1 // Spans are encoded using 1-bit tag and 2 different encoding formats (one for each tag value).
2 // One format is used for keeping span data inline,
3 // another contains index into an out-of-line span interner.
4 // The encoding format for inline spans were obtained by optimizing over crates in rustc/libstd.
5 // See https://internals.rust-lang.org/t/rfc-compiler-refactoring-spans/1357/28
7 use crate::def_id::{DefIndex, LocalDefId};
8 use crate::hygiene::SyntaxContext;
10 use crate::{BytePos, SpanData};
12 use rustc_data_structures::fx::FxIndexSet;
14 /// A compressed span.
16 /// Whereas [`SpanData`] is 16 bytes, which is a bit too big to stick everywhere, `Span`
17 /// is a form that only takes up 8 bytes, with less space for the length, parent and
18 /// context. The vast majority (99.9%+) of `SpanData` instances will fit within
19 /// those 8 bytes; any `SpanData` whose fields don't fit into a `Span` are
20 /// stored in a separate interner table, and the `Span` will index into that
21 /// table. Interning is rare enough that the cost is low, but common enough
22 /// that the code is exercised regularly.
24 /// An earlier version of this code used only 4 bytes for `Span`, but that was
25 /// slower because only 80--90% of spans could be stored inline (even less in
26 /// very large crates) and so the interner was used a lot more.
28 /// Inline (compressed) format with no parent:
29 /// - `span.base_or_index == span_data.lo`
30 /// - `span.len_or_tag == len == span_data.hi - span_data.lo` (must be `<= MAX_LEN`)
31 /// - `span.ctxt_or_tag == span_data.ctxt` (must be `<= MAX_CTXT`)
33 /// Interned format with inline `SyntaxContext`:
34 /// - `span.base_or_index == index` (indexes into the interner table)
35 /// - `span.len_or_tag == LEN_TAG` (high bit set, all other bits are zero)
36 /// - `span.ctxt_or_tag == span_data.ctxt` (must be `<= MAX_CTXT`)
38 /// Inline (compressed) format with root context:
39 /// - `span.base_or_index == span_data.lo`
40 /// - `span.len_or_tag == len == span_data.hi - span_data.lo` (must be `<= MAX_LEN`)
41 /// - `span.len_or_tag` has top bit (`PARENT_MASK`) set
42 /// - `span.ctxt == span_data.parent` (must be `<= MAX_CTXT`)
45 /// - `span.base_or_index == index` (indexes into the interner table)
46 /// - `span.len_or_tag == LEN_TAG` (high bit set, all other bits are zero)
47 /// - `span.ctxt_or_tag == CTXT_TAG`
49 /// The inline form uses 0 for the tag value (rather than 1) so that we don't
50 /// need to mask out the tag bit when getting the length, and so that the
51 /// dummy span can be all zeroes.
53 /// Notes about the choice of field sizes:
54 /// - `base` is 32 bits in both `Span` and `SpanData`, which means that `base`
55 /// values never cause interning. The number of bits needed for `base`
56 /// depends on the crate size. 32 bits allows up to 4 GiB of code in a crate.
57 /// - `len` is 15 bits in `Span` (a u16, minus 1 bit for the tag) and 32 bits
58 /// in `SpanData`, which means that large `len` values will cause interning.
59 /// The number of bits needed for `len` does not depend on the crate size.
60 /// The most common numbers of bits for `len` are from 0 to 7, with a peak usually
61 /// at 3 or 4, and then it drops off quickly from 8 onwards. 15 bits is enough
62 /// for 99.99%+ of cases, but larger values (sometimes 20+ bits) might occur
63 /// dozens of times in a typical crate.
64 /// - `ctxt_or_tag` is 16 bits in `Span` and 32 bits in `SpanData`, which means that
65 /// large `ctxt` values will cause interning. The number of bits needed for
66 /// `ctxt` values depend partly on the crate size and partly on the form of
67 /// the code. No crates in `rustc-perf` need more than 15 bits for `ctxt_or_tag`,
68 /// but larger crates might need more than 16 bits.
70 /// In order to reliably use parented spans in incremental compilation,
71 /// the dependency to the parent definition's span. This is performed
72 /// using the callback `SPAN_TRACK` to access the query engine.
