Auto merge of rust-lang#100150 - notriddle:notriddle/implementors-js, r=GuillaumeGomez

rustdoc: use a more compact encoding for implementors/trait.*.js

The exact amount that this reduces the size of an implementors file depends on whether most of the impls are synthetic or not. For `Send`, it reduces the file from 128K to 112K, while for `Clone` it went from 64K to 44K.

Author: bors

src/librustdoc/html/format.rs

@@ -152,7 +152,7 @@ impl Buffer {
     }
 }
 
-fn comma_sep<T: fmt::Display>(
+pub(crate) fn comma_sep<T: fmt::Display>(
     items: impl Iterator<Item = T>,
     space_after_comma: bool,
 ) -> impl fmt::Display {

src/librustdoc/html/render/write_shared.rs

@@ -1,5 +1,4 @@
 use std::ffi::OsStr;
-use std::fmt::Write;
 use std::fs::{self, File};
 use std::io::prelude::*;
 use std::io::{self, BufReader};
@@ -10,7 +9,8 @@ use std::sync::LazyLock as Lazy;
 use itertools::Itertools;
 use rustc_data_structures::flock;
 use rustc_data_structures::fx::{FxHashMap, FxHashSet};
-use serde::Serialize;
+use serde::ser::SerializeSeq;
+use serde::{Serialize, Serializer};
 
 use super::{collect_paths_for_type, ensure_trailing_slash, Context, BASIC_KEYWORDS};
 use crate::clean::Crate;
@@ -284,25 +284,43 @@ pub(super) fn write_shared(
         cx.write_shared(SharedResource::Unversioned { name }, contents, &options.emit)?;
     }
 
-    fn collect(path: &Path, krate: &str, key: &str) -> io::Result<(Vec<String>, Vec<String>)> {
+    /// Read a file and return all lines that match the `"{crate}":{data},` format,
+    /// and return a tuple `(Vec<DataString>, Vec<CrateNameString>)`.
+    ///
+    /// This forms the payload of files that look like this:
+    ///
+    /// ```javascript
+    /// var data = {
+    /// "{crate1}":{data},
+    /// "{crate2}":{data}
+    /// };
+    /// use_data(data);
+    /// ```
+    ///
+    /// The file needs to be formatted so that *only crate data lines start with `"`*.
+    fn collect(path: &Path, krate: &str) -> io::Result<(Vec<String>, Vec<String>)> {
         let mut ret = Vec::new();
         let mut krates = Vec::new();
 
         if path.exists() {
-            let prefix = format!(r#"{}["{}"]"#, key, krate);
+            let prefix = format!("\"{}\"", krate);
             for line in BufReader::new(File::open(path)?).lines() {
                 let line = line?;
-                if !line.starts_with(key) {
+                if !line.starts_with('"') {
                     continue;
                 }
                 if line.starts_with(&prefix) {
                     continue;
                 }
-                ret.push(line.to_string());
+                if line.ends_with(",") {
+                    ret.push(line[..line.len() - 1].to_string());
+                } else {
+                    // No comma (it's the case for the last added crate line)
+                    ret.push(line.to_string());
+                }
                 krates.push(
-                    line[key.len() + 2..]
-                        .split('"')
-                        .next()
+                    line.split('"')
+                        .find(|s| !s.is_empty())
                         .map(|s| s.to_owned())
                         .unwrap_or_else(String::new),
                 );
@@ -311,6 +329,20 @@ pub(super) fn write_shared(
         Ok((ret, krates))
     }
 
+    /// Read a file and return all lines that match the <code>"{crate}":{data},\</code> format,
+    /// and return a tuple `(Vec<DataString>, Vec<CrateNameString>)`.
+    ///
+    /// This forms the payload of files that look like this:
+    ///
+    /// ```javascript
+    /// var data = JSON.parse('{\
+    /// "{crate1}":{data},\
+    /// "{crate2}":{data}\
+    /// }');
+    /// use_data(data);
+    /// ```
+    ///
+    /// The file needs to be formatted so that *only crate data lines start with `"`*.
     fn collect_json(path: &Path, krate: &str) -> io::Result<(Vec<String>, Vec<String>)> {
         let mut ret = Vec::new();
         let mut krates = Vec::new();
@@ -526,13 +558,27 @@ if (typeof exports !== 'undefined') {exports.searchIndex = searchIndex};
             },
         };
 
