rustdoc-json: switch from HashMap to FxHashMap to fix non-determinism

Cargo.lock

@@ -2294,6 +2294,7 @@ dependencies = [
  "anyhow",
  "clap 4.1.4",
  "fs-err",
+ "rustc-hash",
  "rustdoc-json-types",
  "serde",
  "serde_json",
@@ -4846,6 +4847,7 @@ dependencies = [
 name = "rustdoc-json-types"
 version = "0.1.0"
 dependencies = [
+ "rustc-hash",
  "serde",
  "serde_json",
 ]

src/rustdoc-json-types/Cargo.toml

@@ -8,6 +8,7 @@ path = "lib.rs"
 
 [dependencies]
 serde = { version = "1.0", features = ["derive"] }
+rustc-hash = "1.1.0"
 
 [dev-dependencies]
 serde_json = "1.0"

src/rustdoc-json-types/lib.rs

@@ -3,10 +3,9 @@
 //! These types are the public API exposed through the `--output-format json` flag. The [`Crate`]
 //! struct is the root of the JSON blob and all other items are contained within.
 
-use std::collections::HashMap;
-use std::path::PathBuf;
-
+use rustc_hash::FxHashMap;
 use serde::{Deserialize, Serialize};
+use std::path::PathBuf;
 
 /// rustdoc format-version.
 pub const FORMAT_VERSION: u32 = 24;
@@ -24,11 +23,11 @@ pub struct Crate {
     pub includes_private: bool,
     /// A collection of all items in the local crate as well as some external traits and their
     /// items that are referenced locally.
-    pub index: HashMap<Id, Item>,
+    pub index: FxHashMap<Id, Item>,
     /// Maps IDs to fully qualified paths and other info helpful for generating links.
-    pub paths: HashMap<Id, ItemSummary>,
+    pub paths: FxHashMap<Id, ItemSummary>,
     /// Maps `crate_id` of items to a crate name and html_root_url if it exists.
-    pub external_crates: HashMap<u32, ExternalCrate>,
+    pub external_crates: FxHashMap<u32, ExternalCrate>,
     /// A single version number to be used in the future when making backwards incompatible changes
     /// to the JSON output.
     pub format_version: u32,
@@ -54,8 +53,8 @@ pub struct ItemSummary {
     ///
     /// Note that items can appear in multiple paths, and the one chosen is implementation
     /// defined. Currently, this is the full path to where the item was defined. Eg
-    /// [`String`] is currently `["alloc", "string", "String"]` and [`HashMap`] is
-    /// `["std", "collections", "hash", "map", "HashMap"]`, but this is subject to change.
+    /// [`String`] is currently `["alloc", "string", "String"]` and [`HashMap`][`std::collections::HashMap`]
+    /// is `["std", "collections", "hash", "map", "HashMap"]`, but this is subject to change.
     pub path: Vec<String>,
     /// Whether this item is a struct, trait, macro, etc.
     pub kind: ItemKind,
@@ -80,7 +79,7 @@ pub struct Item {
     /// Some("") if there is some documentation but it is empty (EG `#[doc = ""]`).
     pub docs: Option<String>,
     /// This mapping resolves [intra-doc links](https://github.com/rust-lang/rfcs/blob/master/text/1946-intra-rustdoc-links.md) from the docstring to their IDs
-    pub links: HashMap<String, Id>,
+    pub links: FxHashMap<String, Id>,
     /// Stringified versions of the attributes on this item (e.g. `"#[inline]"`)
     pub attrs: Vec<String>,
     pub deprecation: Option<Deprecation>,

src/tools/jsondoclint/Cargo.toml

@@ -9,6 +9,7 @@ edition = "2021"
 anyhow = "1.0.62"
 clap = { version = "4.0.15", features = ["derive"] }
 fs-err = "2.8.1"
+rustc-hash = "1.1.0"
 rustdoc-json-types = { version = "0.1.0", path = "../../rustdoc-json-types" }
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0.85"

src/tools/jsondoclint/src/validator/tests.rs

@@ -1,5 +1,4 @@
-use std::collections::HashMap;
-
+use rustc_hash::FxHashMap;
 use rustdoc_json_types::{Crate, Item, ItemKind, ItemSummary, Visibility, FORMAT_VERSION};
 
