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
8 use crate::{BytePos, SpanData};
9 use crate::hygiene::SyntaxContext;
11 use rustc_data_structures::fx::FxHashMap;
13 /// A compressed span.
15 /// `SpanData` is 12 bytes, which is a bit too big to stick everywhere. `Span`
16 /// is a form that only takes up 8 bytes, with less space for the length and
17 /// context. The vast majority (99.9%+) of `SpanData` instances will fit within
18 /// those 8 bytes; any `SpanData` whose fields don't fit into a `Span` are
19 /// stored in a separate interner table, and the `Span` will index into that
20 /// table. Interning is rare enough that the cost is low, but common enough
21 /// that the code is exercised regularly.
23 /// An earlier version of this code used only 4 bytes for `Span`, but that was
24 /// slower because only 80--90% of spans could be stored inline (even less in
25 /// very large crates) and so the interner was used a lot more.
27 /// Inline (compressed) format:
28 /// - `span.base_or_index == span_data.lo`
29 /// - `span.len_or_tag == len == span_data.hi - span_data.lo` (must be `<= MAX_LEN`)
30 /// - `span.ctxt == span_data.ctxt` (must be `<= MAX_CTXT`)
33 /// - `span.base_or_index == index` (indexes into the interner table)
34 /// - `span.len_or_tag == LEN_TAG` (high bit set, all other bits are zero)
35 /// - `span.ctxt == 0`
37 /// The inline form uses 0 for the tag value (rather than 1) so that we don't
38 /// need to mask out the tag bit when getting the length, and so that the
39 /// dummy span can be all zeroes.
41 /// Notes about the choice of field sizes:
42 /// - `base` is 32 bits in both `Span` and `SpanData`, which means that `base`
43 /// values never cause interning. The number of bits needed for `base`
44 /// depends on the crate size. 32 bits allows up to 4 GiB of code in a crate.
45 /// `script-servo` is the largest crate in `rustc-perf`, requiring 26 bits
47 /// - `len` is 15 bits in `Span` (a u16, minus 1 bit for the tag) and 32 bits
48 /// in `SpanData`, which means that large `len` values will cause interning.
49 /// The number of bits needed for `len` does not depend on the crate size.
50 /// The most common number of bits for `len` are 0--7, with a peak usually at
51 /// 3 or 4, and then it drops off quickly from 8 onwards. 15 bits is enough
52 /// for 99.99%+ of cases, but larger values (sometimes 20+ bits) might occur
53 /// dozens of times in a typical crate.
54 /// - `ctxt` is 16 bits in `Span` and 32 bits in `SpanData`, which means that
55 /// large `ctxt` values will cause interning. The number of bits needed for
56 /// `ctxt` values depend partly on the crate size and partly on the form of
57 /// the code. No crates in `rustc-perf` need more than 15 bits for `ctxt`,
58 /// but larger crates might need more than 16 bits.
60 #[derive(Clone, Copy, Eq, PartialEq, Hash)]
67 const LEN_TAG: u16 = 0b1000_0000_0000_0000;
68 const MAX_LEN: u32 = 0b0111_1111_1111_1111;
69 const MAX_CTXT: u32 = 0b1111_1111_1111_1111;
71 /// Dummy span, both position and length are zero, syntax context is zero as well.
72 pub const DUMMY_SP: Span = Span { base_or_index: 0, len_or_tag: 0, ctxt_or_zero: 0 };
76 pub fn new(mut lo: BytePos, mut hi: BytePos, ctxt: SyntaxContext) -> Self {
78 std::mem::swap(&mut lo, &mut hi);
81 let (base, len, ctxt2) = (lo.0, hi.0 - lo.0, ctxt.as_u32());
83 if len <= MAX_LEN && ctxt2 <= MAX_CTXT {
85 Span { base_or_index: base, len_or_tag: len as u16, ctxt_or_zero: ctxt2 as u16 }
88 let index = with_span_interner(|interner| interner.intern(&SpanData { lo, hi, ctxt }));
89 Span { base_or_index: index, len_or_tag: LEN_TAG, ctxt_or_zero: 0 }
94 pub fn data(self) -> SpanData {
95 if self.len_or_tag != LEN_TAG {
97 debug_assert!(self.len_or_tag as u32 <= MAX_LEN);
99 lo: BytePos(self.base_or_index),
100 hi: BytePos(self.base_or_index + self.len_or_tag as u32),
101 ctxt: SyntaxContext::from_u32(self.ctxt_or_zero as u32),
105 debug_assert!(self.ctxt_or_zero == 0);
106 let index = self.base_or_index;
107 with_span_interner(|interner| *interner.get(index))
113 pub struct SpanInterner {
114 spans: FxHashMap<SpanData, u32>,
115 span_data: Vec<SpanData>,
119 fn intern(&mut self, span_data: &SpanData) -> u32 {
120 if let Some(index) = self.spans.get(span_data) {
124 let index = self.spans.len() as u32;
125 self.span_data.push(*span_data);
126 self.spans.insert(*span_data, index);
131 fn get(&self, index: u32) -> &SpanData {
132 &self.span_data[index as usize]
136 // If an interner exists, return it. Otherwise, prepare a fresh one.
138 fn with_span_interner<T, F: FnOnce(&mut SpanInterner) -> T>(f: F) -> T {
139 GLOBALS.with(|globals| f(&mut *globals.span_interner.lock()))