Make URL strategy better at recognizing URLs. (#873)

Author: lrhn

pkgs/path/CHANGELOG.md

@@ -1,5 +1,12 @@
 ## 1.9.2-wip
 
+- Make `Url` style better at recognizing schemes and authorities.
+  Only consider a path as being a schemed URL if it starts with a
+  valid scheme.
+  Recognize `#` and `?` as ending an authority or path. Remove
+  queries and fragments when when normalizing, include them 
+  in/as the last segment when splitting.
+
 ## 1.9.1
 
 - Require Dart 3.4

pkgs/path/lib/src/characters.dart

@@ -2,16 +2,24 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
-/// This library contains character-code definitions.
+// Character-code constants.
+
+const hash = 0x23;
+const percent = 0x25;
 const plus = 0x2b;
 const minus = 0x2d;
 const period = 0x2e;
 const slash = 0x2f;
 const zero = 0x30;
 const nine = 0x39;
 const colon = 0x3a;
+const question = 0x3f;
 const upperA = 0x41;
 const upperZ = 0x5a;
 const lowerA = 0x61;
+const lowerE = 0x65;
+const lowerF = 0x66;
+const lowerI = 0x69;
+const lowerL = 0x6c;
 const lowerZ = 0x7a;
 const backslash = 0x5c;

pkgs/path/lib/src/context.dart

@@ -372,18 +372,22 @@ class Context {
   /// Canonicalizes [path].
   ///
   /// This is guaranteed to return the same path for two different input paths
-  /// if and only if both input paths point to the same location. Unlike
+  /// only if both input paths point to the same location. Unlike
   /// [normalize], it returns absolute paths when possible and canonicalizes
-  /// ASCII case on Windows.
+  /// ASCII case on Windows, and scheme and authority case for URLs (but does
+  /// not normalize or canonicalize `%`-escapes.)
   ///
   /// Note that this does not resolve symlinks.
   ///
   /// If you want a map that uses path keys, it's probably more efficient to use
   /// a Map with [equals] and [hash] specified as the callbacks to use for keys
   /// than it is to canonicalize every key.
+  ///
   String canonicalize(String path) {
     path = absolute(path);
-    if (style != Style.windows && !_needsNormalization(path)) return path;
+    // Windows and URL styles need to case-canonicalize, even if it doesn't
+    // need to normalize anything.
+    if (style == Style.posix && !_needsNormalization(path)) return path;
 
     final parsed = _parse(path);
     parsed.normalize(canonicalize: true);
@@ -395,7 +399,7 @@ class Context {
   ///
   /// Note that this is *not* guaranteed to return the same result for two
   /// equivalent input paths. For that, see [canonicalize]. Or, if you're using
-  /// paths as map keys use [equals] and [hash] as the key callbacks.
+  /// paths as map keys, use [equals] and [hash] as the key callbacks.
   ///
   ///     context.normalize('path/./to/..//file.text'); // -> 'path/file.txt'
   String normalize(String path) {
@@ -408,68 +412,76 @@ class Context {
 
   /// Returns whether [path] needs to be normalized.
   bool _needsNormalization(String path) {
-    var start = 0;
-    final codeUnits = path.codeUnits;
-    int? previousPrevious;
-    int? previous;
+    // Empty paths are normalized to ".".
+    if (path.isEmpty) return true;
+
+    // At start, no previous separator.
+    const stateStart = 0;
+
+    // Previous character was a separator.
+    const stateSeparator = 1;
+
+    // Added to state for each `.` seen.
+    const stateDotCount = 2;
+
+    // Path segment that contains anything other than nothing, `.` or `..`.
+    //
+    // Includes any value at or above this one.
+    const stateNotDots = 6;
+
+    // Current state of the last few characters.
+    //
+    // Seeing a separator resets to [stateSeparator].
+    // Seeing a `.` adds [stateDotCount].
+    // Seeing any non-separator or more than two dots will
+    // bring the value above [stateNotDots].
+    // (The separator may be optional at the start, seeing one is fine,
+    // and seeing dots will start counting.)
+    // (That is, `/` has value 1, `/.` value 3, ``/..` value 5, and anything
+    // else is 6 or above, except at the very start where empty path, `.`
+    // and `..` have values 0, 2 and 4.)
+    var state = stateStart;
 
