mathjax / mathjax-a11y Goto Github PK

More an note to self since I can't test right now.

Does the extension disable line-breaking?

Mathml already containing mactions can lead to strange behaviour and potentially incorrect speech output.

We don't handle them yet, right?

My gut feeling would be that they will never get collapsed completely (but might in themselves if they're huge). As in

<...> (<...>)

etc.

More notes from previous F2Fs.

We want to figure out how we can align the new tools (enrichment, collapse, speech/walker) with MathJax-node.

• pre-generate enriched MathML
• pre-generate maction-ified MathML (+ minimal extension for collapsing)
• pre-generate full MathJax-node output (+ minimal extensions for collapsing)
  • enable the walker to use the data attributes in the output.

Via stevefaulkner on a11y slackers. the textarea needs associated label <label for="input">Enter TeX code here:</label>

Sigh.

I would like to propose to move the dist and build files back into this repo.

Since we're not putting the final extension in the third-party repo, we might as well push them to the CDN from this repo; the other one just adds noise and confusion.

NVDA reads out single letters.

They are somewhat exclusive and the results can be confusing. But somehow the configuration I added to the walker does not actually deactivate it.

We sometimes call it walker, sometimes explorer. Should these be identical? If so, we should decide on a unique name.

Looks like the old xpath issues are back again. Was anything changed?

Would it be more consistent to name this "none"?

JAWS voices "blank" before each speech string during exploration.

However, it usually stops doing that after a while and start voicing things correctly.

I have not been able to figure out how to trigger this change in behavior. It might be a bug in JAWS or it might be something we can tweak. Let's reach out.

https://mathjax.github.io/MathJax-RespEq/Semantics-Lab/Struik.html gives me processing errors everywhere.

Maybe 2.6 related breakage?

I see this failure at http://codepen.io/pkra/pen/yOWrWQ (with the dist built from current master).

Also, only one expression (the first display-style expression) has the usual Math processing error in the DOM.

lastError gives

MathJax.Hub.lastError
TypeError: Cannot read property 'root' of null
    at Object.AddSpeech (https://beta.mathjax.org/mathjax/contrib/a11y/Assistive-Explore.js?rev=2.6.1:1:5904)
    at Object.AddEvent (https://beta.mathjax.org/mathjax/contrib/a11y/Assistive-Explore.js?rev=2.6.1:1:5825)
    at Object.Register (https://beta.mathjax.org/mathjax/contrib/a11y/Assistive-Explore.js?rev=2.6.1:1:4595)
    at Function.execute (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:3084)
    at cb (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:2661)
    at Execute (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:7293)
    at ExecuteHooks (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:10599)
    at Function.execute (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:3084)
    at cb (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:2661)
    at Execute (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:7293)
    at Post (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:9345)
    at Object.processOutput (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:40062)
    at Function.execute (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:3084)
    at Function.WAITEXECUTE (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:5827)
    at cb (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:2661)
    at Function.WAITEXECUTE (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:6016)
    at cb (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:2661)
    at Execute (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:7293)
    at Object.loadComplete (https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS_HTML-full:19:15890)
    at https://cdn.mathjax.org/mathjax/2.6-latest/jax/output/SVG/fonts/TeX/Main/Regular/BasicLatin.js?rev=2.6.1:19:19138

Switching to a different renderer + refreshing the page resets ok.

It seems SRE now forces HTTPS, in particular even loading it over http leads to SRE loading its json data via https.

This leads to problems on WinXP (which doesn't support recent encryption cyphers).

A simple way to grab a copy of all relevant files for local testing would be enough.

Just a note for a future iteration. When collapsing while exploring, ChromeVox will repeat the outer label before re-entering the explorer and reading the live-region again. This lead to repetition.

To compare, it would help to pull up a separate version with regular / enhanced line-breaking.

As an AT user, I want to read documents with mathematics the same way I read other documents, using my AT's smart modes, e.g., to read the entire document or read it line-by-line.

This experience falls short because, e.g., NVDA will (sensibly) read the aria-label ("math") when reading line-by-line, i.e., "When math holds, we get math which implies math".

