release: version 0.2.1

Author: mpizenberg

.github/workflows/cargo_publish_dry_run.yml

@@ -0,0 +1,40 @@
+# Runs `cargo publish --dry-run` before another release
+
+name: Check crate publishing works
+on:
+  pull_request:
+    branches: [ release ]
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  cargo_publish_dry_run:
+    name: Publishing works
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - name: Install stable Rust
+        uses: actions-rs/toolchain@v1
+        with:
+          toolchain: stable
+          profile: minimal
+
+      - name: Get Cargo version
+        id: cargo_version
+        run: echo "::set-output name=version::$(cargo -V | tr -d ' ')"
+        shell: bash
+
+      - name: Download cache
+        uses: actions/cache@v2
+        with:
+          path: |
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+            target/
+          key: ${{ runner.os }}-${{ steps.cargo_version.outputs.version }}-${{ hashFiles('Cargo.toml') }}
+          restore-keys: ${{ runner.os }}-${{ steps.cargo_version.outputs.version }}
+
+      - name: Run `cargo publish --dry-run`
+        run: cargo publish --dry-run

.github/workflows/ci.yml

@@ -2,7 +2,7 @@ name: CI
 on:
   pull_request:
   push:
-    branches: [ master, dev ]
+    branches: [ release, dev ]
   schedule: [ cron: "0 6 * * 4" ]
 
 env:

CHANGELOG.md

@@ -2,9 +2,37 @@
 
 All notable changes to this project will be documented in this file.
 
-## Unreleased [(diff)][diff-unreleased]
+## Unreleased [(diff)][unreleased-diff]
 
-## [0.2.0] - 2020-11-19 - [(diff with 0.1.0)][diff-0.2.0]
+## [0.2.1] - 2021-06-30 - [(diff with 0.2.0)][0.2.0-diff]
+
+This release is focused on performance improvements and code readability, without any change to the public API.
+
+The code tends to be simpler around tricky parts of the algorithm such as conflict resolution.
+Some data structures have been rewritten (with no unsafe) to lower memory usage.
+Depending on scenarios, version 0.2.1 is 3 to 8 times faster than 0.2.0.
+As an example, solving all elm package versions existing went from 580ms to 175ms on my laptop.
+While solving a specific subset of packages from crates.io went from 2.5s to 320ms on my laptop.
+
+Below are listed all the important changes in the internal parts of the API.
+
+#### Added
+
+- New `SmallVec` data structure (with no unsafe) using fixed size arrays for up to 2 entries.
+- New `SmallMap` data structure (with no unsafe) using fixed size arrays for up to 2 entries.
+- New `Arena` data structure (with no unsafe) backed by a `Vec` and indexed with `Id<T>` where `T` is phantom data.
+
+#### Changed
+
+- Updated the `large_case` benchmark to run with both u16 and string package identifiers in registries.
+- Use the new `Arena` for the incompatibility store, and use its `Id<T>` identifiers to reference incompatibilities instead of full owned copies in the `incompatibilities` field of the solver `State`.
+- Save satisfier indices of each package involved in an incompatibility when looking for its satisfier. This speeds up the search for the previous satisfier.
+- Early unit propagation loop restart at the first conflict found instead of continuing evaluation for the current package.
+- Index incompatibilities by package in a hash map instead of using a vec.
+- Keep track of already contradicted incompatibilities in a `Set` until the next backtrack to speed up unit propagation.
+- Unify `history` and `memory` in `partial_solution` under a unique hash map indexed by packages. This should speed up access to relevan terms in conflict resolution.
+
+## [0.2.0] - 2020-11-19 - [(diff with 0.1.0)][0.1.0-diff]
 
 This release brings many important improvements to PubGrub.
 The gist of it is:
@@ -78,7 +106,9 @@ The gist of it is:
 
 #### Changed
 
-- CI workflow was improved (`./github/workflows/`), including a check for [Conventional Commits](https://www.conventionalcommits.org/en/v1.0.0/) and [Clippy ](https://github.com/rust-lang/rust-clippy) for source code linting.
+- CI workflow was improved (`./github/workflows/`), including a check for
+  [Conventional Commits](https://www.conventionalcommits.org/en/v1.0.0/) and
+  [Clippy](https://github.com/rust-lang/rust-clippy) for source code linting.
 - Using SPDX license identifiers instead of MPL-2.0 classic file headers.
 - `State.incompatibilities` is now wrapped inside a `Rc`.
 - `DecisionLevel(u32)` is used in place of `usize` for partial solution decision levels.
@@ -131,7 +161,10 @@ The gist of it is:
 - `.gitignore` configured for a Rust project.
 - `.github/workflows/` CI to automatically build, test and document on push and pull requests.
 
