jcsteh

@jcsteh@aus.social

Accessibility engineer/tech lead at Mozilla. Previously co-lead developer of the NVDA screen reader. Father, husband, pianist/keyboardist, tech enthusiast, totally blind.

This profile is from a federated server and may be incomplete. Browse more on the original instance.

drew, to random

@jcsteh Pouring out a stiff glass for you tonight sir.

jcsteh,

@drew ???

jcsteh,

@drew Oh, blarg. I actually haven't read them yet. Is this the comments on the post itself?

dangero, to accessibility

question. I'd like to recommend this to developers so they can make their projects more . I noticed there's no documentation. Is it easy enough to learn without the need for documentation? If not, I'm hesitant to recommend something that I know developers are going to struggle with. It's tough enough asking them to make their apps accessible when they don't know much about us, let alone asking them to learn a new library with no docs.

jcsteh,

@dangero CC @matt in case he has thoughts to provide.

KaraLG84, to random
@KaraLG84@dragonscave.space avatar

For shits and giggles I thought I'd see if Be My AI could tell me the layout of a keyboard I was looking at. I asked it if the alt key was directly to the right of the Windows key. It said no, the key directly to the right of the Windows key is the Caps lock key.
So no, it can't.

jcsteh,

@zersiax @KaraLG84 On a related note, it also seems to be very confused about direction and orientation. I've had it say above when something is beside, left when it's right, etc. I'm fascinated by what aspect of its training or inference leads to that problem. I guess if it's trained on images and descriptions, perhaps there aren't enough similar images with different directions and orientations in order for the correct direction to have a higher probability? No idea what I'm talking about just musing really.

menelion, to github
@menelion@dragonscave.space avatar

Hi #blind folks, especially software developers! I'm taking part in a #GitHub research meeting and I hope to raise as many #accessibility points as I can. I was told this fact is not at all confidential, so I may gather feedback.
Here's what I remember: multiline comments are inaccessible (eternal story); sometimes menu roles are used where they shouldn't be (watching repos, reactions etc.). Anything else that really bugs you at GitHub? Thanks!

jcsteh,

@twynn @menelion Thanks for the ping. I can't function efficiently on GitHub without the AxSGrease script I wrote for GitHub. Here is what it does:

  1. Puts the "add line comment" button logically after the line of code instead of before, which makes code reviews a lot faster for screen reader users.
  2. Makes non-comment events into headings; e.g. closing/referencing an issue, approving/requesting changes to a PR, merging a PR.
  3. Turns issue listing tables into tables so you can navigate between issues by row, rather than having to skip through a bunch of stuff for each issue.
    Something i haven't added yet (but really need to because it drives me absolutely wild) is removing the headings on every single file! in the folders/files table. It completely clutters heading navigation for no good reason.
    Finally, blame is completely inaccessible for screen reader users right now, which is really problematic.
KaraLG84, to random
@KaraLG84@dragonscave.space avatar

@NVAccess What's the current way of legally getting a copy of Eloquence these days? I assume the only option is Code Factory's SAPI 5 version?

jcsteh,

@NVAccess @KaraLG84 Did I miss something? Did they abandon the NVDA add-on?

jcsteh,

@NVAccess @KaraLG84 <sigh> It's very hard to push for people to do things legally when there aren't any legal options and it's nearly impossible for anyone other than a big company to feasibly license it.

jcsteh,

@NVAccess @KaraLG84 Which must have been especially difficult given that by all reports, the CodeFactory add-on wasn't particularly polished or well maintained. My tolerance for half-arsed software development (or anything half-arsed that negatively impacts people really) is... fairly low.

jcsteh, to random

As I understand it, with all current LLMs, having a conversation involves feeding the model the entire conversation up to this point. That is, there is no memory: the prompt you feed it just gets longer and longer. So how does that work with something like GPT-4O which could be processing audio and/or video at a much faster rate? Surely the prompts must get very large very quickly with anything beyond a short interaction? Doesn't that mean the responses take longer and cost more as the conversation gets longer?

jcsteh, to random

Half the time (probably more) when I restart NVDA, usually due to UIA shitting itself in consoles, everything goes to shit on my system. Alt+tab won't report correctly, a bunch of Windows go not responding, etc. I end up having to log out and log back in again to restore functionality. I can't figure out what the hell is causing it and it's really difficult to debug because well... the system is unusable. Damn I hate bugs like this. And anything that makes my screen reader not read things to me to that level has a tendency to frustrate me to the point of rage.

jcsteh, to random

Oh cool. Just got the "We've got an update for you" dialog from Windows... with three unlabelled buttons and no ability to close with alt+f4. Good to see that continued commitment to accessibility right there in a highly visible part of the OS. They must all be busy with that pivot to AI.
The buttons don't appear to have any text on them or at least none that OCR can pick up. I thought I'd check the UIA AutomationId just in case that revealed anything useful. The buttons do have different AutomationIds... of "Button2", "Button1" and "Button0", in that order.

jcsteh,

@ToniBarth I'm not sure. It doesn't give me any useful information. It's probably a cumulative update of some kind, but not sure why it isn't just rudely auto rebooting like it usually does.

jcsteh, to accessibility

GitHub produced a great mini documentary about my work with @michaeldcurran, Scott Chesworth and others to create both #NVDASR and #OSARA.
With audio description: https://www.youtube.com/watch?v=N-y3yomLLSk
Without audio description: https://www.youtube.com/watch?v=L076ngGbBRc&t=0s
#accessibility #a11y

nick, to random
@nick@hkc.social avatar

@jcsteh Hey man, hope you're well...

