Previously, on A Few Hours of WebVTT with Dale…

In my last blog post, I put forth the question of whether voice tags could be nested. The JS Validator complained when one would nest voice tags and I was curious as to why. I couldn’t find anything in the spec to suggest that nested voice tags were not legal in the syntax. Rillian suggested I check out WebKit’s WebVTT parser to see how they handle nested voice cues, and that is what I did.

ProTip: If you want to search for a file in a GitHub repo, hit the ‘t’ key on the repo’s page to get a fuzzy file finder. GitHub, I love thee! After some searching, I believe I found the code in WebKit that handles tags, in WebVTTParser.cpp:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

case WebVTTTokenTypes::StartTag: {
    RefPtr<HTMLElement> child;
    if (isRecognizedTag(tokenTagName))
        child = HTMLElement::create(tagName, document);
    else if (m_token.name().size() == 1 && m_token.name()[0] == 'c')
        child = HTMLElement::create(spanTag, document);
    else if (m_token.name().size() == 1 && m_token.name()[0] == 'v')
        child = HTMLElement::create(qTag, document);

    if (child) {
        if (m_token.classes().size() > 0)
            child->setAttribute(classAttr, AtomicString(m_token.classes().data(), m_token.classes().size()));
        if (child->hasTagName(qTag))
            child->setAttribute(titleAttr, AtomicString(m_token.annotation().data(), m_token.annotation().size()));
        m_currentNode->parserAppendChild(child);
        m_currentNode = child;
    }
    break;
}

This is how I understand the code: First, we check if the tag is a recognized tag name. That method, defined in WebVTTParser.h, looks like this:

static inline bool isRecognizedTag(const AtomicString& tagName)
{
    return tagName == iTag
        || tagName == bTag
        || tagName == uTag
        || tagName == rubyTag
        || tagName == rtTag;
}

None of these tags seem like a voice tag. The parser moves on! Next, on line 7, there is mention of a token with the name ‘v’. If the token is a v tag, an element is created of type ‘qTag’ and a reference to the element is returned to child. If the element was created, further processing on the element, such as setting attributes, is completed before attaching the objects to the DOM. No special handling of nested tags seems to be present.

I’m not really sure what a ‘qTag’ is. Out of all the WebVTT files in WebKit, this code block seems to have the only mention of a qTag. I really wish GitHub had the ability to grep a repo…

The spec says nothing disallowing nested voice tags. WebKit does nothing special to disallow nested voice tags. Therefore, I guess we are allowing nested voice tags! Jordan contributed the code for this one. Thanks dude!

Links:

The Other Fixes

I just wanted to recap what else I did for release 0.2. I fixed strip-vtt.py so it will smartly separate the metadata from the WebVTT data, regardless of line endings. Thanks for Caitlin for help on fine tuning the regular expression.

Links:

I also fixed the validator not accepting valid escape values for non-breaking whitespace and the right-to-left/left-to-right marker.

Links:

I have to open another pull request moving the rest of the tests from known-good into good, but I’m waiting for the strip-vtt.py fix to land first, which already moves a few tests along with it.

WebVTT Release 0.2 Reflections

The issues found in in known-good have been resolved but there is still the known-bad folder to tackle. Also, I’d like to get in touch with the W3C about clarifying the spec in regards to the legality of nested tags.

When I started this task, I thought blowing through all the bugs in the known-good and known-bad folders would be a piece of cake. Little did I know that each issue takes research, testing, analyzing other people’s code and patience.

I was originally going to work on a JavaScript polyfill for WebVTT once I was done fixing validator bugs but it seeems there are already quite a few of those around. So I need to find a new task to work on. That’s getting ahead of myself though, as there is still more bugs to fix in the validator.