epub: fix fatal errors while parsing EPUB files

After generating the EPUB file for the Elixir docs with this version,
and reviewing the result with `epubcheck`, I got the following summary:

```console
$ epubcheck doc/elixir/Elixir.epub --json elixir_docs.json                                                                                                    (base)

Check finished with errors
Messages: 0 fatals / 141 errors / 0 warnings / 0 infos
```

If you compare the previous result with what we had on #1851

```
Messages: 9 fatals / 425 errors / 0 warnings / 0 infos
```

you can see that now we don't have messages with `fatal` severity and we
have reduced considerably the number of errors =)

I manually checked the generated EPUB on Apple Books and the previous
truncated sections are solved, I don't see the banner _Below is a
rendering of the page up to the first error_ and also the links to
anchor different anchor seems to work.

Fixes: #1851

Author: milmazz

lib/ex_doc/formatter/epub.ex

@@ -50,6 +50,23 @@ defmodule ExDoc.Formatter.EPUB do
     Path.relative_to_cwd(epub)
   end
 
+  @doc """
+  Helper that replaces anchor names and links that could potentially cause problems on EPUB documents
+
+  This helper replaces all the `&` with `&` found in anchors like
+  `Kernel.xhtml#&&/2` or `<h2 id="&&&/2-examples" class="section-heading">...</h2>`
+
+  These anchor names cause a fatal error while EPUB readers parse the files,
+  resulting in truncated content.
+
+  For more details, see: https://github.com/elixir-lang/ex_doc/issues/1851
+  """
+  def fix_anchors(content) do
+    content
+    |> String.replace(~r{id="&+/\d+[^"]*}, &String.replace(&1, "&", "&amp;"))
+    |> String.replace(~r{href="[^#"]*#&+/\d+[^"]*}, &String.replace(&1, "&", "&amp;"))
+  end
+
   defp normalize_config(config) do
     output =
       config.output
@@ -63,7 +80,11 @@ defmodule ExDoc.Formatter.EPUB do
     for {_title, extras} <- config.extras do
       Enum.each(extras, fn %{id: id, title: title, title_content: title_content, content: content} ->
         output = "#{config.output}/OEBPS/#{id}.xhtml"
-        html = Templates.extra_template(config, title, title_content, content)
+
+        html =
+          config
+          |> Templates.extra_template(title, title_content, content)
+          |> fix_anchors()
 
         if File.regular?(output) do
           ExDoc.Utils.warn("file #{Path.relative_to_cwd(output)} already exists", [])
@@ -157,7 +178,11 @@ defmodule ExDoc.Formatter.EPUB do
   end
 
   defp generate_module_page(module_node, config) do
-    content = Templates.module_page(config, module_node)
+    content =
+      config
+      |> Templates.module_page(module_node)
+      |> fix_anchors()
+
     File.write("#{config.output}/OEBPS/#{module_node.id}.xhtml", content)
   end
 

test/ex_doc/formatter/epub_test.exs

@@ -151,6 +151,9 @@ defmodule ExDoc.Formatter.EPUBTest do
     assert content =~
              ~r{<a href="TypesAndSpecs.Sub.xhtml"><code(\sclass="inline")?>TypesAndSpecs.Sub</code></a>}
 
+    assert content =~
+             ~r{<a href="https://hexdocs.pm/elixir/Kernel.html#&amp;&amp;/2"><code(\sclass="inline")?>&amp;&amp;/2</code></a>}
+
     content = File.read!(tmp_dir <> "/epub/OEBPS/nav.xhtml")
     assert content =~ ~r{<li><a href="readme.xhtml">README</a></li>}
   end
@@ -248,4 +251,19 @@ defmodule ExDoc.Formatter.EPUBTest do
   after
     File.rm_rf!("test/tmp/epub_assets")
   end
+
+  describe "fix_anchors/1" do
+    test "adapts anchor names to avoid parsing errors from EPUB readers" do
+      for {source, expected} <- [
+            {~S|<a href="Kernel.SpecialForms.xhtml#&/1">its documentation</a>|,
+             ~S|<a href="Kernel.SpecialForms.xhtml#&amp;/1">its documentation</a>|},
+            {~S|<a href="Kernel.xhtml#&&/2"><code class="inline">&amp;&amp;/2</code></a>|,
+             ~S|<a href="Kernel.xhtml#&amp;&amp;/2"><code class="inline">&amp;&amp;/2</code></a>|},
+            {~S|<h2 id="&&&/2-examples" class="section-heading">title</h2>|,
+             ~S|<h2 id="&amp;&amp;&amp;/2-examples" class="section-heading">title</h2>|}
+          ] do
+        assert ExDoc.Formatter.EPUB.fix_anchors(source) == expected
+      end
+    end
+  end
 end

test/fixtures/README.md

@@ -15,3 +15,7 @@ hello
 ## more > than
 
 <p><strong>raw content</strong></p>
+
+The following text includes a reference to an anchor that causes problems in EPUB documents.
+
+To remove this anti-pattern, we can replace `&&/2`, `||/2`, and `!/1` by `and/2`, `or/2`, and `not/1` respectively.