Use `ArcStr` for marker values (#10453)

N.B. After fixing #10430, `ArcStr` became the fastest implementation
(and the gains were significantly reduced, down to 1-2%). See:
https://github.com/astral-sh/uv/pull/10453#issuecomment-2583344414.

## Summary

I tried out a variety of small string crates, but `Arc<str>`
outperformed them, giving a ~10% speed-up:

```console
❯ hyperfine "../arcstr lock" "../flexstr lock" "uv lock" "../arc lock" "../compact_str lock" --prepare "rm -f uv.lock" --min-runs 50 --warmup 20
Benchmark 1: ../arcstr lock
  Time (mean ± σ):     304.6 ms ±   2.3 ms    [User: 302.9 ms, System: 117.8 ms]
  Range (min … max):   299.0 ms … 311.3 ms    50 runs

Benchmark 2: ../flexstr lock
  Time (mean ± σ):     319.2 ms ±   1.7 ms    [User: 317.7 ms, System: 118.2 ms]
  Range (min … max):   316.8 ms … 323.3 ms    50 runs

Benchmark 3: uv lock
  Time (mean ± σ):     330.6 ms ±   1.5 ms    [User: 328.1 ms, System: 139.3 ms]
  Range (min … max):   326.6 ms … 334.2 ms    50 runs

Benchmark 4: ../arc lock
  Time (mean ± σ):     303.0 ms ±   1.2 ms    [User: 301.6 ms, System: 118.4 ms]
  Range (min … max):   300.3 ms … 305.3 ms    50 runs

Benchmark 5: ../compact_str lock
  Time (mean ± σ):     320.4 ms ±   2.0 ms    [User: 318.7 ms, System: 120.8 ms]
  Range (min … max):   317.3 ms … 326.7 ms    50 runs

Summary
  ../arc lock ran
    1.01 ± 0.01 times faster than ../arcstr lock
    1.05 ± 0.01 times faster than ../flexstr lock
    1.06 ± 0.01 times faster than ../compact_str lock
    1.09 ± 0.01 times faster than uv lock
```

Cargo.lock

@@ -5013,6 +5013,7 @@ name = "uv-distribution-types"
 version = "0.0.1"
 dependencies = [
  "anyhow",
+ "arcstr",
  "bitflags 2.6.0",
  "fs-err 3.0.0",
  "itertools 0.14.0",
@@ -5272,6 +5273,7 @@ dependencies = [
 name = "uv-pep508"
 version = "0.6.0"
 dependencies = [
+ "arcstr",
  "boxcar",
  "indexmap",
  "insta",
@@ -5508,6 +5510,7 @@ dependencies = [
 name = "uv-resolver"
 version = "0.0.1"
 dependencies = [
+ "arcstr",
  "clap",
  "dashmap",
  "either",

crates/uv-distribution-types/Cargo.toml

@@ -29,6 +29,7 @@ uv-platform-tags = { workspace = true }
 uv-pypi-types = { workspace = true }
 
 anyhow = { workspace = true }
+arcstr = { workspace = true }
 bitflags = { workspace = true }
 fs-err = { workspace = true }
 itertools = { workspace = true }

crates/uv-distribution-types/src/prioritized_distribution.rs

@@ -1,5 +1,6 @@
 use std::fmt::{Display, Formatter};
 
+use arcstr::ArcStr;
 use tracing::debug;
 
 use uv_distribution_filename::{BuildTag, WheelFilename};
@@ -662,12 +663,12 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                 let mut tag_marker = MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "win32".to_string(),
+                    value: arcstr::literal!("win32"),
                 });
                 tag_marker.and(MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::PlatformMachine,
                     operator: MarkerOperator::Equal,
-                    value: "x86".to_string(),
+                    value: arcstr::literal!("x86"),
                 }));
                 marker.or(tag_marker);
             }
@@ -675,12 +676,12 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                 let mut tag_marker = MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "win32".to_string(),
+                    value: arcstr::literal!("win32"),
                 });
                 tag_marker.and(MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::PlatformMachine,
                     operator: MarkerOperator::Equal,
-                    value: "x86_64".to_string(),
+                    value: arcstr::literal!("x86_64"),
                 }));
                 marker.or(tag_marker);
             }
@@ -688,12 +689,12 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                 let mut tag_marker = MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "win32".to_string(),
+                    value: arcstr::literal!("win32"),
                 });
                 tag_marker.and(MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::PlatformMachine,
                     operator: MarkerOperator::Equal,
-                    value: "arm64".to_string(),
+                    value: arcstr::literal!("arm64"),
                 }));
                 marker.or(tag_marker);
             }
@@ -703,7 +704,7 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                 let mut tag_marker = MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "darwin".to_string(),
+                    value: arcstr::literal!("darwin"),
                 });
 
