after constructing the interner.
1101 #[derive(Clone, Eq, PartialOrd, Ord)]
1102 pub struct SymbolStr {
1103 string: &'static str,
1106 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1108 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1109 fn eq(&self, other: &T) -> bool {
1110 self.string == other.deref()
1114 impl !Send for SymbolStr {}
1115 impl !Sync for SymbolStr {}
1117 /// This impl means that if `ss` is a `SymbolStr`:
1118 /// - `*ss` is a `str`;
1119 /// - `&*ss` is a `&str`;
1120 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1121 /// function expecting a `&str`.
1122 impl std::ops::Deref for SymbolStr {
1125 fn deref(&self) -> &str { self.string }
1128 impl fmt::Debug for SymbolStr {
1129 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1130 fmt::Debug::fmt(self.string, f)
1134 impl fmt::Display for SymbolStr {
1135 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1136 fmt::Display::fmt(self.string, f)