Add 'E' modifier to Regex for :export option (#14907)

Author: sabiwara

lib/elixir/lib/inspect.ex

@@ -569,6 +569,7 @@ defimpl Inspect, for: Regex do
   defp translate_options([:firstline | t], acc), do: translate_options(t, [?f | acc])
   defp translate_options([:ungreedy | t], acc), do: translate_options(t, [?U | acc])
   defp translate_options([:multiline | t], acc), do: translate_options(t, [?m | acc])
+  defp translate_options([:export | t], acc), do: translate_options(t, [?E | acc])
   defp translate_options([], acc), do: acc
   defp translate_options(_t, _acc), do: :error
 

lib/elixir/lib/regex.ex

@@ -101,6 +101,13 @@ defmodule Regex do
     * `:ungreedy` (U) - inverts the "greediness" of the regexp
       (the previous `r` option is deprecated in favor of `U`)
 
+    * `:export` (E) (since Elixir 1.20) - uses an exported pattern
+      which can be shared across nodes or through config, at the cost of a runtime
+      overhead every time to re-import it every time it is executed.
+      This modifier only has an effect starting on Erlang/OTP 28, and it is ignored
+      on older versions (i.e. `~r/foo/E == ~r/foo/`). This is because patterns cannot
+      and do not need to be exported in order to be shared in these versions.
+
   ## Captures
 
   Many functions in this module handle what to capture in a regex
@@ -515,7 +522,7 @@ defmodule Regex do
   """
   @spec names(t) :: [String.t()]
   def names(%Regex{re_pattern: re_pattern}) do
-    {:namelist, names} = :re.inspect(re_pattern, :namelist)
+    {:namelist, names} = :re.inspect(maybe_import_pattern(re_pattern), :namelist)
     names
   end
 
@@ -585,10 +592,16 @@ defmodule Regex do
       %Regex{source: source, opts: compile_opts} = regex
       :re.run(string, source, compile_opts ++ options)
     else
-      _ -> :re.run(string, re_pattern, options)
+      _ -> :re.run(string, maybe_import_pattern(re_pattern), options)
     end
   end
 
+  @compile {:inline, maybe_import_pattern: 1}
+  defp maybe_import_pattern({:re_exported_pattern, _, _, _, _} = exported),
+    do: :re.import(exported)
+
+  defp maybe_import_pattern(pattern), do: pattern
+
   @typedoc """
   Options for regex functions that capture matches.
   """
@@ -1007,6 +1020,16 @@ defmodule Regex do
     translate_options(t, [:ungreedy | acc])
   end
 
+  defp translate_options(<<?E, t::binary>>, acc) do
+    # on OTP 27-, the E modifier is a no-op since the feature doesn't exist but isn't needed
+    # (regexes aren't using references and can be shared across nodes or stored in config)
+    # TODO: remove this check on Erlang/OTP 28+ and update docs
+    case Code.ensure_loaded?(:re) and function_exported?(:re, :import, 1) do
+      true -> translate_options(t, [:export | acc])
+      false -> translate_options(t, acc)
+    end
+  end
+
   defp translate_options(<<>>, acc), do: acc
   defp translate_options(t, _acc), do: {:error, t}
 
@@ -1022,6 +1045,9 @@ defmodule Regex do
         :erlang.system_info(:otp_release) < [?2, ?8] ->
           Macro.escape(regex.re_pattern)
 
+        :lists.member(:export, regex.opts) ->
+          Macro.escape(regex.re_pattern)
+
         # OTP 28.1+ introduced the ability to export and import regexes from compiled binaries
         Code.ensure_loaded?(:re) and function_exported?(:re, :import, 1) ->
           {:ok, exported} = :re.compile(regex.source, [:export] ++ regex.opts)

lib/elixir/test/elixir/inspect_test.exs

@@ -906,6 +906,11 @@ defmodule Inspect.OthersTest do
     assert inspect(Regex.compile!("foo", [:ucp])) == ~S'Regex.compile!("foo", [:ucp])'
   end
 
