Fix and catch speech computation errors, add missing braille option

[zorkow]

PR contains some corrections.

• Builds the correct speech string from the SSML also for the mjx-container element's aria-label.
• No speech is computed when enableEnrichment is set to false. Also catches error on speech computation when enableEnrichment is true but the element is actually not enriched. I think this currently only happens in the lab or when an enrichment error is caught.
• Adds missing enableBraille option.

[dpvc]

There are a couple of items identified here for your consideration.

Also, it looks like the function

MathJax-src/ts/a11y/semantic-enrich.ts

Lines 290 to 306 in 799ad89

	protected getSpeech(node: MmlNode): string {
	const attributes = node.attributes;
	if (!attributes) return '';
	const speech = attributes.getExplicit('data-semantic-speech') as string;
	// TODO (explorer) For tree role move all speech etc. to container
	// element.
	if (!attributes.hasExplicit('data-semantic-parent') && speech) {
	return speech;
	}
	for (let child of node.childNodes) {
	let value = this.getSpeech(child);
	if (value) {
	return value;
	}
	}
	return '';
	}

is no longer used, so can be eliminated.

Also, I'm wondering if

if (this.options.enableSpeech || this.options.enableBraille) {
...
}

should be placed around lines

MathJax-src/ts/a11y/semantic-enrich.ts

Lines 417 to 419 in 799ad89

	for (const math of this.math) {
	(math as EnrichedMathItem<N, T, D>).attachSpeech(this);
	}

like is done at line 430 below. That would save having to loop through all the math when we aren't adding speech or braille.

[dpvc]

Do any of the calls below cause a MathJax retry error? (That is, do any call mathjax.retryAfter()?)

If not, then you can do this.state(STATE.ATTACHSPEECH) above line 247, and then just return here and the other places where the state is set. That would avoid having to set the state in several places. But it only works if the rest of the code will always complete and not restart the typesetting via retryAfter().

[zorkow]

It should not. The only place that triggers a retryAfter is in line 183:

MathJax-src/ts/a11y/semantic-enrich.ts

Line 183 in 799ad89

this.generatorPool.init(document.options, document.adaptor);

But that should have run before attachSpeech at all times.

[dpvc]

I didn't catch this earlier, but I think this condition is wrong. That is, if document.options.enableSpeech is false, you should not generate speech even if !speech is true, and similarly for braille. So I think this should be

if ((!speech && document.options.enableSpeech) || (!bralle && document.options.enableBraille)) {

[zorkow]

Thanks. Good catch.

[zorkow]

I've made changes except the attachSpeech one and I've made a change in package.json to the new SRE release.

[dpvc]

OK, this all looks good. If you want to do the STATE.ATTACHSPEECH change, that is great. Approved either way.

[zorkow]

OK, this all looks good. If you want to do the STATE.ATTACHSPEECH change, that is great. Approved either way.

I've tested with the states and I've made the simplification.