BREAKING CHANGE: Drop grammars

[marcoscaceres]

grammars are underspecified - they lack a format, so they are kinda useless... so removes grammars attribute, drops SpeechGrammarList, and SpeechGrammar.

Closes #32
Closes #57

---

Preview | Diff

[saschanaz]

What's the implementation status on Chrome? What does it support?

[foolip]

I'd like to hold off on removing this from the spec until I understand a bit better why this was added and if it ever worked, or perhaps still works on some platform.

[kdavis-mozilla]

@foolip I'd speculate that when the spec was initially written, many years ago now, the production STT engines were not able to handle LVCSR and required grammars to get reasonable results. Things have changed somewhat now.

[marcoscaceres]

Noting also that src attribute has implied fetching behavior and semantics, yet the spec doesn't actually specify how things are fetched, parsed, error handling (e.g., what happens if one fails, the rest download, is order significant, and so on), referrer policy, etc., etc. so, there is no way anyone implemented any of this in an interoperable manner.

Although well intentioned, fetching such grammars should really be done with fetch() or even good ol' xhr if dealing with XML.

[foolip]

@marcoscaceres there's no argument that the spec defines this well enough for interoperable implementation, and doing nothing wouldn't resolve #57. I've poked some more in #57 (comment) and my only concern with dropping this is what then happens with the implementation in Chromium, since I can't guess yet what the impact of removing it would be. Would you be OK with waiting for usage numbers before we axe it from the spec, to avoid flip-flopping? I'd of course not expect anyone to try implementing it in the interim, and warnings to that effect would be fine.

[marcoscaceres]

Yes, absolutely: no objection and no rush :)

[maccman]

Any update? Does Chrome support SpeechGrammarList?

[evanbliu]

The WebSpeech API implementation in Chrome supports SpeechGrammarList. However, the backend speech recognition service does not so they don't actually do anything. I'm in favor of dropping this from the spec.

[beaufortfrancois]

With #117 now being merged, I believe we can close this PR.