74 #[derive(Clone, Copy, Eq, PartialEq, Hash)]
75 #[rustc_pass_by_value]
82 const LEN_TAG: u16 = 0b1111_1111_1111_1111;
83 const PARENT_MASK: u16 = 0b1000_0000_0000_0000;
84 const MAX_LEN: u32 = 0b0111_1111_1111_1111;
85 const CTXT_TAG: u32 = 0b1111_1111_1111_1111;
86 const MAX_CTXT: u32 = CTXT_TAG - 1;
88 /// Dummy span, both position and length are zero, syntax context is zero as well.
89 pub const DUMMY_SP: Span = Span { base_or_index: 0, len_or_tag: 0, ctxt_or_tag: 0 };
97 parent: Option<LocalDefId>,
100 std::mem::swap(&mut lo, &mut hi);
103 let (base, len, ctxt2) = (lo.0, hi.0 - lo.0, ctxt.as_u32());
105 if len <= MAX_LEN && ctxt2 <= MAX_CTXT {
106 let len_or_tag = len as u16;
107 debug_assert_eq!(len_or_tag & PARENT_MASK, 0);
109 if let Some(parent) = parent {
110 // Inline format with parent.
111 let len_or_tag = len_or_tag | PARENT_MASK;
112 let parent2 = parent.local_def_index.as_u32();
113 if ctxt2 == SyntaxContext::root().as_u32() && parent2 <= MAX_CTXT {
114 return Span { base_or_index: base, len_or_tag, ctxt_or_tag: parent2 as u16 };
117 // Inline format with ctxt.
120 len_or_tag: len as u16,
121 ctxt_or_tag: ctxt2 as u16,
128 with_span_interner(|interner| interner.intern(&SpanData { lo, hi, ctxt, parent }));
129 let ctxt_or_tag = if ctxt2 <= MAX_CTXT { ctxt2 } else { CTXT_TAG } as u16;
130 Span { base_or_index: index, len_or_tag: LEN_TAG, ctxt_or_tag }
134 pub fn data(self) -> SpanData {
135 let data = self.data_untracked();
136 if let Some(parent) = data.parent {
137 (*SPAN_TRACK)(parent);
142 /// Internal function to translate between an encoded span and the expanded representation.
143 /// This function must not be used outside the incremental engine.
145 pub fn data_untracked(self) -> SpanData {
146 if self.len_or_tag != LEN_TAG {
148 if self.len_or_tag & PARENT_MASK == 0 {
149 debug_assert!(self.len_or_tag as u32 <= MAX_LEN);
151 lo: BytePos(self.base_or_index),
152 hi: BytePos(self.base_or_index + self.len_or_tag as u32),
153 ctxt: SyntaxContext::from_u32(self.ctxt_or_tag as u32),
157 let len = self.len_or_tag & !PARENT_MASK;
158 debug_assert!(len as u32 <= MAX_LEN);
160 LocalDefId { local_def_index: DefIndex::from_u32(self.ctxt_or_tag as u32) };
162 lo: BytePos(self.base_or_index),
163 hi: BytePos(self.base_or_index + len as u32),
164 ctxt: SyntaxContext::root(),
165 parent: Some(parent),
170 let index = self.base_or_index;
171 with_span_interner(|interner| interner.spans[index as usize])
175 /// This function is used as a fast path when decoding the full `SpanData` is not necessary.
177 pub fn ctxt(self) -> SyntaxContext {
178 let ctxt_or_tag = self.ctxt_or_tag as u32;
179 if ctxt_or_tag <= MAX_CTXT {
180 if self.len_or_tag == LEN_TAG || self.len_or_tag & PARENT_MASK == 0 {
181 // Inline format or interned format with inline ctxt.
182 SyntaxContext::from_u32(ctxt_or_tag)
184 // Inline format or interned format with inline parent.
185 // We know that the SyntaxContext is root.
186 SyntaxContext::root()
190 let index = self.base_or_index;
191 with_span_interner(|interner| interner.spans[index as usize].ctxt)
197 pub struct SpanInterner {
198 spans: FxIndexSet<SpanData>,
202 fn intern(&mut self, span_data: &SpanData) -> u32 {
203 let (index, _) = self.spans.insert_full(*span_data);
208 // If an interner exists, return it. Otherwise, prepare a fresh one.
210 fn with_span_interner<T, F: FnOnce(&mut SpanInterner) -> T>(f: F) -> T {
211 crate::with_session_globals(|session_globals| f(&mut session_globals.span_interner.lock()))