From 6edf05b740070429e57ec18e5c027167dbb17ab4 Mon Sep 17 00:00:00 2001 From: Yotam Ofek Date: Tue, 30 Sep 2025 00:14:03 +0300 Subject: [PATCH 1/2] Add `#[bench]` for librustdoc's syntax highlighter --- src/bootstrap/src/core/builder/mod.rs | 2 +- src/librustdoc/html/highlight/tests.rs | 14 ++++++++++++++ 2 files changed, 15 insertions(+), 1 deletion(-) diff --git a/src/bootstrap/src/core/builder/mod.rs b/src/bootstrap/src/core/builder/mod.rs index 006dea4b98d13..fc06db8f80b9d 100644 --- a/src/bootstrap/src/core/builder/mod.rs +++ b/src/bootstrap/src/core/builder/mod.rs @@ -1145,7 +1145,7 @@ impl<'a> Builder<'a> { test::RunMakeCargo, ), Kind::Miri => describe!(test::Crate), - Kind::Bench => describe!(test::Crate, test::CrateLibrustc), + Kind::Bench => describe!(test::Crate, test::CrateLibrustc, test::CrateRustdoc), Kind::Doc => describe!( doc::UnstableBook, doc::UnstableBookGen, diff --git a/src/librustdoc/html/highlight/tests.rs b/src/librustdoc/html/highlight/tests.rs index 2603e887bead5..4d1bee9b3a1b3 100644 --- a/src/librustdoc/html/highlight/tests.rs +++ b/src/librustdoc/html/highlight/tests.rs @@ -1,6 +1,7 @@ use expect_test::expect_file; use rustc_data_structures::fx::FxIndexMap; use rustc_span::create_default_session_globals_then; +use test::Bencher; use super::{DecorationInfo, write_code}; @@ -81,3 +82,16 @@ let a = 4;"; expect_file!["fixtures/decorations.html"].assert_eq(&html); }); } + +#[bench] +fn bench_html_highlighting(b: &mut Bencher) { + let src = include_str!("../../../../compiler/rustc_ast/src/visit.rs"); + + create_default_session_globals_then(|| { + b.iter(|| { + let mut out = String::new(); + write_code(&mut out, src, None, None, None); + out + }); + }); +} From 2d03ab1486c11f2c5f8a19ae040b94edc090b279 Mon Sep 17 00:00:00 2001 From: Yotam Ofek Date: Tue, 30 Sep 2025 17:12:31 +0300 Subject: [PATCH 2/2] Replace `rustc_span::Span` with a stripped down version for librustdoc's highlighter --- src/librustdoc/html/highlight.rs | 3 +- src/librustdoc/html/render/context.rs | 3 +- src/librustdoc/html/render/mod.rs | 2 +- src/librustdoc/html/render/span_map.rs | 68 ++++++++++++++++++++------ src/librustdoc/html/sources.rs | 7 ++- 5 files changed, 65 insertions(+), 18 deletions(-) diff --git a/src/librustdoc/html/highlight.rs b/src/librustdoc/html/highlight.rs index fad15573cde0c..1dcb4dcc3ff83 100644 --- a/src/librustdoc/html/highlight.rs +++ b/src/librustdoc/html/highlight.rs @@ -12,15 +12,16 @@ use std::iter; use rustc_data_structures::fx::FxIndexMap; use rustc_lexer::{Cursor, FrontmatterAllowed, LiteralKind, TokenKind}; +use rustc_span::BytePos; use rustc_span::edition::Edition; use rustc_span::symbol::Symbol; -use rustc_span::{BytePos, DUMMY_SP, Span}; use super::format; use crate::clean::PrimitiveType; use crate::display::Joined as _; use crate::html::escape::EscapeBodyText; use crate::html::macro_expansion::ExpandedCode; +use crate::html::render::span_map::{DUMMY_SP, Span}; use crate::html::render::{Context, LinkFromSrc}; /// This type is needed in case we want to render links on items to allow to go to their definition. diff --git a/src/librustdoc/html/render/context.rs b/src/librustdoc/html/render/context.rs index 5f92ab2fada9c..4c06d0da47013 100644 --- a/src/librustdoc/html/render/context.rs +++ b/src/librustdoc/html/render/context.rs @@ -30,6 +30,7 @@ use crate::formats::item_type::ItemType; use crate::html::escape::Escape; use crate::html::macro_expansion::ExpandedCode; use crate::html::markdown::{self, ErrorCodes, IdMap, plain_text_summary}; +use crate::html::render::span_map::Span; use