+[0.2.1]: https://github.com/pubgrub-rs/pubgrub/releases/tag/v0.2.1
 [0.2.0]: https://github.com/pubgrub-rs/pubgrub/releases/tag/v0.2.0
 [0.1.0]: https://github.com/pubgrub-rs/pubgrub/releases/tag/v0.1.0
-[diff-unreleased]: https://github.com/pubgrub-rs/pubgrub/compare/release...dev
-[diff-0.2.0]: https://github.com/mpizenberg/elm-pointer-events/compare/v0.1.0...v0.2.0
+
+[unreleased-diff]: https://github.com/pubgrub-rs/pubgrub/compare/release...dev
+[0.2.0-diff]: https://github.com/pubgrub-rs/pubgrub/compare/v0.2.0...v0.2.1
+[0.1.0-diff]: https://github.com/pubgrub-rs/pubgrub/compare/v0.1.0...v0.2.0

Cargo.toml

@@ -2,12 +2,16 @@
 
 [package]
 name = "pubgrub"
-version = "0.2.0"
-authors = ["Matthieu Pizenberg <[email protected]>", "Alex Tokarev <[email protected]>", "Jacob Finkelman <[email protected]>"]
+version = "0.2.1"
+authors = [
+    "Matthieu Pizenberg <[email protected]>",
+    "Alex Tokarev <[email protected]>",
+    "Jacob Finkelman <[email protected]>",
+]
 edition = "2018"
 description = "PubGrub version solving algorithm"
 readme = "README.md"
-repository = "https://github.com/mpizenberg/pubgrub-rs"
+repository = "https://github.com/pubgrub-rs/pubgrub"
 license = "MPL-2.0"
 keywords = ["dependency", "pubgrub", "semver", "solver", "version"]
 categories = ["algorithms"]
@@ -22,8 +26,8 @@ serde = { version = "1.0", features = ["derive"], optional = true }
 
 [dev-dependencies]
 proptest = "0.10.1"
-ron="0.6"
-varisat="0.2.2"
+ron = "0.6"
+varisat = "0.2.2"
 criterion = "0.3"
 
 [[bench]]

benches/large_case.rs

@@ -4,30 +4,44 @@ use std::time::Duration;
 extern crate criterion;
 use self::criterion::*;
 
+use pubgrub::package::Package;
 use pubgrub::solver::{resolve, OfflineDependencyProvider};
-use pubgrub::version::NumberVersion;
+use pubgrub::version::{NumberVersion, SemanticVersion, Version};
+use serde::de::Deserialize;
+use std::hash::Hash;
+
+fn bench<'a, P: Package + Deserialize<'a>, V: Version + Hash + Deserialize<'a>>(
+    b: &mut Bencher,
+    case: &'a str,
+) {
+    let dependency_provider: OfflineDependencyProvider<P, V> = ron::de::from_str(&case).unwrap();
+
+    b.iter(|| {
+        for p in dependency_provider.packages() {
+            for n in dependency_provider.versions(p).unwrap() {
+                let _ = resolve(&dependency_provider, p.clone(), n.clone());
+            }
+        }
+    });
+}
 
 fn bench_nested(c: &mut Criterion) {
     let mut group = c.benchmark_group("large_cases");
     group.measurement_time(Duration::from_secs(20));
 
     for case in std::fs::read_dir("test-examples").unwrap() {
         let case = case.unwrap().path();
-
-        group.bench_function(
-            format!("{}", case.file_name().unwrap().to_string_lossy()),
-            |b| {
-                let s = std::fs::read_to_string(&case).unwrap();
-                let dependency_provider: OfflineDependencyProvider<u16, NumberVersion> =
-                    ron::de::from_str(&s).unwrap();
-
-                b.iter(|| {
-                    for &n in dependency_provider.versions(&0).unwrap() {
-                        let _ = resolve(&dependency_provider, 0, n);
-                    }
-                });
-            },
-        );
+        let name = case.file_name().unwrap().to_string_lossy();
+        let data = std::fs::read_to_string(&case).unwrap();
+        if name.ends_with("u16_NumberVersion.ron") {
+            group.bench_function(name, |b| {
+                bench::<u16, NumberVersion>(b, &data);
+            });
+        } else if name.ends_with("str_SemanticVersion.ron") {
+            group.bench_function(name, |b| {
+                bench::<&str, SemanticVersion>(b, &data);
+            });
+        }
     }
 