While some navigation is availabe to reach back (ctrl+tab to go to previous focusable), it is not always reliable (jumping too far behind). Also, once you're on the focused element, I couldn't figure out if there's a way to get back into the flow.

Some ideas.

• talk to AT community on expected / best-of-class behavior.
• add complete speech text as aria-label (preceded by "math", perhaps, just like headings are preceded by "heading level n")
• consider if we could/should/would get AT to stop at each equation (treating them as newlines, even inline math), and enable smooth UX for activating and continuing(/skipping) the walker.

What else could we do?

Right now, we can't share the demos using https://rawgit.com/mathjax/MathJax-RespEq/master/Semantics-Lab/Semantics-Lab-MML.html since it enforces https.

(@zorkow plans to enable ssl on progressiveaccess.com.)

Some thoughts on what we need for production use

• responsive equations
  • minimal build of SRE + respEq extension
  • pre-processing abilities
    • generate MathML with mactions + minimal JS for collapsing
    • pre-generate SVG/CommonHTML output to a set of breakpoints with media queries
• a11y tool
  • build minimal version for pre-generated enriched MathML, SVG, CommonHTML

Give them names for usability.
(Suggested by Sina.)

I can't explore beyond it (neither horizontally or vertically).

This files needs at least a loadComplete call. Since we'll be pushing it to the CDN's third party folder, we need to decide on a path structure, e.g.,

MathJax.Callback.Queue(
  ["loadComplete",MathJax.Ajax,"[Contrib]/mathjax-a11y/MathJax-accessibility.js"]
);

HammerJS be helpful here.

Hammer.js supports most common single and multi-touch gestures, and is fully extendable to add custom gestures.

I tested Dolphin again.

It reads aria-labels so we have now some support now. However, Dolphin adds a lot of unfortunate noise.

a) it announces all potential click handlers (single/double/right) and the application
b) when navigating passed the label, it will read all characters, so you'd need SVG (but #119)

Nothing for this release but worth thinking about in the next iteration.

Should be spoken once when the first math element gets focus not at start of page as most blind people will stop speech immediately after page load.
(Suggestion by Sina.)

@dpvc
I would have expected the collapsible elements in SVG to have class mjx-svg-maction.
However, they have no classname at all, so hover and flame do not work.

to fix mobile rendering

Here's what I come up with.

• explain the two-in-one extensions or split them up
• document configuration options
• document menu options
• document behavior
  • global aria-label
  • walker
  • collapse
  • interaction between the various parts, dependencies etc.

It seems in documents with more than one equation, the walker is adding an announcement ("nav. math in page, explore with shift+space") for each equation, leading to as many announcements as there are equations in the page -- that's a bit annoying 😉

Using the walker is difficult in everyday documents because the extension does not have good defaults (i.e., dummy walker).

I was wondering about the name. -Extensions (plural) seems a bit more fitting (but might lead to typos). Something like Accessibility-Loader / a11y-loader is more fitting but perhaps not as easy to remember. Not a huge thing but wanted to raise it (esp in the context of #115)

Just to record this from other conversations. It would be good to revive the older demos that only show some features, e.g.,

• just enrichment (right now, the demos add mactions for collapsing), e.g., http://codepen.io/pkra/pen/oXLGZQ
• just collapsing (no speech text or walker)

The sample from our MathUI paper is throwing an error in the Semantics-Lab (but not in regular MathJax).