 use crate::json_find::SelectorPart;
@@ -27,7 +26,7 @@ fn errors_on_missing_links() {
         root: id("0"),
         crate_version: None,
         includes_private: false,
-        index: HashMap::from_iter([(
+        index: FxHashMap::from_iter([(
             id("0"),
             Item {
                 name: Some("root".to_owned()),
@@ -36,7 +35,7 @@ fn errors_on_missing_links() {
                 span: None,
                 visibility: Visibility::Public,
                 docs: None,
-                links: HashMap::from_iter([("Not Found".to_owned(), id("1"))]),
+                links: FxHashMap::from_iter([("Not Found".to_owned(), id("1"))]),
                 attrs: vec![],
                 deprecation: None,
                 inner: ItemEnum::Module(Module {
@@ -46,8 +45,8 @@ fn errors_on_missing_links() {
                 }),
             },
         )]),
-        paths: HashMap::new(),
-        external_crates: HashMap::new(),
+        paths: FxHashMap::default(),
+        external_crates: FxHashMap::default(),
         format_version: rustdoc_json_types::FORMAT_VERSION,
     };
 
@@ -73,7 +72,7 @@ fn errors_on_local_in_paths_and_not_index() {
         root: id("0:0:1572"),
         crate_version: None,
         includes_private: false,
-        index: HashMap::from_iter([
+        index: FxHashMap::from_iter([
             (
                 id("0:0:1572"),
                 Item {
@@ -83,7 +82,7 @@ fn errors_on_local_in_paths_and_not_index() {
                     span: None,
                     visibility: Visibility::Public,
                     docs: None,
-                    links: HashMap::from_iter([(("prim@i32".to_owned(), id("0:1:1571")))]),
+                    links: FxHashMap::from_iter([(("prim@i32".to_owned(), id("0:1:1571")))]),
                     attrs: Vec::new(),
                     deprecation: None,
                     inner: ItemEnum::Module(Module {
@@ -102,22 +101,22 @@ fn errors_on_local_in_paths_and_not_index() {
                     span: None,
                     visibility: Visibility::Public,
                     docs: None,
-                    links: HashMap::default(),
+                    links: FxHashMap::default(),
                     attrs: Vec::new(),
                     deprecation: None,
                     inner: ItemEnum::Primitive(Primitive { name: "i32".to_owned(), impls: vec![] }),
                 },
             ),
         ]),
-        paths: HashMap::from_iter([(
+        paths: FxHashMap::from_iter([(
             id("0:1:1571"),
             ItemSummary {
                 crate_id: 0,
                 path: vec!["microcore".to_owned(), "i32".to_owned()],
                 kind: ItemKind::Primitive,
             },
         )]),
-        external_crates: HashMap::default(),
+        external_crates: FxHashMap::default(),
         format_version: rustdoc_json_types::FORMAT_VERSION,
     };
 
@@ -137,7 +136,7 @@ fn checks_local_crate_id_is_correct() {
         root: id("root"),
         crate_version: None,
         includes_private: false,
-        index: HashMap::from_iter([(
+        index: FxHashMap::from_iter([(
             id("root"),
             Item {
                 id: id("root"),
@@ -146,7 +145,7 @@ fn checks_local_crate_id_is_correct() {
                 span: None,
                 visibility: Visibility::Public,
                 docs: None,
-                links: HashMap::default(),
+                links: FxHashMap::default(),
                 attrs: Vec::new(),
                 deprecation: None,
                 inner: ItemEnum::Module(Module {
@@ -156,8 +155,8 @@ fn checks_local_crate_id_is_correct() {
                 }),
             },
         )]),
-        paths: HashMap::default(),
-        external_crates: HashMap::default(),
+        paths: FxHashMap::default(),
+        external_crates: FxHashMap::default(),
         format_version: FORMAT_VERSION,
     };
     check(&krate, &[]);

tests/run-make/rustdoc-verify-output-files/Makefile

@@ -22,15 +22,11 @@ all:
 	# Check if expected json file is generated
 	[ -e $(OUTPUT_DIR)/foobar.json ]
 
-	# TODO
-	# We should re-generate json doc once again and compare the diff with previously
-	# generated one. Because layout of json docs changes in each compilation, we can't
-	# do that currently.
-	#
-	# See https://github.com/rust-lang/rust/issues/103785#issuecomment-1307425590 for details.
+	# Copy first json output to check if it's exactly same after second compilation
+	cp -R $(OUTPUT_DIR)/foobar.json $(TMP_OUTPUT_DIR)/foobar.json
 
-	# remove generated json doc
-	rm $(OUTPUT_DIR)/foobar.json
+	# Generate json doc on the same output
+	$(RUSTDOC) src/lib.rs --crate-name foobar --crate-type lib --out-dir $(OUTPUT_DIR) -Z unstable-options --output-format json
 
-	# Check if json doc compilation broke any of the html files generated previously
+	# Check if all docs(including both json and html formats) are still the same after multiple compilations
 	$(DIFF) -r -q $(OUTPUT_DIR) $(TMP_OUTPUT_DIR)