+  @tag :re_import
+  test "exported regex" do
+    assert inspect(~r/foo/E) == "~r/foo/E"
+  end
+
   test "inspect_fun" do
     fun = fn
       integer, _opts when is_integer(integer) ->

lib/elixir/test/elixir/regex_test.exs

@@ -52,6 +52,20 @@ defmodule RegexTest do
     assert Regex.match?(~r/^b$/m, "a\nb\nc")
   end
 
+  @tag :re_import
+  test "export" do
+    # exported patterns have no structs, so these are structurally equal
+    assert ~r/foo/E == Regex.compile!("foo", [:export])
+
+    assert Regex.match?(~r/foo/E, "foo")
+    refute Regex.match?(~r/foo/E, "Foo")
+
+    assert Regex.run(~r/c(d)/E, "abcd") == ["cd", "d"]
+    assert Regex.run(~r/e/E, "abcd") == nil
+
+    assert Regex.names(~r/(?<FOO>foo)/E) == ["FOO"]
+  end
+
   test "precedence" do
     assert {"aa", :unknown} |> elem(0) =~ ~r/(a)\1/
   end

lib/mix/lib/mix/tasks/compile.app.ex

@@ -221,6 +221,14 @@ defmodule Mix.Tasks.Compile.App do
     end
   end
 
+  defp to_erl_term(%Regex{re_pattern: {:re_pattern, _, _, _, ref}} = regex)
+       when is_reference(ref) do
+    Mix.raise("""
+    \"def application\" has a term which cannot be written to .app files: #{inspect(regex)}.
+    Use the E modifier to store regexes in application config.
+    """)
+  end
+
   defp to_erl_term(map) when is_map(map) do
     inner =
       Enum.map_intersperse(

lib/mix/test/mix/tasks/compile.app_test.exs

@@ -263,6 +263,33 @@ defmodule Mix.Tasks.Compile.AppTest do
     end)
   end
 
+  @tag :re_import
+  test "accepts only regexes without a reference" do
+    in_fixture("no_mixfile", fn ->
+      Mix.Project.push(CustomProject)
+
+      Process.put(:application, env: [regex: ~r/foo/])
+
+      message = """
+      "def application" has a term which cannot be written to \.app files: ~r\/foo\/.
+      Use the E modifier to store regexes in application config.
+      """
+
+      assert_raise Mix.Error, message, fn ->
+        Mix.Tasks.Compile.App.run([])
+      end
+
+      Process.put(:application, env: [exported: ~r/foo/E])
+
+      Mix.Tasks.Compile.Elixir.run([])
+      Mix.Tasks.Compile.App.run([])
+
+      properties = parse_resource_file(:custom_project)
+
+      assert properties[:env] == [exported: ~r/foo/E]
+    end)
+  end
+
   test ".app contains description and registered (as required by systools)" do
     in_fixture("no_mixfile", fn ->
       Mix.Project.push(MixTest.Case.Sample)

lib/mix/test/test_helper.exs

@@ -43,12 +43,22 @@ cover_exclude =
     []
   end
 
+# OTP 28.1+
+re_import_exclude =
+  if Code.ensure_loaded?(:re) and function_exported?(:re, :import, 1) do
+    []
+  else
+    [:re_import]
+  end
+
 Code.require_file("../../elixir/scripts/cover_record.exs", __DIR__)
 CoverageRecorder.maybe_record("mix")
 
 ExUnit.start(
   trace: !!System.get_env("TRACE"),
-  exclude: epmd_exclude ++ os_exclude ++ git_exclude ++ line_exclude ++ cover_exclude,
+  exclude:
+    epmd_exclude ++
+      os_exclude ++ git_exclude ++ line_exclude ++ cover_exclude ++ re_import_exclude,
   include: line_include,
   assert_receive_timeout: String.to_integer(System.get_env("ELIXIR_ASSERT_TIMEOUT", "300"))
 )