Load independent and minimal syntax sets when using --language (#1787)

This significantly speeds up the startup time of bat, since only a single
linked SyntaxDefinition is loaded for each file. The size increase of the
binary is just ~400 kB.

In order for startup time to be improved, the --language arg must be used, and
it must match one of the following names:

"Plain Text", "ActionScript", "AppleScript", "Batch File", "NAnt Build File",
"C#", "C", "CSS", "D", "Diff", "Erlang", "Go", "Haskell", "JSON", "Java
Properties", "BibTeX", "LaTeX Log", "TeX", "Lisp", "Lua", "MATLAB", "Pascal",
"R", "Regular Expression", "Rust", "SQL", "Scala", "Tcl", "XML", "YAML", "Apache
Conf", "ARM Assembly", "Assembly (x86_64)", "CMakeCache", "Comma Separated
Values", "Cabal", "CoffeeScript", "CpuInfo", "Dart Analysis Output", "Dart",
"Dockerfile", "DotENV", "F#", "Friendly Interactive Shell (fish)", "Fortran
(Fixed Form)", "Fortran (Modern)", "Fortran Namelist", "fstab", "GLSL",
"GraphQL", "Groff/troff", "group", "hosts", "INI", "Jinja2", "jsonnet",
"Kotlin", "Less", "LLVM", "Lean", "MemInfo", "Nim", "Ninja", "Nix", "passwd",
"PowerShell", "Protocol Buffer (TEXT)", "Puppet", "Rego", "resolv", "Robot
Framework", "SML", "Strace", "Stylus", "Solidity", "Vyper", "Swift",
"SystemVerilog", "TOML", "Terraform", "TypeScript", "TypeScriptReact",
"Verilog", "VimL", "Zig", "gnuplot", "log", "requirements.txt", "Highlight
non-printables", "Private Key", "varlink"

Later commits will improve startup time for more code paths.

* fix some typos and misspellings

* CHANGELOG.md: Add Performance section (preliminary)

* Add a CHANGELOG.md entry for this PR

CHANGELOG.md

@@ -9,12 +9,15 @@
 
 - Python syntax highlighting no longer suffers from abysmal performance in specific scenarios. See #1688 (@keith-hall)
 
+## Performance
+
+- Load cached assets as fast as integrated assets, see #1753 (@Enselic)
+- Greatly reduce startup time in loop-through mode, e.g. when redirecting output. Instead of *50 ms* - *100 ms*, startup takes *5 ms* - *10 ms*. See #1747 (@Enselic)
+- Reduce startup time by approximately 80% for 91 out of 168 syntaxes when using `--language`. See #1787 (@Enselic)
+
 ## Other
 
 - Add PowerShell completion, see #1826 (@rashil2000)
-- Load cached assets as fast as integrated assets, see #1753 (@Enselic)
-- Greatly reduce startup time in loop-through mode, e.g. when redirecting output. Instead of *50 ms* - *100 ms*, startup takes *5 ms* - *10 ms*. See #1747 (@Enselic)
-
 
 ## Syntaxes
 

src/assets.rs

@@ -1,3 +1,4 @@
+use std::collections::HashMap;
 use std::ffi::OsStr;
 use std::fs;
 use std::path::{Path, PathBuf};
@@ -18,6 +19,14 @@ use crate::syntax_mapping::{MappingTarget, SyntaxMapping};
 pub struct HighlightingAssets {
     syntax_set_cell: LazyCell<SyntaxSet>,
     serialized_syntax_set: SerializedSyntaxSet,
+
+    minimal_syntaxes: MinimalSyntaxes,
+
+    /// Lazily load serialized [SyntaxSet]s from [Self.minimal_syntaxes]. The
+    /// index in this vec matches the index in
+    /// [Self.minimal_syntaxes.serialized_syntax_sets]
+    deserialized_minimal_syntaxes: Vec<LazyCell<SyntaxSet>>,
+
     theme_set: ThemeSet,
     fallback_theme: Option<&'static str>,
 }
@@ -28,12 +37,39 @@ pub struct SyntaxReferenceInSet<'a> {
     pub syntax_set: &'a SyntaxSet,
 }
 