Interesting Firefox behavior in 126. Things that formerly exposed to the screen reader as "tabs" are now Links, as things that used to show as "menu items" are now "buttons". Nothing in release notes. Intended behavior?

jcsteh,

@nick Nothing i can think of. Do you have a particular test case or site where this occurs?

jcsteh,

@nick Ah, actually, we did tighten up contextual roles to match the spec. For example, it is a spec violation to have a menuitem outside of a menu or a tab outside of a tablist. Previously, Firefox didn't validate that, but it does now. But that was always a spec violation by the author; Firefox just chose to ignore it previously, which is itself a spec violation. Personally, I'm not a big fan of the spec enforcing this on browsers, but that's what it currently says.

jcsteh,

@matt It's a tricky one because the spec specifically requires user agents to enforce this, and there are web platform tests we fail if we don't. It is not just an author requirement; it is also a UA requirement. @nick

jcsteh,

@matt That said, there is a bit of a debate going on about this at the moment. Mozilla originally protested implementing this, then went ahead and did it anyway assuming in good faith that the spec was written like this for a good reason. But now Apple, after initially agreeing with the spec, are contesting it as well. So there might be scope to get this struck from the spec. @nick

jcsteh,

@matt Whether it's good for accessibility is debatable in this case. A menuitem without a parent menu isn't going to behave as one might expect and will possibly behave inconsistently across screen readers. @nick

jcsteh,

@matt Regardless, I commented on the relevant spec issue: https://github.com/w3c/aria/issues/2166#issuecomment-2111381939 @nick

jcsteh,

@matt Also, to be clear, we conceded in this case because we didn't have any evidence that this would hurt users and there is some validity in trying to ensure a consistent experience across assistive technology products. We can and do push back on spec requirements when we believe they are problematic, and generally, given a sufficiently strong case and no strong counter-arguments, these things get changed. The spec groups are a lot more pragmatic these days and I do believe that everyone wants to do the best thing for the web as a whole. @nick

jcsteh, to random

I spent half an hour on the weekend, maybe more, braille labelling all the reusable storage containers we use in our pantry. Today, most of the labels have come off, with the remaining ones not far behind. Why the hell is the adhesive on braille labelling tape so utterly useless? Sighted folks have labels which just stick and stay stuck. It's hard enough for blind folk to label things without this bullshit. Fuck!

TheQuinbox, to random

@jcsteh I've made a bit of headway on some of the random crashes and freezes I've been talking about with NVDA. Most recently my alt tab was bugging out majorly, so I alt tabbed and closed my WSL window. Once I did that, I started getting absolutely bombarded with NVDA error noises, and it completely froze up and was hardly reading anything, exactly like I had seen before. However, it was playing error sounds this time (because I turned them on), and this was in the log. The only way to get out of it was to restart NVDA with the desktop icon. Exception in IUIAutomationEventHandler.HandleAutomationEvent implementation:
Traceback (most recent call last):
File "comtypes_comobject.pyc", line 154, in call_without_this
File "UIAHandler_init_.pyc", line 569, in IUIAutomationEventHandler_HandleAutomationEvent
File "comtypes_init_.pyc", line 278, in getattr
File "monkeyPatches\comtypesMonkeyPatches.pyc", line 32, in call
_ctypes.COMError: (-2147220991, 'An event was unable to invoke any of the subscribers', (None, None, None, 0, None))
Not sure if there's any other useful debug info I can give here but I'm saving this log.

jcsteh,

@TheQuinbox Does this kind of thing tend to happen more if a console spews a shit tonne of output?

jcsteh,

@TheQuinbox <sigh> I'm not surprised. I haven't seen quite the issues you're seeing, but I've definitely seen consoles just stop reading properly or interfering with UIA events elsewhere. My suspicion is that UIA Core throws a tantrum globally.

TheQuinbox, to random

I didn't see this advertised very many places, but it's saved my ass so many times. One of my friends wrote a C program for Windows that works like a pipe, and puts any output from a command line application into a multiline text field that you can search through and read. For example, astyle -h|show. https://github.com/samtupy/pipe2textbox

jcsteh,

@TheQuinbox Neat. I have a similar thing set up here which shoves piped output in a temporary file and then opens it in a separate Notepad++ session, but this is likely a lot more convenient for most people.

jcsteh, to random

Just switched back from Enafore to Semaphore. I love that Enafore has additional features like post editing, etc. Unfortunately, it also breaks a bunch of a11y focus stuff - dialog focus, re-focusing posts when returning to timelines, etc. - and I don't have time to chase it down at the moment.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • megavids
  • thenastyranch
  • rosin
  • GTA5RPClips
  • osvaldo12
  • love
  • Youngstown
  • slotface
  • khanakhh
  • everett
  • kavyap
  • mdbf
  • DreamBathrooms
  • ngwrru68w68
  • provamag3
  • magazineikmin
  • InstantRegret
  • normalnudes
  • tacticalgear
  • cubers
  • ethstaker
  • modclub
  • cisconetworking
  • Durango
  • anitta
  • Leos
  • tester
  • JUstTest
  • All magazines
    •