                 // Parse the macOS version from the tag.
@@ -787,7 +788,7 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                     arch_marker.or(MarkerTree::expression(MarkerExpression::String {
                         key: MarkerValueString::PlatformMachine,
                         operator: MarkerOperator::Equal,
-                        value: (*arch).to_string(),
+                        value: ArcStr::from(*arch),
                     }));
                 }
                 tag_marker.and(arch_marker);
@@ -800,7 +801,7 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                 let mut tag_marker = MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "linux".to_string(),
+                    value: arcstr::literal!("linux"),
                 });
 
                 // Parse the architecture from the tag.
@@ -866,7 +867,7 @@ pub fn implied_markers(filename: &WheelFilename) -> MarkerTree {
                 tag_marker.and(MarkerTree::expression(MarkerExpression::String {
                     key: MarkerValueString::PlatformMachine,
                     operator: MarkerOperator::Equal,
-                    value: arch.to_string(),
+                    value: ArcStr::from(arch),
                 }));
 
                 marker.or(tag_marker);

crates/uv-pep508/Cargo.toml

@@ -23,6 +23,7 @@ uv-fs = { workspace = true }
 uv-normalize = { workspace = true }
 uv-pep440 = { workspace = true }
 
+arcstr = { workspace = true}
 boxcar = { workspace = true }
 indexmap = { workspace = true }
 itertools = { workspace = true }

crates/uv-pep508/src/lib.rs

@@ -1334,17 +1334,17 @@ mod tests {
         let mut b = MarkerTree::expression(MarkerExpression::String {
             key: MarkerValueString::SysPlatform,
             operator: MarkerOperator::Equal,
-            value: "win32".to_string(),
+            value: arcstr::literal!("win32"),
         });
         let mut c = MarkerTree::expression(MarkerExpression::String {
             key: MarkerValueString::OsName,
             operator: MarkerOperator::Equal,
-            value: "linux".to_string(),
+            value: arcstr::literal!("linux"),
         });
         let d = MarkerTree::expression(MarkerExpression::String {
             key: MarkerValueString::ImplementationName,
             operator: MarkerOperator::Equal,
-            value: "cpython".to_string(),
+            value: arcstr::literal!("cpython"),
         });
 
         c.and(d);

crates/uv-pep508/src/marker/algebra.rs

@@ -51,6 +51,7 @@ use std::ops::Bound;
 use std::sync::Mutex;
 use std::sync::MutexGuard;
 