     group.finish();

src/internal/arena.rs

@@ -0,0 +1,122 @@
+use std::{
+    fmt,
+    hash::{Hash, Hasher},
+    marker::PhantomData,
+    ops::{Index, Range},
+};
+
+/// The index of a value allocated in an arena that holds `T`s.
+///
+/// The Clone, Copy and other traits are defined manually because
+/// deriving them adds some additional constraints on the `T` generic type
+/// that we actually don't need since it is phantom.
+///
+/// <https://github.com/rust-lang/rust/issues/26925>
+pub struct Id<T> {
+    raw: u32,
+    _ty: PhantomData<fn() -> T>,
+}
+
+impl<T> Clone for Id<T> {
+    fn clone(&self) -> Self {
+        *self
+    }
+}
+
+impl<T> Copy for Id<T> {}
+
+impl<T> PartialEq for Id<T> {
+    fn eq(&self, other: &Id<T>) -> bool {
+        self.raw == other.raw
+    }
+}
+
+impl<T> Eq for Id<T> {}
+
+impl<T> Hash for Id<T> {
+    fn hash<H: Hasher>(&self, state: &mut H) {
+        self.raw.hash(state)
+    }
+}
+
+impl<T> fmt::Debug for Id<T> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let mut type_name = std::any::type_name::<T>();
+        if let Some(id) = type_name.rfind(':') {
+            type_name = &type_name[id + 1..]
+        }
+        write!(f, "Id::<{}>({})", type_name, self.raw)
+    }
+}
+
+impl<T> Id<T> {
+    pub fn into_raw(self) -> usize {
+        self.raw as usize
+    }
+    fn from(n: u32) -> Self {
+        Self {
+            raw: n as u32,
+            _ty: PhantomData,
+        }
+    }
+    pub fn range_to_iter(range: Range<Self>) -> impl Iterator<Item = Self> {
+        let start = range.start.raw;
+        let end = range.end.raw;
+        (start..end).map(Self::from)
+    }
+}
+
+/// Yet another index-based arena.
+///
+/// An arena is a kind of simple grow-only allocator, backed by a `Vec`
+/// where all items have the same lifetime, making it easier
+/// to have references between those items.
+/// They are all dropped at once when the arena is dropped.
+#[derive(Clone, PartialEq, Eq)]
+pub struct Arena<T> {
+    data: Vec<T>,
+}
+
+impl<T: fmt::Debug> fmt::Debug for Arena<T> {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result {
+        fmt.debug_struct("Arena")
+            .field("len", &self.data.len())
+            .field("data", &self.data)
+            .finish()
+    }
+}
+
+impl<T> Arena<T> {
+    pub fn new() -> Arena<T> {
+        Arena { data: Vec::new() }
+    }
+
+    pub fn alloc(&mut self, value: T) -> Id<T> {
+        let raw = self.data.len();
+        self.data.push(value);
+        Id::from(raw as u32)
+    }
+
+    pub fn alloc_iter<I: Iterator<Item = T>>(&mut self, values: I) -> Range<Id<T>> {
+        let start = Id::from(self.data.len() as u32);
+        values.for_each(|v| {
+            self.alloc(v);
+        });
+        let end = Id::from(self.data.len() as u32);
+        Range { start, end }
+    }
+}
+
+impl<T> Index<Id<T>> for Arena<T> {
+    type Output = T;
+    fn index(&self, id: Id<T>) -> &T {
+        &self.data[id.raw as usize]
+    }
+}
+
+impl<T> Index<Range<Id<T>>> for Arena<T> {
+    type Output = [T];
+    fn index(&self, id: Range<Id<T>>) -> &[T] {
+        &self.data[(id.start.raw as usize)..(id.end.raw as usize)]
+    }
+}