+/// Stores and allows lookup of minimal [SyntaxSet]s. The [SyntaxSet]s are
+/// stored in serialized form, and are deserialized on-demand. This gives good
+/// startup performance since only the necessary [SyntaxReference]s needs to be
+/// deserialized.
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub(crate) struct MinimalSyntaxes {
+    /// Lookup the index into `serialized_syntax_sets` of a [SyntaxSet] by the
+    /// name of any [SyntaxReference] inside the [SyntaxSet]
+    /// (We will later add `by_extension`, `by_first_line`, etc.)
+    pub(crate) by_name: HashMap<String, usize>,
+
+    /// Serialized [SyntaxSet]s. Whether or not this data is compressed is
+    /// decided by [COMPRESS_SERIALIZED_MINIMAL_SYNTAXES]
+    pub(crate) serialized_syntax_sets: Vec<Vec<u8>>,
+}
+
 // Compress for size of ~700 kB instead of ~4600 kB at the cost of ~30% longer deserialization time
 pub(crate) const COMPRESS_SYNTAXES: bool = true;
 
 // Compress for size of ~20 kB instead of ~200 kB at the cost of ~30% longer deserialization time
 pub(crate) const COMPRESS_THEMES: bool = true;
 
+// Compress for size of ~400 kB instead of ~2100 kB at the cost of ~30% longer deserialization time
+pub(crate) const COMPRESS_SERIALIZED_MINIMAL_SYNTAXES: bool = true;
+
+// Whether or not to compress the serialized form of [MinimalSyntaxes]. Shall
+// always be `false`, because the data in
+// [MinimalSyntaxes.serialized_syntax_sets] has already been compressed
+// (assuming [COMPRESS_SERIALIZED_MINIMAL_SYNTAXES] is `true`). The "outer" data
+// structures like `by_name` are tiny. If we compress, deserialization can't do
+// efficient byte-by-byte copy of `serialized_syntax_sets`.
+pub(crate) const COMPRESS_MINIMAL_SYNTAXES: bool = false;
+
 const IGNORED_SUFFIXES: [&str; 13] = [
     // Editor etc backups
     "~",
@@ -55,10 +91,20 @@ const IGNORED_SUFFIXES: [&str; 13] = [
 ];
 
 impl HighlightingAssets {
-    fn new(serialized_syntax_set: SerializedSyntaxSet, theme_set: ThemeSet) -> Self {
+    fn new(
+        serialized_syntax_set: SerializedSyntaxSet,
+        minimal_syntaxes: MinimalSyntaxes,
+        theme_set: ThemeSet,
+    ) -> Self {
+        // Prepare so we can lazily load minimal syntaxes without a mut reference
+        let deserialized_minimal_syntaxes =
+            vec![LazyCell::new(); minimal_syntaxes.serialized_syntax_sets.len()];
+
         HighlightingAssets {
             syntax_set_cell: LazyCell::new(),
             serialized_syntax_set,
+            deserialized_minimal_syntaxes,
+            minimal_syntaxes,
             theme_set,
             fallback_theme: None,
         }
@@ -71,6 +117,11 @@ impl HighlightingAssets {
     pub fn from_cache(cache_path: &Path) -> Result<Self> {
         Ok(HighlightingAssets::new(
             SerializedSyntaxSet::FromFile(cache_path.join("syntaxes.bin")),
+            asset_from_cache(
+                &cache_path.join("minimal_syntaxes.bin"),
+                "minimal syntax sets",
+                COMPRESS_MINIMAL_SYNTAXES,
+            )?,
             asset_from_cache(&cache_path.join("themes.bin"), "theme set", COMPRESS_THEMES)?,
         ))
     }
@@ -78,6 +129,7 @@ impl HighlightingAssets {
     pub fn from_binary() -> Self {
         HighlightingAssets::new(
             SerializedSyntaxSet::FromBinary(get_serialized_integrated_syntaxset()),
+            get_integrated_minimal_syntaxes(),
             get_integrated_themeset(),
         )
     }
@@ -111,6 +163,41 @@ impl HighlightingAssets {
         self.get_theme_set().themes.keys().map(|s| s.as_ref())
     }
 
