The problem with accessibility-focused frontends is that they lack in features. There's a much smaller number of users able to detect bugs and a much smaller number of developers willing to fix them and add new features whenever they appear in the mainstream app. This works pretty well for i.e. accessibility-first Twitter clients, but it probably wouldn't for your local utility company that serves ten or so blind customers, none of whom are programmers.
Isn't that a consequence of building different tools that fit people with different situations? We do exactly this with physical tools and constructions: we build a wheelchair-friendly ramp alongside a traditional stairway. It's going to be a trade-off either way, right?
If we "choose" (I fully understand this is not an actual choice done by anybody) to stick to tools which try to auto-magically translate visual-first frontends into accessible frontends - we must accept that there will be things that mess with that translation process and cause it to fail. It's unrealistic to expect a translation that works 100% of the time, reliably. Sometimes it's a minor technicality that can be worked around - but other times it's because data presented visually in a certain way, can't always be translated to something else (ie. a heatmap, a 3D multiplayer game, etc). How do we make the Mona Lisa accessible to a blind person? On that same strength - should TUIs not be allowed to have progress bars or other visual elements?
To follow your example, we don't build separate bank branches for wheelchair users, we just extend existing branches with ramps, elevators and anything else they might need.
We already do this with other kinds of user interfaces. For example, web pages can be annotated with ARIA attributes, which don't influence how a page looks, but tell the screen reader that something is a checkbox which is currently unchecked. This is only required if you don't use the native HTML checkbox control, of course. Other platforms have their own ways of doing this, see iAccessible and UIA on Windows, ATSPI on Linux and AXElement on the Mac, for example.
Terminals were never meant for use cases like this, look at how hacky tty control codes are. If not for the fact that they're basically a bunch of ugly cludges that somehow work, there would probably be an aPI for that too. For now, though, we have to live with what we have.
So we are in agreement that there should be different interfaces that suit each experience, we just disagree on the exact implementation details. I submit it's best to build entirely seperate experiences, but you submit it's enough to take the graphical-first experience and annotate it enough so that a screen-reader can generate an equivalent experience on the fly (using different kinds of annotation technologies). My response to this is:
1. Annotations and metadata (ARIA-labels, et-al) make it easier for the screen-reader to display relevant information in an accessible manner - but they create an unnecessary coupling between the visual-first frontend and the accessible frontend, when in reality they are built for different kinds of users.
2. Annotations are a decent starting point, but they are NOT a substitute for building an "accessibility-first" experience because they are too limited. You can't annotate a graph, or a progress bar, for example. But you could've built an entirely seperate experience which conveys the same data a graph would, in an accessible manner (given the right tools and frameworks).
This is exactly why section 508 and ADA liability rules exist. If they don’t know how to make their programs accessible, a very expensive lawsuit will show them how. Also, why would a small utility be writing their own software? Shouldn’t they be using COTS?
I think its clear from your post that you have never dealt with the visually impaired or even explored accessebility options of iOS/MacOS/Windows. We have working accessebility for all applications throughfully designed with native OS controls.
Having separate front ends will never happen, most businesse sproduce a single bloated JS app that crashes equally on all platforms.
...which is a simple consequence of Windows people not being used to polyvalent software. You're putting horse before the cart.
The comment I'm responding to says that progress bars in the terminals, and that TUIs in general are an accessibility nightmare.
Are you disagreeing? Are you claiming that it's a solved problem? Would you mind elaborating on what you think can be done to solve this?
The bulk majority of applications people use today are not "thoughtfully designed with native OS controls". Sprinkling some ARIA labels around is better than nothing, but the experience itself is extremely limited and is not on-par with what those people deserve.
I'm talking about Javascript/CSS-heavy websites in a web-browser, computer games, or software such as Blender/SketchUp/Google Earth to name a few. ARIA labels aren't good enough here - we would need to develop an entirely different interface to accomodate for each individual accessibility scenario. You can imagine a blind person and a deaf person using entirely different versions of Blender, each built with a different interface accomodating a specific disability.
You will still find many that don't do this properly, especially if you run the commands through a CI tool which is typically a dead giveaway. Browser based log viewers aren't going to handle VT100 escapes and so you see garbage in the output. In this regard using unicode emojis to pretty things up works more reliably.
I would presume that a screen reader should do the same thing. Turn off tty mode so that the stream is easier to parse.
And there is nothing that says that it should be accessible. Which is exactly why a terminal, going by the very definition you yourself provide, is very difficult to work with from an accessibility perspective. There are simply too many degrees of freedom, allowing you to represent data in too many different ways, making it very difficult for something like a screen reader to "parse" the buffer (which can contain anything, not just text) and convert it to something meaningful.
Using a terminal, one can implement a progress bar in hundreds of different ways - and it would be impossible for a screen reader to handle all such use-cases.
There's no sense in expecting a visual application (such as the terminal) to be somehow accessible out-of-the-box.
Trying to automatically translate a visual interface to an accessible interface is bound to be error prone and not work in all cases. What's accessible for a blind person, may not be accessible to a deaf person.
Applications should share the core business logic, but have entirely seperate "frontends" for every situation. A graphical frontend, and an accessible frontend. We need an open cross-platform standard for this, and enforcement.