Improve comparison performance with pre-computed priority

Closes: #151
Resolves: #148

Author: halostatue

CHANGELOG.md

@@ -1,29 +1,31 @@
 # Changelog
 
-## NEXT / YYYY-MM-DD
-
-- 1 deprecation:
-
-  - Deprecated `MIME::Type#priority_compare`. In a future release, this will be
-    will be renamed to `MIME::Type#<=>`. This method is used in tight loops, so
-    there is no warning message for either `MIME::Type#priority_compare` or
-    `MIME::Type#<=>`.
-
-- 1 enhancement:
-
-  - Improved the performance of sorting by eliminating the complex comparison
-    flow from `MIME::Type#priority_compare`. The old version shows under 600
-    i/s, and the new version shows over 900 i/s. In sorting the full set of MIME
-    data, there are three differences between the old and new versions; after
-    comparison, these differences are considered acceptable.
-
-- 1 bug fix:
-
-  - Simplified the default compare implementation (`MIME::Type#<=>`) to use the
-    new `MIME::Type#priority_compare` operation and simplify the fallback to
-    `String` comparison. This _may_ result in exceptions where there had been
-    none, as explicit support for several special values (which should have
-    caused errors in any case) have been removed.
+## 3.7.0.pre2 / YYYY-MM-DD
+
+- Deprecated `MIME::Type#priority_compare`. In a future release, this will be
+  will be renamed to `MIME::Type#<=>`. This method is used in tight loops, so
+  there is no warning message for either `MIME::Type#priority_compare` or
+  `MIME::Type#<=>`.
+
+- Improved the performance of sorting by eliminating the complex comparison flow
+  from `MIME::Type#priority_compare`. The old version shows under 600 i/s, and
+  the new version shows over 900 i/s. In sorting the full set of MIME data,
+  there are three differences between the old and new versions; after
+  comparison, these differences are considered acceptable.
+
+- Simplified the default compare implementation (`MIME::Type#<=>`) to use the
+  new `MIME::Type#priority_compare` operation and simplify the fallback to
+  `String` comparison. This _may_ result in exceptions where there had been
+  none, as explicit support for several special values (which should have caused
+  errors in any case) have been removed.
+
+- When sorting the result of `MIME::Types#type_for`, provided a priority boost
+  if one of the target extensions is the type's preferred extension. This means
+  that for the case in [#148][issue-148], when getting the type for `foo.webm`,
+  the type `video/webm` will be returned before the type `audio/webm`, because
+  `.webm` is the preferred extension for `video/webm` but not `audio/webm`
+  (which has a preferred extension of `.weba`). Added tests to ensure MIME types
+  are retrieved in a stable order (which is alphabetical).
 
 ## 3.6.2 / 2025-03-25
 
@@ -375,6 +377,7 @@ there are some validation changes and updated code with formatting.
 [issue-127]: https://github.com/mime-types/ruby-mime-types/issues/127
 [issue-134]: https://github.com/mime-types/ruby-mime-types/issues/134
 [issue-136]: https://github.com/mime-types/ruby-mime-types/issues/136
+[issue-148]: https://github.com/mime-types/ruby-mime-types/issues/148
 [issue-166]: https://github.com/mime-types/ruby-mime-types/issues/166
 [issue-177]: https://github.com/mime-types/ruby-mime-types/issues/177
 [mime-types-data]: https://github.com/mime-types/mime-types-data

Rakefile

@@ -2,6 +2,7 @@ require "rubygems"
 require "hoe"
 require "rake/clean"
 require "minitest"
+require "minitest/test_task"
 
 Hoe.plugin :halostatue
 Hoe.plugin :rubygems
@@ -10,6 +11,7 @@ Hoe.plugins.delete :debug
 Hoe.plugins.delete :newb
 Hoe.plugins.delete :publish
 Hoe.plugins.delete :signing
+Hoe.plugins.delete :test
 
 spec = Hoe.spec "mime-types" do
   developer("Austin Ziegler", "[email protected]")
@@ -24,7 +26,7 @@ spec = Hoe.spec "mime-types" do
     val.merge!({"rubygems_mfa_required" => "true"})
   }
 
-  extra_deps << ["mime-types-data", "~> 3.2015"]
+  extra_deps << ["mime-types-data", "~> 3.2025", ">= 3.2025.0506.pre2"]
   extra_deps << ["logger", ">= 0"]
 
   extra_dev_deps << ["hoe", "~> 4.0"]
@@ -65,6 +67,8 @@ Minitest::TestTask.create :coverage do |t|
   RUBY
 end
 
+task default: :test
+
 namespace :benchmark do
   task :support do
     %w[lib support].each { |path|
@@ -174,6 +178,11 @@ namespace :convert do
   task docs: "convert:docs:run"
 end
 