+use arcstr::ArcStr;
 use itertools::{Either, Itertools};
 use rustc_hash::FxHashMap;
 use std::sync::LazyLock;
@@ -287,28 +288,31 @@ impl InternerGuard<'_> {
                 // values in `exclusions`.
                 //
                 // See: https://discuss.python.org/t/clarify-usage-of-platform-system/70900
-                let (key, value) = match (key, value.as_str()) {
+                let (key, value) = match (key, value.as_ref()) {
-                    (MarkerValueString::PlatformSystem, "Windows") => {
+                    (MarkerValueString::PlatformSystem, "Windows") => (
-                        (CanonicalMarkerValueString::SysPlatform, "win32".to_string())
+                        CanonicalMarkerValueString::SysPlatform,
-                    }
+                        arcstr::literal!("win32"),
+                    ),
                     (MarkerValueString::PlatformSystem, "Darwin") => (
                         CanonicalMarkerValueString::SysPlatform,
-                        "darwin".to_string(),
+                        arcstr::literal!("darwin"),
+                    ),
+                    (MarkerValueString::PlatformSystem, "Linux") => (
+                        CanonicalMarkerValueString::SysPlatform,
+                        arcstr::literal!("linux"),
+                    ),
+                    (MarkerValueString::PlatformSystem, "AIX") => (
+                        CanonicalMarkerValueString::SysPlatform,
+                        arcstr::literal!("aix"),
                     ),
-                    (MarkerValueString::PlatformSystem, "Linux") => {
-                        (CanonicalMarkerValueString::SysPlatform, "linux".to_string())
-                    }
-                    (MarkerValueString::PlatformSystem, "AIX") => {
-                        (CanonicalMarkerValueString::SysPlatform, "aix".to_string())
-                    }
                     (MarkerValueString::PlatformSystem, "Emscripten") => (
                         CanonicalMarkerValueString::SysPlatform,
-                        "emscripten".to_string(),
+                        arcstr::literal!("emscripten"),
                     ),
                     // See: https://peps.python.org/pep-0738/#sys
                     (MarkerValueString::PlatformSystem, "Android") => (
                         CanonicalMarkerValueString::SysPlatform,
-                        "android".to_string(),
+                        arcstr::literal!("android"),
                     ),
                     _ => (key.into(), value),
                 };
@@ -869,48 +873,48 @@ impl InternerGuard<'_> {
                 MarkerExpression::String {
                     key: MarkerValueString::OsName,
                     operator: MarkerOperator::Equal,
-                    value: "nt".to_string(),
+                    value: arcstr::literal!("nt"),
                 },
                 MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "linux".to_string(),
+                    value: arcstr::literal!("linux"),
                 },
             ),
             (
                 MarkerExpression::String {
                     key: MarkerValueString::OsName,
                     operator: MarkerOperator::Equal,
-                    value: "nt".to_string(),
+                    value: arcstr::literal!("nt"),
                 },
                 MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "darwin".to_string(),
+                    value: arcstr::literal!("darwin"),
                 },
             ),
             (
                 MarkerExpression::String {
                     key: MarkerValueString::OsName,
                     operator: MarkerOperator::Equal,
-                    value: "nt".to_string(),
+                    value: arcstr::literal!("nt"),
                 },
                 MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "ios".to_string(),
+                    value: arcstr::literal!("ios"),
                 },
             ),
             (
                 MarkerExpression::String {
                     key: MarkerValueString::OsName,
                     operator: MarkerOperator::Equal,
-                    value: "posix".to_string(),
+                    value: arcstr::literal!("posix"),
                 },
                 MarkerExpression::String {
                     key: MarkerValueString::SysPlatform,
                     operator: MarkerOperator::Equal,
-                    value: "win32".to_string(),
+                    value: arcstr::literal!("win32"),
                 },
             ),
         ];
@@ -950,12 +954,12 @@ impl InternerGuard<'_> {
                     MarkerExpression::String {
                         key: MarkerValueString::PlatformSystem,
                         operator: MarkerOperator::Equal,
-                        value: platform_system.to_string(),
+                        value: ArcStr::from(platform_system),
                     },
                     MarkerExpression::String {
                         key: MarkerValueString::SysPlatform,
                         operator: MarkerOperator::Equal,
-                        value: sys_platform.to_string(),
+                        value: ArcStr::from(sys_platform),
                     },
                 ));
             }
@@ -996,13 +1000,13 @@ pub(crate) enum Variable {
     /// string marker and value.
     In {
         key: CanonicalMarkerValueString,
-        value: String,
+        value: ArcStr,
     },
     /// A variable representing a `<value> in <key>` expression for a particular
     /// string marker and value.
     Contains {
         key: CanonicalMarkerValueString,
-        value: String,
+        value: ArcStr,
     },
     /// A variable representing the existence or absence of a given extra.
     ///
@@ -1128,7 +1132,7 @@ pub(crate) enum Edges {
     // Invariant: All ranges are simple, meaning they can be represented by a bounded
     // interval without gaps. Additionally, there are at least two edges in the set.
     String {
-        edges: SmallVec<(Ranges<String>, NodeId)>,
+        edges: SmallVec<(Ranges<ArcStr>, NodeId)>,
     },
     // The edges of a boolean variable, representing the values `true` (the `high` child)
     // and `false` (the `low` child).
@@ -1158,8 +1162,8 @@ impl Edges {
     ///
     /// This function will panic for the `In` and `Contains` marker operators, which
     /// should be represented as separate boolean variables.
-    fn from_string(operator: MarkerOperator, value: String) -> Edges {
+    fn from_string(operator: MarkerOperator, value: ArcStr) -> Edges {
-        let range: Ranges<String> = match operator {
+        let range: Ranges<ArcStr> = match operator {
             MarkerOperator::Equal => Ranges::singleton(value),
             MarkerOperator::NotEqual => Ranges::singleton(value).complement(),
             MarkerOperator::GreaterThan => Ranges::strictly_higher_than(value),

crates/uv-pep508/src/marker/parse.rs

@@ -1,5 +1,5 @@
+use arcstr::ArcStr;
 use std::str::FromStr;
-
 use uv_normalize::ExtraName;
 use uv_pep440::{Version, VersionPattern, VersionSpecifier};
 
