relativize_paths silently, incorrectly transforms `gcse:search`

[sirosen]

relativize_paths silently transforms gcse:search

Steps to reproduce

1. Create an html file with Google's GCSE search snippet
   This includes an element for the Google JS to manipulate, <gcse:search></gcse:search>
2. Run nanoc with rules to apply relativize_paths on that html file, with type: html
3. Observe output will read <search></search>

Expected behavior

I think that nanoc should not be transforming that element at all. However, maybe there's some subtlety I'm missing about strict HTML specification.

Either of the below would be also acceptable:

• nanoc should print a warning that the element is transformed -- it can surprise people, after all
• nanoc should crashfail the build on the element

Actual behavior

Instead, nanoc succeeds, but changes the element.

Details

I'm surprised to be the first person reporting this -- maybe not that many people are using GCSE, or using relativize_paths, or maybe I'm doing something else wrong.

[denisdefreyne]

I can confirm this is a problem!

This appears to be problematic behavior with Nokogiri/libxml for parsing HTML… when reassembling the content, the namespaces are somehow left out.

[denisdefreyne]

This appears to be a common problem without a proper solution (see e.g. jekyll/jemoji#43).

Suggestions for workarounds that I’ve found are along the form of

<div id='gcse'></div>
<script>
  $('#gcse').html("<gcse:search></gcse:search>");
</script>

… which is pretty icky.

This might not be fixable within Nanoc itself, unfortunately, but perhaps Nanoc can offer an automatic workaround.

[denisdefreyne]

Icky workaround in #1320.

[sirosen]

Thanks for the really quick detective work on this.
There's been an available patch to fix libxml sitting unattended for a few years now, so this might get fixed someday... but probably not soon.

Given that GCSE is all JS-based, it will never work when blocked by NoScript, etc.
In that light, the workaround you've noted doesn't have any significant practical faults.

This is all a matter of where the grossness exists, not whether or not it exists.

If you decide that you dislike #1320 very strongly, I think it would be satisfactory to detect gcse:search in content prior to applying filters and print a warning with a link:

Warning!
  Nanoc relies on libxml to parse HTML. This is known to cause issues with GCSE.
  Recommended workaround: https://github.com/nanoc/nanoc/issues/1319#issuecomment-367234453

[denisdefreyne]

Sucks to hear about the libxml patch that has been lying around for so long!

My main problem with #1320 is that it only applies to gcse:search, not g:plusone, fb:like, …. The only way to solve this problem in general is to fix the HTML parser, or use a different one.

It might be worth looking into using Oga, for instance.