+task :version do
+  require "mime/types/version"
+  puts MIME::Types::VERSION
+end
+
 namespace :deps do
   task :top, [:number] => "benchmark:support" do |_, args|
     require "deps"

lib/mime/type.rb

@@ -133,7 +133,8 @@ def to_s
   def initialize(content_type) # :yields: self
     @friendly = {}
     @obsolete = @registered = @provisional = false
-    @preferred_extension = @docs = @use_instead = nil
+    @preferred_extension = @docs = @use_instead = @__sort_priority = nil
+
     self.extensions = []
 
     case content_type
@@ -164,6 +165,8 @@ def initialize(content_type) # :yields: self
     self.xrefs ||= {}
 
     yield self if block_given?
+
+    update_sort_priority
   end
 
   # Indicates that a MIME type is like another type. This differs from
@@ -182,60 +185,54 @@ def like?(other)
   # simplified type (the simplified type will be used if comparing against
   # something that can be treated as a String with #to_s). In comparisons, this
   # is done against the lowercase version of the MIME::Type.
+  #
+  # Note that this implementation of #<=> is deprecated and will be changed
+  # in the next major version to be the same as #priority_compare.
+  #
+  # Note that MIME::Types no longer compare against nil.
   def <=>(other)
-    if other.nil?
-      -1
-    elsif other.respond_to?(:simplified)
+    return priority_compare(other) if other.is_a?(MIME::Type)
+    simplified <=> other
+  end
+
+  # Compares the +other+ MIME::Type using a pre-computed sort priority value,
+  # then the simplified representation for an alphabetical sort.
+  #
+  # For the next major version of MIME::Types, this method will become #<=> and
+  # #priority_compare will be removed.
+  def priority_compare(other)
+    if (cmp = __sort_priority <=> other.__sort_priority) == 0
       simplified <=> other.simplified
     else
-      filtered = "silent" if other == :silent
-      filtered ||= "true" if other == true
-      filtered ||= other.to_s
-
-      simplified <=> MIME::Type.simplified(filtered)
+      cmp
     end
   end
 
-  # Compares the +other+ MIME::Type based on how reliable it is before doing a
-  # normal <=> comparison. Used by MIME::Types#[] to sort types. The
-  # comparisons involved are:
-  #
-  # 1. self.simplified <=> other.simplified (ensures that we
-  #    do not try to compare different types)
-  # 2. IANA-registered definitions < other definitions.
-  # 3. Complete definitions < incomplete definitions.
-  # 4. Current definitions < obsolete definitions.
-  # 5. Obselete with use-instead names < obsolete without.
-  # 6. Obsolete use-instead definitions are compared.
+  # Uses a modified pre-computed sort priority value based on whether one of the provided
+  # extensions is the preferred extension for a type.
   #
-  # While this method is public, its use is strongly discouraged by consumers
-  # of mime-types. In mime-types 3, this method is likely to see substantial
-  # revision and simplification to ensure current registered content types sort
-  # before unregistered or obsolete content types.
-  def priority_compare(other)
-    pc = simplified <=> other.simplified
-    if pc.zero? || !(extensions & other.extensions).empty?
-      pc =
-        if (reg = registered?) != other.registered?
-          reg ? -1 : 1 # registered < unregistered
-        elsif (comp = complete?) != other.complete?
-          comp ? -1 : 1 # complete < incomplete
-        elsif (obs = obsolete?) != other.obsolete?
-          obs ? 1 : -1 # current < obsolete
-        elsif obs && ((ui = use_instead) != (oui = other.use_instead))
-          if ui.nil?
-            1
-          elsif oui.nil?
-            -1
-          else
-            ui <=> oui
-          end
-        else
-          0
-        end
+  # This is an internal function. If an extension provided is a preferred extension either
+  # for this instance or the compared instance, the corresponding extension has its top
+  # _extension_ bit cleared from its sort priority. That means that a type with between
+  # 0 and 8 extensions will be treated as if it had 9 extensions.
+  def __extension_priority_compare(other, exts) # :nodoc:
+    tsp = __sort_priority
+
+    if exts.include?(preferred_extension) && tsp & 0b1000 != 0
+      tsp = tsp & 0b11110111 | 0b0111
+    end
+
+    osp = other.__sort_priority
+
+    if exts.include?(other.preferred_extension) && osp & 0b1000 != 0
+      osp = osp & 0b11110111 | 0b0111
     end
 
-    pc
+    if (cmp = tsp <=> osp) == 0
+      simplified <=> other.simplified
+    else
+      cmp
+    end
   end
 
   # Returns +true+ if the +other+ object is a MIME::Type and the content types
@@ -270,6 +267,13 @@ def hash
     simplified.hash
   end
 
