Do not build regexes at compile-time

Author: josevalim

lib/ex_doc/autolink.ex

@@ -163,12 +163,12 @@ defmodule ExDoc.Autolink do
     end
   end
 
-  @ref_regex ~r/^`(.+)`$/
+  defp ref_regex, do: ~r/^`(.+)`$/
 
   def custom_link(attrs, config) do
     case Keyword.fetch(attrs, :href) do
       {:ok, href} ->
-        case Regex.scan(@ref_regex, href) do
+        case Regex.scan(ref_regex(), href) do
           [[_, custom_link]] ->
             custom_link
             |> url(:custom_link, config)
@@ -214,7 +214,7 @@ defmodule ExDoc.Autolink do
 
   defp build_extra_link(link, config) do
     with %{scheme: nil, host: nil, path: path} = uri <- URI.parse(link),
-         true <- is_binary(path) and path != "" and not (path =~ @ref_regex),
+         true <- is_binary(path) and path != "" and not (path =~ ref_regex()),
          true <- Path.extname(path) in @builtin_ext do
       if file = config.extras[Path.basename(path)] do
         append_fragment(file <> config.ext, uri.fragment)

lib/ex_doc/formatter/html/templates.ex

@@ -182,9 +182,8 @@ defmodule ExDoc.Formatter.HTML.Templates do
   end
 
   # TODO: split into sections in Formatter.HTML instead (possibly via DocAST)
-  @h2_regex ~r/<h2.*?>(.*?)<\/h2>/m
   defp extract_headers(content) do
-    @h2_regex
+    ~r/<h2.*?>(.*?)<\/h2>/m
     |> Regex.scan(content, capture: :all_but_first)
     |> List.flatten()
     |> Enum.filter(&(&1 != ""))
@@ -222,13 +221,12 @@ defmodule ExDoc.Formatter.HTML.Templates do
 
   We only link `h2` and `h3` headers. This is kept consistent in ExDoc.SearchData.
   """
-  @heading_regex ~r/<(h[23]).*?>(.*?)<\/\1>/m
   @spec link_headings(String.t() | nil, String.t()) :: String.t() | nil
   def link_headings(content, prefix \\ "")
   def link_headings(nil, _), do: nil
 
   def link_headings(content, prefix) do
-    @heading_regex
+    ~r/<(h[23]).*?>(.*?)<\/\1>/m
     |> Regex.scan(content)
     |> Enum.reduce({content, %{}}, fn [match, tag, title], {content, occurrences} ->
       possible_id = text_to_id(title)
@@ -243,7 +241,6 @@ defmodule ExDoc.Formatter.HTML.Templates do
     |> elem(0)
   end
 
-  @class_regex ~r/<h[23].*?(\sclass="(?<class>[^"]+)")?.*?>/
   @class_separator " "
   defp link_heading(match, _tag, _title, "", _prefix), do: match
 
@@ -273,7 +270,7 @@ defmodule ExDoc.Formatter.HTML.Templates do
     # it was setting `#{section_header_class_name}` as the only CSS class
     # associated with the given header.
     class_attribute =
-      case Regex.named_captures(@class_regex, match) do
+      case Regex.named_captures(~r/<h[23].*?(\sclass="(?<class>[^"]+)")?.*?>/, match) do
         %{"class" => ""} ->
           section_header_class_name
 

lib/ex_doc/utils.ex

@@ -85,13 +85,13 @@ defmodule ExDoc.Utils do
     end)
   end
 
-  @clean_html_regex ~r/<\/?\s*[a-zA-Z]+(?:[^>=]|='[^']*'|="[^"]*"|=[^'"][^\s>]*)*>/
 
   @doc """
   Strips HTML tags from text leaving their text content
   """
   def strip_tags(text, replace_with \\ "") when is_binary(text) do
-    String.replace(text, @clean_html_regex, replace_with)
+    clean_html_regex = ~r/<\/?\s*[a-zA-Z]+(?:[^>=]|='[^']*'|="[^"]*"|=[^'"][^\s>]*)*>/
+    String.replace(text, clean_html_regex, replace_with)
   end
 
   @doc """