\begin{align*}
I_\nu(\nu^{-1},1)
&=\underbrace{\frac{\pi^2}{4}\ln\left(\frac{(1+\nu)^{1+\nu}}{\nu^\nu}
\right)-\frac{7\zeta(3)}{8}\nu}_{\text{Let this be
C}}+2\int^\frac{1-\nu}{1+\nu}_1\frac{\chi_3(v)}{(1+v)^2}{\rm d}v\\
&=C-\left.\frac{2\chi_3(v)}{1+v}\right|^\frac{1-\nu}{1+\nu}_1+2\int^\frac{1-\nu}
{1+\nu}_1\frac{\chi_2(v)}{v(1+v)}{\rm d}v\\
&=C+(1-\nu)\chi_3\left(\frac{1-\nu}{1+\nu}\right)-\frac{7\zeta(3)}{8}
-\left.{\color{white}{\frac{1}{1}}}2\chi_2(v)\ln(1+v)\right|^\frac{1-\nu}{1+\nu}
_1+\int^\frac{1-\nu}{1+\nu}_1\frac{\ln(1+v)\ln\left(\frac{1+v}{1-v}\right)}{v}{
\rm d}v\\
&=C+(1-\nu)\chi_3\left(\frac{1-\nu}{1+\nu}\right)-\frac{7\zeta(3)}{8}
+2\chi_2\left(\frac{1-\nu}{1+\nu}\right)\ln\left(\frac{1+\nu}{2}\right)+\frac{
\pi^2}{4}\ln{2}\\
&\ \ \ \
+\frac{1}{2}\int^\frac{1-\nu}{1+\nu}_1\frac{
\ln^2(1+v)-\ln^2(1-v)+\ln^2\left(\frac{1-v}{1+v}\right)}{v}{\rm d}v
\end{align*}

Where is that entry's text stored/generated?

Also, should we harmonize the naming of this option and Semantic-Collapse.js (where I would have assumed this comes from but I don't see it)?

Something I noticed without time to look into. Due to the bug in 2.6.0 (fixed in 2.6.1), we can't use combined configurations while disabling the AssistiveMML extension. Yet when I switched to just inline configurations, I got Math Processing errors. Probably just a timing issue.

If a MathML semantics annotation element (apparently Mathtype adds those) stops rendering of the enriched expression.
Simple example:

<math><semantics><mrow><mi>a</mi><mi>b</mi></mrow><annotation>implicit times</annotation></semantics></math>

Punctuated with role endpunct or startpunct should not be collapsed.
These are expressions that contain a single punctuation at start or end, only. In most maths expressions this is irrelevant.
For example:
a = b.
will lead to
<.>
Instead it should lead to
<=>.

Just to track this in an issue. The only big regression after switching to role=application.

Walking up sometimes jumps two levels.

Example: summation formula, navigating into the sigma (go all the way down, then one up will take you up two levels)

Currently, all extensions skip inline math. That makes sense for responsive rendering (though I might argue for that there, too). But for the a11y walker we need to enable inline math elements.

When entering a formula including something similar to:
( x + 1 ) ^2 =
The exponent is applied to the last term within the brackets, rather than the brackets as a hole.

(I've attached a screenshot to demonstrate)

We should add links to GitHub (e.g., https://github.com/mathjax/MathJax-RespEq/wiki/Support-Matrix-a11y-tool) to the demos.

We need to somehow alert the user that the tool is available.

Maybe role=application helps (though @zorkow had problems with it in the past). Otheriwse, we could add a label. I'm thinking it might be something equivalent to role=math, i.e., something that will make AT indicate there's some math.

If an expression is already collapsed on load the aria label reads the full expression.
Should either contain the expression corresponding to the visible expression or the shortest possible one for faster page reading.

data-semantic-complexity can appear twice in MathML element. That leads to an XML error at my end.
I believe it only happens for mactions on the top level.
Example expression:

i = 1, 2, \ldots, n+1

• NVDA consistently loses first speech text
  • could be fixed by adding a 100ms delay, cf. #36
• NVDA consistently loses speech text updates.
  • This is somewhat more erratic. It only happens in certain situations (and always when entering a lower level of an expression) but it is consistent, e.g.,
  • in the quadratic equation, x is never voiced when entering the first level (but exploring left, then right gets x voiced). Similarly, -b is never voiced when exploring the fraction's numerator.

Since no other AT has these problems, maybe it's a bug on NVDA's side. Let's reach out.

Something is wrong with the transferring of complexity value to the DOM elements created.

http://beta.mathjax.org/mathjax/develop/unpacked/MathJax.js

is a copy of today's develop branch. Let's switch the demos over please. (I keep running into rawgit issues).

Reminder to update gh-pages