crate::html::render::write_shared::write_shared; use crate::html::url_parts_builder::UrlPartsBuilder; use crate::html::{layout, sources, static_files}; @@ -139,7 +140,7 @@ pub(crate) struct SharedContext<'tcx> { /// Correspondence map used to link types used in the source code pages to allow to click on /// links to jump to the type's definition. - pub(crate) span_correspondence_map: FxHashMap, + pub(crate) span_correspondence_map: FxHashMap, pub(crate) expanded_codes: FxHashMap>, /// The [`Cache`] used during rendering. pub(crate) cache: Cache, diff --git a/src/librustdoc/html/render/mod.rs b/src/librustdoc/html/render/mod.rs index 97dcaf57cdfa6..d6371e4dbab3e 100644 --- a/src/librustdoc/html/render/mod.rs +++ b/src/librustdoc/html/render/mod.rs @@ -36,7 +36,7 @@ mod ordered_json; mod print_item; pub(crate) mod sidebar; mod sorted_template; -mod span_map; +pub(crate) mod span_map; mod type_layout; mod write_shared; diff --git a/src/librustdoc/html/render/span_map.rs b/src/librustdoc/html/render/span_map.rs index ef7ce33298d91..bc9417b1bb1de 100644 --- a/src/librustdoc/html/render/span_map.rs +++ b/src/librustdoc/html/render/span_map.rs @@ -8,11 +8,48 @@ use rustc_hir::{ExprKind, HirId, Item, ItemKind, Mod, Node, QPath}; use rustc_middle::hir::nested_filter; use rustc_middle::ty::TyCtxt; use rustc_span::hygiene::MacroKind; -use rustc_span::{BytePos, ExpnKind, Span}; +use rustc_span::{BytePos, ExpnKind}; use crate::clean::{self, PrimitiveType, rustc_span}; use crate::html::sources; +/// This is a stripped down version of [`rustc_span::Span`] that only contains the start and end byte positions of the span. +/// +/// Profiling showed that the `Span` interner was taking up a lot of the run-time when highlighting, and since we +/// never actually use the context and parent that are stored in a normal `Span`, we can replace its usages with this +/// one, which is much cheaper to construct. +#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)] +pub(crate) struct Span { + lo: BytePos, + hi: BytePos, +} + +impl From for Span { + fn from(value: rustc_span::Span) -> Self { + Self { lo: value.lo(), hi: value.hi() } + } +} + +impl Span { + pub(crate) fn lo(self) -> BytePos { + self.lo + } + + pub(crate) fn hi(self) -> BytePos { + self.hi + } + + pub(crate) fn with_lo(self, lo: BytePos) -> Self { + Self { lo, hi: self.hi() } + } + + pub(crate) fn with_hi(self, hi: BytePos) -> Self { + Self { lo: self.lo(), hi } + } +} + +pub(crate) const DUMMY_SP: Span = Span { lo: BytePos(0), hi: BytePos(0) }; + /// This enum allows us to store two different kinds of information: /// /// In case the `span` definition comes from the same crate, we can simply get the `span` and use @@ -96,7 +133,7 @@ impl SpanMapVisitor<'_> { }) .unwrap_or(path.span) }; - self.matches.insert(span, link); + self.matches.insert(span.into(), link); } Res::Local(_) if let Some(span) = self.tcx.hir_res_span(path.res) => { let path_span = if only_use_last_segment @@ -106,11 +143,12 @@ impl SpanMapVisitor<'_> { } else { path.span }; - self.matches.insert(path_span, LinkFromSrc::Local(clean::Span::new(span))); + self.matches.insert(path_span.into(), LinkFromSrc::Local(clean::Span::new(span))); } Res::PrimTy(p) => { // FIXME: Doesn't handle "path-like" primitives like arrays or tuples. - self.matches.insert(path.span, LinkFromSrc::Primitive(PrimitiveType::from(p))); + self.matches + .insert(path.span.into(), LinkFromSrc::Primitive(PrimitiveType::from(p))); } Res::Err => {} _ => {} @@ -127,7 +165,7 @@ impl SpanMapVisitor<'_> { if cspan.inner().is_dummy() || cspan.cnum(self.tcx.sess) != LOCAL_CRATE { return; } - self.matches.insert(span, LinkFromSrc::Doc(item.owner_id.to_def_id())); + self.matches.insert(span.into(), LinkFromSrc::Doc(item.owner_id.to_def_id())); } } @@ -138,7 +176,7 @@ impl SpanMapVisitor<'_> { /// so, we loop until we find the macro definition by using `outer_expn_data` in a loop. /// Finally, we get the information about the macro itself (`span` if "local", `DefId` /// otherwise) and store it inside the span map. - fn handle_macro(&mut self, span: Span) -> bool { + fn handle_macro(&mut self, span: rustc_span::Span) -> bool { if !span.from_expansion() { return false; } @@ -176,7 +214,7 @@ impl SpanMapVisitor<'_> { // The "call_site" includes the whole macro with its "arguments". We only want // the macro name. let new_span = new_span.with_hi(new_span.lo() + BytePos(macro_name.len() as u32)); - self.matches.insert(new_span, link_from_src); + self.matches.insert(new_span.into(), link_from_src); true } @@ -233,7 +271,7 @@ impl<'tcx> Visitor<'tcx> for SpanMapVisitor<'tcx> { intravisit::walk_path(self, path); } - fn visit_qpath(&mut self, qpath: &QPath<'tcx>, id: HirId, _span: Span) { + fn visit_qpath(&mut self, qpath: &QPath<'tcx>, id: HirId, _span: rustc_span::Span) { match *qpath { QPath::TypeRelative(qself, path) => { if matches!(path.res, Res::Err) { @@ -249,7 +287,7 @@ impl<'tcx> Visitor<'tcx> for SpanMapVisitor<'tcx> { self.handle_path(&path, false); } } else { - self.infer_id(path.hir_id, Some(id), path.ident.span); + self.infer_id(path.hir_id, Some(id), path.ident.span.into()); } rustc_ast::visit::try_visit!(self.visit_ty_unambig(qself)); @@ -267,7 +305,7 @@ impl<'tcx> Visitor<'tcx> for SpanMapVisitor<'tcx> { } } - fn visit_mod(&mut self, m: &'tcx Mod<'tcx>, span: Span, id: HirId) { + fn visit_mod(&mut self, m: &'tcx Mod<'tcx>, span: rustc_span::Span, id: HirId) { // To make the difference between "mod foo {}" and "mod foo;". In case we "import" another // file, we want to link to it. Otherwise no need to create a link. if !span.overlaps(m.spans.inner_span) { @@ -275,8 +313,10 @@ impl<'tcx> Visitor<'tcx> for SpanMapVisitor<'tcx> { // name only and not all the "mod foo;". if let Node::Item(item) = self.tcx.hir_node(id) { let (ident, _) = item.expect_mod(); - self.matches - .insert(ident.span, LinkFromSrc::Local(clean::Span::new(m.spans.inner_span))); + self.matches.insert( + ident.span.into(), + LinkFromSrc::Local(clean::Span::new(m.spans.inner_span)), + ); } } else { // If it's a "mod foo {}", we want to look to its documentation page. @@ -288,9 +328,9 @@ impl<'tcx> Visitor<'tcx> for SpanMapVisitor<'tcx> { fn visit_expr(&mut self, expr: &'tcx rustc_hir::Expr<'tcx>) { match expr.kind { ExprKind::MethodCall(segment, ..) => { - self.infer_id(segment.hir_id, Some(expr.hir_id), segment.ident.span) + self.infer_id(segment.hir_id, Some(expr.hir_id), segment.ident.span.into()) } - ExprKind::Call(call, ..) => self.infer_id(call.hir_id, None, call.span), + ExprKind::Call(call, ..) => self.infer_id(call.hir_id, None, call.span.into()), _ => { if self.handle_macro(expr.span) { // We don't want to go deeper into the macro. diff --git a/src/librustdoc/html/sources.rs b/src/librustdoc/html/sources.rs index 9c5518a780e44..c79f63fbc2032 100644 --- a/src/librustdoc/html/sources.rs +++ b/src/librustdoc/html/sources.rs @@ -348,7 +348,12 @@ pub(crate) fn print_src( highlight::write_code( fmt, s, - Some(highlight::HrefContext { context, file_span, root_path, current_href }), + Some(highlight::HrefContext { + context, + file_span: file_span.into(), + root_path, + current_href, + }), Some(decoration_info), Some(line_info), );