+    /// Finds a [SyntaxSet] that contains a [SyntaxReference] by its name. First
+    /// tries to find a minimal [SyntaxSet]. If none is found, returns the
+    /// [SyntaxSet] that contains all syntaxes.
+    fn get_syntax_set_by_name(&self, name: &str) -> Result<&SyntaxSet> {
+        let minimal_syntax_set = self
+            .minimal_syntaxes
+            .by_name
+            .get(&name.to_ascii_lowercase())
+            .and_then(|index| self.get_minimal_syntax_set_with_index(*index));
+
+        match minimal_syntax_set {
+            Some(syntax_set) => Ok(syntax_set),
+            None => self.get_syntax_set(),
+        }
+    }
+
+    fn load_minimal_syntax_set_with_index(&self, index: usize) -> Result<SyntaxSet> {
+        let serialized_syntax_set = &self.minimal_syntaxes.serialized_syntax_sets[index];
+        asset_from_contents(
+            &serialized_syntax_set[..],
+            &format!("minimal syntax set {}", index),
+            COMPRESS_SERIALIZED_MINIMAL_SYNTAXES,
+        )
+        .map_err(|_| format!("Could not parse minimal syntax set {}", index).into())
+    }
+
+    fn get_minimal_syntax_set_with_index(&self, index: usize) -> Option<&SyntaxSet> {
+        self.deserialized_minimal_syntaxes
+            .get(index)
+            .and_then(|cell| {
+                cell.try_borrow_with(|| self.load_minimal_syntax_set_with_index(index))
+                    .ok()
+            })
+    }
+
     /// Use [Self::get_syntax_for_file_name] instead
     #[deprecated]
     pub fn syntax_for_file_name(
@@ -167,7 +254,7 @@ impl HighlightingAssets {
         mapping: &SyntaxMapping,
     ) -> Result<SyntaxReferenceInSet> {
         if let Some(language) = language {
-            let syntax_set = self.get_syntax_set()?;
+            let syntax_set = self.get_syntax_set_by_name(language)?;
             syntax_set
                 .find_syntax_by_token(language)
                 .map(|syntax| SyntaxReferenceInSet { syntax, syntax_set })
@@ -320,6 +407,13 @@ pub(crate) fn get_integrated_themeset() -> ThemeSet {
     from_binary(include_bytes!("../assets/themes.bin"), COMPRESS_THEMES)
 }
 
+fn get_integrated_minimal_syntaxes() -> MinimalSyntaxes {
+    from_binary(
+        include_bytes!("../assets/minimal_syntaxes.bin"),
+        COMPRESS_MINIMAL_SYNTAXES,
+    )
+}
+
 pub(crate) fn from_binary<T: serde::de::DeserializeOwned>(v: &[u8], compressed: bool) -> T {
     asset_from_contents(v, "n/a", compressed)
         .expect("data integrated in binary is never faulty, but make sure `compressed` is in sync!")

src/bin/bat/assets.rs

@@ -20,6 +20,7 @@ pub fn cache_dir() -> Cow<'static, str> {
 pub fn clear_assets() {
     clear_asset("themes.bin", "theme set cache");
     clear_asset("syntaxes.bin", "syntax set cache");
+    clear_asset("minimal_syntaxes.bin", "minimal syntax sets cache");
     clear_asset("metadata.yaml", "metadata file");
 }
 

src/build_assets.rs

@@ -37,17 +37,19 @@ pub fn build_assets(
 
     let syntax_set_builder = build_syntax_set_builder(source_dir, include_integrated_assets)?;
 
-    if std::env::var("BAT_PRINT_SYNTAX_DEPENDENCIES").is_ok() {
-        // To trigger this code, run:
-        // BAT_PRINT_SYNTAX_DEPENDENCIES=1 cargo run -- cache --build --source assets --blank --target /tmp
-        print_syntax_dependencies(&syntax_set_builder);
-    }
+    let minimal_syntaxes = build_minimal_syntaxes(&syntax_set_builder, include_integrated_assets)?;
 
     let syntax_set = syntax_set_builder.build();
 
     print_unlinked_contexts(&syntax_set);
 
-    write_assets(&theme_set, &syntax_set, target_dir, current_version)
+    write_assets(
+        &theme_set,
+        &syntax_set,
+        &minimal_syntaxes,
+        target_dir,
+        current_version,
+    )
 }
 
 fn build_theme_set(source_dir: &Path, include_integrated_assets: bool) -> ThemeSet {
@@ -116,6 +118,7 @@ fn print_unlinked_contexts(syntax_set: &SyntaxSet) {
 fn write_assets(
     theme_set: &ThemeSet,
     syntax_set: &SyntaxSet,
+    minimal_syntaxes: &MinimalSyntaxes,
     target_dir: &Path,
     current_version: &str,
 ) -> Result<()> {
@@ -132,6 +135,12 @@ fn write_assets(
         "syntax set",
         COMPRESS_SYNTAXES,
     )?;
+    asset_to_cache(
+        minimal_syntaxes,
+        &target_dir.join("minimal_syntaxes.bin"),
+        "minimal syntax sets",
+        COMPRESS_MINIMAL_SYNTAXES,
+    )?;
 
     print!(
         "Writing metadata to folder {} ... ",
@@ -143,26 +152,65 @@ fn write_assets(
     Ok(())
 }
 