-        #[derive(Serialize)]
         struct Implementor {
             text: String,
             synthetic: bool,
             types: Vec<String>,
         }
 
+        impl Serialize for Implementor {
+            fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+            where
+                S: Serializer,
+            {
+                let mut seq = serializer.serialize_seq(None)?;
+                seq.serialize_element(&self.text)?;
+                if self.synthetic {
+                    seq.serialize_element(&1)?;
+                    seq.serialize_element(&self.types)?;
+                }
+                seq.end()
+            }
+        }
+
         let implementors = imps
             .iter()
             .filter_map(|imp| {
@@ -563,9 +609,9 @@ if (typeof exports !== 'undefined') {exports.searchIndex = searchIndex};
         }
 
         let implementors = format!(
-            r#"implementors["{}"] = {};"#,
+            r#""{}":{}"#,
             krate.name(cx.tcx()),
-            serde_json::to_string(&implementors).unwrap()
+            serde_json::to_string(&implementors).expect("failed serde conversion"),
         );
 
         let mut mydst = dst.clone();
@@ -576,16 +622,15 @@ if (typeof exports !== 'undefined') {exports.searchIndex = searchIndex};
         mydst.push(&format!("{}.{}.js", remote_item_type, remote_path[remote_path.len() - 1]));
 
         let (mut all_implementors, _) =
-            try_err!(collect(&mydst, krate.name(cx.tcx()).as_str(), "implementors"), &mydst);
+            try_err!(collect(&mydst, krate.name(cx.tcx()).as_str()), &mydst);
         all_implementors.push(implementors);
         // Sort the implementors by crate so the file will be generated
         // identically even with rustdoc running in parallel.
         all_implementors.sort();
 
-        let mut v = String::from("(function() {var implementors = {};\n");
-        for implementor in &all_implementors {
-            writeln!(v, "{}", *implementor).unwrap();
-        }
+        let mut v = String::from("(function() {var implementors = {\n");
+        v.push_str(&all_implementors.join(",\n"));
+        v.push_str("\n};");
         v.push_str(
             "if (window.register_implementors) {\
                  window.register_implementors(implementors);\

src/librustdoc/html/static/js/main.js

@@ -501,6 +501,10 @@ function loadCss(cssFileName) {
         const synthetic_implementors = document.getElementById("synthetic-implementors-list");
         const inlined_types = new Set();
 
+        const TEXT_IDX = 0;
+        const SYNTHETIC_IDX = 1;
+        const TYPES_IDX = 2;
+
         if (synthetic_implementors) {
             // This `inlined_types` variable is used to avoid having the same implementation
             // showing up twice. For example "String" in the "Sync" doc page.
@@ -536,10 +540,12 @@ function loadCss(cssFileName) {
 
             struct_loop:
             for (const struct of structs) {
-                const list = struct.synthetic ? synthetic_implementors : implementors;
+                const list = struct[SYNTHETIC_IDX] ? synthetic_implementors : implementors;
 
-                if (struct.synthetic) {
-                    for (const struct_type of struct.types) {
+                // The types list is only used for synthetic impls.
+                // If this changes, `main.js` and `write_shared.rs` both need changed.
+                if (struct[SYNTHETIC_IDX]) {
+                    for (const struct_type of struct[TYPES_IDX]) {
                         if (inlined_types.has(struct_type)) {
                             continue struct_loop;
                         }
@@ -548,7 +554,7 @@ function loadCss(cssFileName) {
                 }
 
                 const code = document.createElement("h3");
-                code.innerHTML = struct.text;
+                code.innerHTML = struct[TEXT_IDX];
                 addClass(code, "code-header");
                 addClass(code, "in-band");