     // Skip past the root before we start looking for snippets that need
     // normalization. We want to normalize "//", but not when it's part of
     // "http://".
-    final root = style.rootLength(path);
-    if (root != 0) {
-      start = root;
-      previous = chars.slash;
-
+    final start = style.rootLength(path);
+    if (start != 0) {
+      if (style.isSeparator(path.codeUnitAt(start - 1))) {
+        state = stateSeparator;
+      }
       // On Windows, the root still needs to be normalized if it contains a
       // forward slash.
       if (style == Style.windows) {
-        for (var i = 0; i < root; i++) {
-          if (codeUnits[i] == chars.slash) return true;
+        for (var i = 0; i < start; i++) {
+          if (path.codeUnitAt(i) == chars.slash) return true;
         }
       }
     }
 
-    for (var i = start; i < codeUnits.length; i++) {
-      final codeUnit = codeUnits[i];
+    for (var i = start; i < path.length; i++) {
+      final codeUnit = path.codeUnitAt(i);
       if (style.isSeparator(codeUnit)) {
+        // If ending empty, `.` or `..` path segment.
+        if (state >= stateSeparator && state < stateNotDots) return true;
         // Forward slashes in Windows paths are normalized to backslashes.
         if (style == Style.windows && codeUnit == chars.slash) return true;
-
-        // Multiple separators are normalized to single separators.
-        if (previous != null && style.isSeparator(previous)) return true;
-
-        // Single dots and double dots are normalized to directory traversals.
-        //
-        // This can return false positives for ".../", but that's unlikely
-        // enough that it's probably not going to cause performance issues.
-        if (previous == chars.period &&
-            (previousPrevious == null ||
-                previousPrevious == chars.period ||
-                style.isSeparator(previousPrevious))) {
+        state = stateSeparator;
+      } else if (codeUnit == chars.period) {
+        state += stateDotCount;
+      } else {
+        state = stateNotDots;
+        if (style == Style.url &&
+            (codeUnit == chars.question || codeUnit == chars.hash)) {
+          // Normalize away `?` query parts and `#` fragment parts in URL
+          // styled paths.
           return true;
         }
       }
-
-      previousPrevious = previous;
-      previous = codeUnit;
-    }
-
-    // Empty paths are normalized to ".".
-    if (previous == null) return true;
-
-    // Trailing separators are removed.
-    if (style.isSeparator(previous)) return true;
-
-    // Single dots and double dots are normalized to directory traversals.
-    if (previous == chars.period &&
-        (previousPrevious == null ||
-            style.isSeparator(previousPrevious) ||
-            previousPrevious == chars.period)) {
-      return true;
     }
 
-    return false;
+    // Otherwise only normalize if there are separators and single/double dots.
+    return state >= stateSeparator && state < stateNotDots;
   }
 
   /// Attempts to convert [path] to an equivalent relative path relative to
@@ -1020,7 +1032,9 @@ class Context {
   /// Returns the path represented by [uri], which may be a [String] or a [Uri].
   ///
   /// For POSIX and Windows styles, [uri] must be a `file:` URI. For the URL
-  /// style, this will just convert [uri] to a string.
+  /// style, this will just convert [uri] to a string, but if the input was
+  /// a string, it will be parsed and normalized as a [Uri] first.
+  ///
   ///
   ///     // POSIX
   ///     context.fromUri('file:///path/to/foo')
@@ -1036,7 +1050,11 @@ class Context {
   ///
   /// If [uri] is relative, a relative path will be returned.
   ///
+  ///     // POSIX
   ///     path.fromUri('path/to/foo'); // -> 'path/to/foo'
+  ///
+  ///     // Windows
+  ///     path.fromUri('/C:/foo'); // -> r'C:\foo`
   String fromUri(Object? uri) => style.pathFromUri(_parseUri(uri!));
 
   /// Returns the URI that represents [path].