@@ -92,7 +92,7 @@ pub(crate) fn parse_marker_value<T: Pep508Url>(
         Some((start_pos, quotation_mark @ ('"' | '\''))) => {
             cursor.next();
             let (start, len) = cursor.take_while(|c| c != quotation_mark);
-            let value = cursor.slice(start, len).to_string();
+            let value = ArcStr::from(cursor.slice(start, len));
             cursor.next_expect_char(quotation_mark, start_pos)?;
             Ok(MarkerValue::QuotedString(value))
         }

crates/uv-pep508/src/marker/simplify.rs

@@ -1,12 +1,14 @@
 use std::fmt;
 use std::ops::Bound;
 
+use arcstr::ArcStr;
 use indexmap::IndexMap;
 use itertools::Itertools;
 use rustc_hash::FxBuildHasher;
-use uv_pep440::{Version, VersionSpecifier};
 use version_ranges::Ranges;
 
+use uv_pep440::{Version, VersionSpecifier};
+
 use crate::{ExtraOperator, MarkerExpression, MarkerOperator, MarkerTree, MarkerTreeKind};
 
 /// Returns a simplified DNF expression for a given marker tree.
@@ -131,7 +133,7 @@ fn collect_dnf(
 
                 let expr = MarkerExpression::String {
                     key: marker.key().into(),
-                    value: marker.value().to_owned(),
+                    value: ArcStr::from(marker.value()),
                     operator,
                 };
 
@@ -150,7 +152,7 @@ fn collect_dnf(
 
                 let expr = MarkerExpression::String {
                     key: marker.key().into(),
-                    value: marker.value().to_owned(),
+                    value: ArcStr::from(marker.value()),
                     operator,
                 };
 

crates/uv-pep508/src/marker/tree.rs

@@ -3,6 +3,7 @@ use std::fmt::{self, Display, Formatter};
 use std::ops::{Bound, Deref};
 use std::str::FromStr;
 
+use arcstr::ArcStr;
 use itertools::Itertools;
 use serde::{de, Deserialize, Deserializer, Serialize, Serializer};
 use version_ranges::Ranges;
@@ -129,7 +130,7 @@ pub enum MarkerValue {
     /// `extra`. This one is special because it's a list and not env but user given
     Extra,
     /// Not a constant, but a user given quoted string with a value inside such as '3.8' or "windows"
-    QuotedString(String),
+    QuotedString(ArcStr),
 }
 
 impl FromStr for MarkerValue {
@@ -272,8 +273,8 @@ impl MarkerOperator {
 
     /// Returns the marker operator and value whose union represents the given range.
     pub fn from_bounds(
-        bounds: (&Bound<String>, &Bound<String>),
+        bounds: (&Bound<ArcStr>, &Bound<ArcStr>),
-    ) -> impl Iterator<Item = (MarkerOperator, String)> {
+    ) -> impl Iterator<Item = (MarkerOperator, ArcStr)> {
         let (b1, b2) = match bounds {
             (Bound::Included(v1), Bound::Included(v2)) if v1 == v2 => {
                 (Some((MarkerOperator::Equal, v1.clone())), None)
@@ -291,7 +292,7 @@ impl MarkerOperator {
     }
 
     /// Returns a value specifier representing the given lower bound.
-    pub fn from_lower_bound(bound: &Bound<String>) -> Option<(MarkerOperator, String)> {
+    pub fn from_lower_bound(bound: &Bound<ArcStr>) -> Option<(MarkerOperator, ArcStr)> {
         match bound {
             Bound::Included(value) => Some((MarkerOperator::GreaterEqual, value.clone())),
             Bound::Excluded(value) => Some((MarkerOperator::GreaterThan, value.clone())),
@@ -300,7 +301,7 @@ impl MarkerOperator {
     }
 