+  # The computed sort priority value. This is _not_ intended to be used by most
+  # callers.
+  def __sort_priority # :nodoc:
+    update_sort_priority if !instance_variable_defined?(:@__sort_priority) || @__sort_priority.nil?
+    @__sort_priority
+  end
+
   # Returns the whole MIME content-type string.
   #
   # The content type is a presentation value from the MIME type registry and
@@ -324,6 +328,7 @@ def extensions
 
   ##
   def extensions=(value) # :nodoc:
+    clear_sort_priority
     @extensions = Set[*Array(value).flatten.compact].freeze
     MIME::Types.send(:reindex_extensions, self)
   end
@@ -350,9 +355,7 @@ def preferred_extension
 
   ##
   def preferred_extension=(value) # :nodoc:
-    if value
-      add_extensions(value)
-    end
+    add_extensions(value) if value
     @preferred_extension = value
   end
 
@@ -405,9 +408,17 @@ def use_instead
   attr_writer :use_instead
 
   # Returns +true+ if the media type is obsolete.
-  attr_accessor :obsolete
+  #
+  # :attr_accessor: obsolete
+  attr_reader :obsolete
   alias_method :obsolete?, :obsolete
 
+  ##
+  def obsolete=(value)
+    clear_sort_priority
+    @obsolete = !!value
+  end
+
   # The documentation for this MIME::Type.
   attr_accessor :docs
 
@@ -465,11 +476,27 @@ def xref_urls
   end
 
   # Indicates whether the MIME type has been registered with IANA.
-  attr_accessor :registered
+  #
+  # :attr_accessor: registered
+  attr_reader :registered
   alias_method :registered?, :registered
 
+  ##
+  def registered=(value)
+    clear_sort_priority
+    @registered = !!value
+  end
+
   # Indicates whether the MIME type's registration with IANA is provisional.
-  attr_accessor :provisional
+  #
+  # :attr_accessor: provisional
+  attr_reader :provisional
+
+  ##
+  def provisional=(value)
+    clear_sort_priority
+    @provisional = !!value
+  end
 
   # Indicates whether the MIME type's registration with IANA is provisional.
   def provisional?
@@ -552,6 +579,7 @@ def encode_with(coder)
     coder["registered"] = registered?
     coder["provisional"] = provisional? if provisional?
     coder["signature"] = signature? if signature?
+    coder["sort-priority"] = __sort_priority || 0b11111111
     coder
   end
 
@@ -560,6 +588,7 @@ def encode_with(coder)
   #
   # This method should be considered a private implementation detail.
   def init_with(coder)
+    @__sort_priority = 0
     self.content_type = coder["content-type"]
     self.docs = coder["docs"] || ""
     self.encoding = coder["encoding"]
@@ -573,6 +602,8 @@ def init_with(coder)
     self.use_instead = coder["use-instead"]
 
     friendly(coder["friendly"] || {})
+
+    update_sort_priority
   end
 
   def inspect # :nodoc:
@@ -628,6 +659,37 @@ def simplify_matchdata(matchdata, remove_x = false, joiner: "/")
 
   private
 
+  def clear_sort_priority
+    @__sort_priority = nil
+  end
+
+  # Update the __sort_priority value. Lower numbers sort better, so the
+  # bitmapping may seem a little odd. The _best_ sort priority is 0.
+  #
+  # | bit | meaning         | details   |
+  # | --- | --------------- | --------- |
+  # | 7   | obsolete        | 1 if true |
+  # | 6   | provisional     | 1 if true |
+  # | 5   | registered      | 0 if true |
+  # | 4   | complete        | 0 if true |
+  # | 3   | # of extensions | see below |
+  # | 2   | # of extensions | see below |
+  # | 1   | # of extensions | see below |
+  # | 0   | # of extensions | see below |
+  #
+  # The # of extensions is marked as the number of extensions subtracted from
+  # 16, to a minimum of 0.
+  def update_sort_priority
+    extension_count = @extensions.length
+    obsolete = (instance_variable_defined?(:@obsolete) && @obsolete) ? 1 << 7 : 0
+    provisional = (instance_variable_defined?(:@provisional) && @provisional) ? 1 << 6 : 0
+    registered = (instance_variable_defined?(:@registered) && @registered) ? 0 : 1 << 5
+    complete = extension_count.nonzero? ? 0 : 1 << 4
+    extension_count = [0, 16 - extension_count].max
+
+    @__sort_priority = obsolete | registered | provisional | complete | extension_count
+  end
+
   def content_type=(type_string)
     match = MEDIA_TYPE_RE.match(type_string)
     fail InvalidContentType, type_string if match.nil?