pkgs/path/lib/src/internal_style.dart

@@ -32,12 +32,21 @@ abstract class InternalStyle extends Style {
   /// "usr", an additional "/" is needed (making "file:///usr").
   bool needsSeparator(String path);
 
-  /// Returns the number of characters of the root part.
+  /// The number of characters of the root part.
   ///
   /// Returns 0 if the path is relative and 1 if the path is root-relative.
   ///
   /// If [withDrive] is `true`, this should include the drive letter for `file:`
   /// URLs. Non-URL styles may ignore the parameter.
+  ///
+  /// For URL style, the root includes any URI scheme and host/authority part.
+  /// If either of those are included, a leading `/` of the path is not
+  /// included unless it's followed by a drive letter. A slash after a
+  /// drive letter is included. If a URL style path starts with a single `/`,
+  /// it is included in the root.
+  /// If a URL style path has a scheme, but not authority, and the path
+  /// does not start with a `/`, the first path segment of the path is
+  /// considered part of the root.
   int rootLength(String path, {bool withDrive = false});
 
   /// Gets the root prefix of [path] if path is absolute. If [path] is relative,

pkgs/path/lib/src/parsed_path.dart

@@ -2,8 +2,10 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
+import 'characters.dart' as chars;
 import 'internal_style.dart';
 import 'style.dart';
+import 'utils.dart' show endOfScheme, removeQueryFragment;
 
 class ParsedPath {
   /// The [InternalStyle] that was used to parse this path.
@@ -35,7 +37,7 @@ class ParsedPath {
   /// one.
   String extension([int level = 1]) => _splitExtension(level)[1];
 
-  /// `true` if this is an absolute path.
+  /// Whether this is an absolute path.
   bool get isAbsolute => root != null;
 
   factory ParsedPath.parse(String path, InternalStyle style) {
@@ -58,11 +60,17 @@ class ParsedPath {
     }
 
     for (var i = start; i < path.length; i++) {
-      if (style.isSeparator(path.codeUnitAt(i))) {
+      final codeUnit = path.codeUnitAt(i);
+      if (style.isSeparator(codeUnit)) {
         parts.add(path.substring(start, i));
         separators.add(path[i]);
         start = i + 1;
       }
+      if (style == Style.url &&
+          (codeUnit == chars.question || codeUnit == chars.hash)) {
+        // Include `?` and `#` in final path segment.
+        break;
+      }
     }
 
     // Add the final part, if any.
@@ -101,6 +109,13 @@ class ParsedPath {
     // Handle '.', '..', and empty parts.
     var leadingDoubles = 0;
     final newParts = <String>[];
+    if (style == Style.url && parts.isNotEmpty) {
+      parts.last = removeQueryFragment(parts.last);
+      if (canonicalize && endOfScheme(parts.first, 0) > 0) {
+        // Normalize scheme and authority.
+        parts.first = parts.first.toLowerCase();
+      }
+    }
     for (var part in parts) {
       if (part == '.' || part == '') {
         // Do nothing. Ignore it.
@@ -131,14 +146,21 @@ class ParsedPath {
     parts = newParts;
     separators =
         List.filled(newParts.length + 1, style.separator, growable: true);
-    if (!isAbsolute || newParts.isEmpty || !style.needsSeparator(root!)) {
+
+    final root = this.root;
+    if (root == null || newParts.isEmpty || !style.needsSeparator(root)) {
       separators[0] = '';
     }
 
-    // Normalize the Windows root if needed.
-    if (root != null && style == Style.windows) {
-      if (canonicalize) root = root!.toLowerCase();
-      root = root!.replaceAll('/', '\\');
+    if (root != null) {
+      if (style == Style.windows) {
+        // Normalize the Windows root if needed.
+        final canonRoot = canonicalize ? root.toLowerCase() : root;
+        this.root = canonRoot.replaceAll('/', r'\');
+      } else if (canonicalize && style == Style.url) {
+        // Canonicalize the URL scheme and authority.
+        this.root = root.toLowerCase();
+      }
     }
     removeTrailingSeparators();
   }