     /// Returns a value specifier representing the given upper bound.
-    pub fn from_upper_bound(bound: &Bound<String>) -> Option<(MarkerOperator, String)> {
+    pub fn from_upper_bound(bound: &Bound<ArcStr>) -> Option<(MarkerOperator, ArcStr)> {
         match bound {
             Bound::Included(value) => Some((MarkerOperator::LessEqual, value.clone())),
             Bound::Excluded(value) => Some((MarkerOperator::LessThan, value.clone())),
@@ -485,7 +486,7 @@ pub enum MarkerExpression {
     String {
         key: MarkerValueString,
         operator: MarkerOperator,
-        value: String,
+        value: ArcStr,
     },
     /// `extra <extra op> '...'` or `'...' <extra op> extra`.
     Extra {
@@ -1383,7 +1384,7 @@ impl Ord for VersionMarkerTree<'_> {
 pub struct StringMarkerTree<'a> {
     id: NodeId,
     key: CanonicalMarkerValueString,
-    map: &'a [(Ranges<String>, NodeId)],
+    map: &'a [(Ranges<ArcStr>, NodeId)],
 }
 
 impl StringMarkerTree<'_> {
@@ -1393,7 +1394,7 @@ impl StringMarkerTree<'_> {
     }
 
     /// The edges of this node, corresponding to possible output ranges of the given variable.
-    pub fn children(&self) -> impl ExactSizeIterator<Item = (&Ranges<String>, MarkerTree)> {
+    pub fn children(&self) -> impl ExactSizeIterator<Item = (&Ranges<ArcStr>, MarkerTree)> {
         self.map
             .iter()
             .map(|(range, node)| (range, MarkerTree(node.negate(self.id))))
@@ -1418,7 +1419,7 @@ impl Ord for StringMarkerTree<'_> {
 #[derive(PartialEq, Eq, Clone, Debug)]
 pub struct InMarkerTree<'a> {
     key: CanonicalMarkerValueString,
-    value: &'a str,
+    value: &'a ArcStr,
     high: NodeId,
     low: NodeId,
 }
@@ -1430,7 +1431,7 @@ impl InMarkerTree<'_> {
     }
 
     /// The value (RHS) for this expression.
-    pub fn value(&self) -> &str {
+    pub fn value(&self) -> &ArcStr {
         self.value
     }
 
@@ -1654,6 +1655,7 @@ mod test {
     use std::str::FromStr;
 
     use insta::assert_snapshot;
+
     use uv_normalize::ExtraName;
     use uv_pep440::Version;
 
@@ -2041,7 +2043,7 @@ mod test {
             MarkerExpression::String {
                 key: MarkerValueString::OsName,
                 operator: MarkerOperator::Equal,
-                value: "nt".to_string(),
+                value: arcstr::literal!("nt")
             }
         );
     }

crates/uv-resolver/Cargo.toml

@@ -38,6 +38,7 @@ uv-types = { workspace = true }
 uv-warnings = { workspace = true }
 uv-workspace = { workspace = true }
 
+arcstr = { workspace = true }
 clap = { workspace = true, features = ["derive"], optional = true }
 dashmap = { workspace = true }
 either = { workspace = true }

crates/uv-resolver/src/resolution/output.rs

@@ -7,6 +7,7 @@ use petgraph::{
     Directed, Direction,
 };
 use rustc_hash::{FxBuildHasher, FxHashMap, FxHashSet};
+
 use uv_configuration::{Constraints, Overrides};
 use uv_distribution::Metadata;
 use uv_distribution_types::{
@@ -726,7 +727,7 @@ impl ResolverOutput {
                     MarkerExpression::String {
                         key: value_string.into(),
                         operator: MarkerOperator::Equal,
-                        value: from_env.to_string(),
+                        value: from_env.into(),
                     }
                 }
             };

crates/uv-resolver/src/resolver/mod.rs

@@ -1424,11 +1424,15 @@ impl<InstalledPackages: InstalledPackagesProvider> ResolverState<InstalledPackag
         // macOS. But if _neither_ version supports Intel macOS, we'd rather use `sys_platform == 'darwin'`
         // instead of `sys_platform == 'darwin' and platform_machine == 'arm64'`, since it's much
         // simpler, and _neither_ version will succeed with Intel macOS anyway.
-        for sys_platform in &["darwin", "linux", "win32"] {
+        for value in [
+            arcstr::literal!("darwin"),
+            arcstr::literal!("linux"),
+            arcstr::literal!("win32"),
+        ] {
             let sys_platform = MarkerTree::expression(MarkerExpression::String {
                 key: MarkerValueString::SysPlatform,
                 operator: MarkerOperator::Equal,
-                value: (*sys_platform).to_string(),
+                value,
             });
             if dist.implied_markers().is_disjoint(sys_platform)
                 && !remainder.is_disjoint(sys_platform)