add Recovery.Prefix module

lib/feed_reader/recovery/prefix.ex

@@ -0,0 +1,45 @@
+defmodule FeedReader.Recovery.Prefix do
+  @moduledoc """
+  Recovers the beginning of an XML document before SAX parsing.
+
+  Feeds commonly contain a UTF-8 BOM or whitespace before the XML declaration,
+  while Saxy expects the document to begin at the XML prolog or root element.
+  This module removes those prefixes incrementally and retains incomplete BOM
+  bytes across chunks, so it is safe for arbitrarily small stream chunks.
+  """
+
+  def normalize(input) do
+    Stream.transform(input, fn -> {:leading, <<>>} end, &normalize_chunk/2, &flush/1)
+  end
+
+  defp normalize_chunk(chunk, {:leading, prefix}) do
+    data = prefix <> chunk
+
+    if bom_prefix?(data) do
+      {[], {:leading, data}}
+    else
+      output =
+        if starts_with_bom?(data), do: binary_part(data, 3, byte_size(data) - 3), else: data
+
+      output = trim_whitespace(output)
+      if output == <<>>, do: {[], {:leading, <<>>}}, else: {[output], :ready}
+    end
+  end
+
+  defp normalize_chunk(chunk, :ready), do: {[chunk], :ready}
+
+  defp bom_prefix?(data) when byte_size(data) < 3,
+    do: data == binary_part(<<0xEF, 0xBB, 0xBF>>, 0, byte_size(data))
+
+  defp bom_prefix?(_data), do: false
+  defp starts_with_bom?(<<0xEF, 0xBB, 0xBF, _rest::binary>>), do: true
+  defp starts_with_bom?(_data), do: false
+
+  defp trim_whitespace(<<char, rest::binary>>) when char in [9, 10, 13, 32],
+    do: trim_whitespace(rest)
+
+  defp trim_whitespace(data), do: data
+  defp flush({:leading, <<>>}), do: []
+  defp flush({:leading, prefix}), do: [prefix]
+  defp flush(:ready), do: []
+end

test/feed_reader/recovery/prefix_test.exs

@@ -0,0 +1,28 @@
+defmodule FeedReader.Recovery.PrefixTest do
+  use ExUnit.Case, async: true
+
+  alias FeedReader.Recovery.Prefix
+
+  test "removes a BOM and leading XML whitespace" do
+    assert normalize([<<0xEF, 0xBB, 0xBF, 13, 10, 9, 32>>, "<rss/>"]) == "<rss/>"
+  end
+
+  test "removes a BOM split across chunks" do
+    assert normalize([<<0xEF>>, <<0xBB>>, <<0xBF>>, " \n", "<rss/>"]) == "<rss/>"
+  end
+
+  test "removes leading whitespace split across chunks" do
+    assert normalize(["\r", "\n", "\t", " ", "<rss/>"]) == "<rss/>"
+  end
+
+  test "preserves content that does not have a removable prefix" do
+    assert normalize(["<rss/>", "<tail/>"]) == "<rss/><tail/>"
+  end
+
+  test "returns an empty stream for empty or whitespace-only input" do
+    assert normalize([]) == ""
+    assert normalize([" ", "\n", "\r\t"]) == ""
+  end
+
+  defp normalize(chunks), do: chunks |> Prefix.normalize() |> Enum.join()
+end