-/// Generates independent [SyntaxSet]s after analyzing dependencies between syntaxes
-/// in a [SyntaxSetBuilder], and then prints the reults.
-fn print_syntax_dependencies(syntax_set_builder: &SyntaxSetBuilder) {
-    println!("Constructing independent SyntaxSets...");
-    let independent_syntax_sets = build_independent_syntax_sets(syntax_set_builder);
+fn print_syntax_set_names(syntax_set: &SyntaxSet) {
+    let names = syntax_set
+        .syntaxes()
+        .iter()
+        .map(|syntax| &syntax.name)
+        .collect::<Vec<_>>();
+    println!("{:?}", names);
+}
 
-    println!("Independent SyntaxSets:");
-    for syntax_set in independent_syntax_sets {
-        let names = syntax_set
-            .syntaxes()
-            .iter()
-            .map(|syntax| &syntax.name)
-            .collect::<Vec<_>>();
-        println!("{:?}", names);
+fn build_minimal_syntaxes(
+    syntax_set_builder: &'_ SyntaxSetBuilder,
+    include_integrated_assets: bool,
+) -> Result<MinimalSyntaxes> {
+    let mut minimal_syntaxes = MinimalSyntaxes {
+        by_name: HashMap::new(),
+        serialized_syntax_sets: vec![],
+    };
+
+    if include_integrated_assets {
+        // Dependency info is not present in integrated assets, so we can't
+        // calculate minimal syntax sets. Return early without any data filled
+        // in. This means that no minimal syntax sets will be available to use, and
+        // the full, slow-to-deserialize, fallback syntax set will be used instead.
+        return Ok(minimal_syntaxes);
     }
+
+    let minimal_syntax_sets_to_serialize = build_minimal_syntax_sets(syntax_set_builder)
+        // For now, only store syntax sets with one syntax, otherwise
+        // the binary grows by several megs
+        .filter(|syntax_set| syntax_set.syntaxes().len() == 1);
+
+    for minimal_syntax_set in minimal_syntax_sets_to_serialize {
+        // Remember what index it is found at
+        let current_index = minimal_syntaxes.serialized_syntax_sets.len();
+
+        for syntax in minimal_syntax_set.syntaxes() {
+            minimal_syntaxes
+                .by_name
+                .insert(syntax.name.to_ascii_lowercase().clone(), current_index);
+        }
+
+        let serialized_syntax_set = asset_to_contents(
+            &minimal_syntax_set,
+            &format!("failed to serialize minimal syntax set {}", current_index),
+            COMPRESS_SERIALIZED_MINIMAL_SYNTAXES,
+        )?;
+
+        // Add last so that it ends up at `current_index`
+        minimal_syntaxes
+            .serialized_syntax_sets
+            .push(serialized_syntax_set);
+    }
+
+    Ok(minimal_syntaxes)
 }
 
 /// Analyzes dependencies between syntaxes in a [SyntaxSetBuilder].
-/// From that, it builds independent [SyntaxSet]s.
-fn build_independent_syntax_sets(
+/// From that, it builds minimal [SyntaxSet]s.
+fn build_minimal_syntax_sets(
     syntax_set_builder: &'_ SyntaxSetBuilder,
 ) -> impl Iterator<Item = SyntaxSet> + '_ {
     let syntaxes = syntax_set_builder.syntaxes();
@@ -170,7 +218,7 @@ fn build_independent_syntax_sets(
     // Build the data structures we need for dependency resolution
     let (syntax_to_dependencies, dependency_to_syntax) = generate_maps(syntaxes);
 
-    // Create one independent SyntaxSet from each (non-hidden) SyntaxDefinition
+    // Create one minimal SyntaxSet from each (non-hidden) SyntaxDefinition
     syntaxes.iter().filter_map(move |syntax| {
         if syntax.hidden {
             return None;
@@ -178,7 +226,15 @@ fn build_independent_syntax_sets(
 
         let mut builder = SyntaxSetDependencyBuilder::new();
         builder.add_with_dependencies(syntax, &syntax_to_dependencies, &dependency_to_syntax);
-        Some(builder.build())
+        let syntax_set = builder.build();
+
+        if std::env::var("BAT_PRINT_SYNTAX_DEPENDENCIES").is_ok() {
+            // To trigger this code, run:
+            // BAT_PRINT_SYNTAX_DEPENDENCIES=1 cargo run -- cache --build --source assets --blank --target /tmp
+            print_syntax_set_names(&syntax_set);
+        }
+
+        Some(syntax_set)
     })
 }