"Design" as practiced today by computer people tends to be heavily based on the idea of negative space: that good design is what's NOT in a system, and by extension, what is NOT ALLOWED TO BE ADDED to the system by a user.
A "design-heavy" system, then, is inevitably highly restrictive about user actions, lest they "ruin the cool design" by adding their own desired features that "make it messy".
But users WANT control over their own space.
Is there a "design" that empowers users?
@natecull linux cli
@pnathan Sadly the CLI concept didn't extend even as far as the X Windowing System.
There may be a reason for that.
hypercard was an amazing thing and i still miss aspects of it, but language-wise hypertalk suffered a lot from mistaken ideas about english-like ergonomics, i think.
right, but that's not using X *as X* or to interact *with* CLI. It's using X just as a set of terminals for CLI that's completely unaware of X.
X, the system (as I understand it), does not use CLI. It sends and receives messages of its own devising and format but doesn't expose these to the user or use the CLI or pipe system to do it (except for some command options for programs which are mostly now ignored)
Tcl/tk, though, I think does use Unix pipes?
and yeah, X is a giant weird beast and i've never actually written code against it in any meaningful way, but i use stuff like xclip, wmctrl, dmenu, rofi, etc., to control facets of it pretty routinely, so it's at least situated in a cli & scripting-driven environment that i use to control my own space.
surely not one without substantial design flaws and a gnarly learning curve, but i refuse to accept systems that lack many of its properties, especially the ones that let me add and re-shape features.
Well, because X is about *graphical components* of which 'windows' are just one tiny, tiny subset
Like you might want your program to accept input from and send data to buttons? Grids? Charts? Images? Textboxes?
None of that is even really possible with the 'a simulated VT-100 running just inside a window' use case of X.
So CLI is not interfacing with X 'as X', just X 'as a VT-100'.
@dredmorbius @pnathan @brennen all of that though is probably possible with Tcl/tk and maybe with some of the oldschool CLI tools you mention? (which really, really don't get much press at all even among hardcore Linux fans)
like to properly represent the state of X as CLI I think you'd need at least an object model? cos things on the screen ARE objects, they have persistence and state
Maybe all I want is a modern Tcl/tk that works with all the shiny GUI toolkits and OOP systems?
@natecull @pnathan @brennen NB: I'm massively hampered in all this discussion principally by /never having really grokked GUI application design in the first place/. My sense of "application" has always been "something that works on a stream of inputs or polls regularly for state", but not "and presents this with a bow and cherry on top".
I could use pointers on some good books on GUI UI/UX. Don Norman seems to be one source. Brett Victor another.
It seems quite hard to think of how to encapsulate the IO flow and state of a GUI component.
ie the input stream could be a stream of events (mouse/keyboard or underlying widgets, data model) and a similar output stream... but it all seems a bit confused and intertwined compared to a simple script that reads a stream of records and writes a stream of updated records.
Could *maybe* separate the IO flow into 'channels' like 'standard output' and 'standard error'
and maybe MVC could be pictured at toplevel as:
(stream of view-modification requests + stream of model updates) --> UI component --> (stream of view-updates + stream of model updates)
but it still seems not right. It's a whole nonlinear *data* flow. And that nonlinearity is what makes it hard to modularise. Too many wires snaking out in all directions, all needing precisely matching sockets.
There is also the distinction between "nodal" utilities and "complexity hubs". The latter do stdin/stdout poorly, if at all (though some do in fact do it). Most complexity hubs have *multiple* inputs and outputs.
For sufficiently simple stuff, I write utilities that I'd otherwise use a spreadsheet for. E.g., summing a sequence, or generating univariate moments.
excel is a funcprog environment that
1., normal people use, and normal people use to great effect to get things done
2. actually offers a non-linear way to do programming; instead of a linear program, you operate on cells of a 2D space containing either data or formulae, and operating on ranges is easy
adding VBA was a terrible mistake that added complexity and inelegance
What if the whole GUI was like an Excel spreadsheet, and widgets were cells? They have a value, the value can be changed, it doesn't care how the value changes but if it does, that change propagates to all observers.
http://www.haskellforall.com/2014/06/spreadsheet-like-programming-in-haskell.html introduces this, http://www.haskellforall.com/2015/11/haskell-native-spreadsheets.html goes over the initial release of that lib, and http://www.haskellforall.com/2015/11/interactive-and-composable-charts.html goes over diagrams using it.
I'm contesting that it's a /good/ FPE.
It is *VASTLY* too error-prone and difficult to debug. Not to mention awkward. OTOH, *it is often the only tool available since "real" programming tools are denied to front-office workers.*
Ray Panko, Univ. Hawaii, has studied Excel errors since the 1990s:
Excel offers a *grid of code* model. That's hard to see, visualise, debug, QC, validate, etc.
It's *possible* to write Very Robust Excel. It's not *easy* to do so.
I'll also note that I'm talking of my own personal use: I'm not saying that *everyone* who has an Excel-like task should use Excel. I'm saying that when I've got a simple "accumulate a store of values, compute something, kick out report", I prefer awk.
I suspect this is because spreadsheets don't offer *grouping* constructs, which are pretty important for humans to organise information.
A 2D grid gets you a long way, but not really far enough.
If you had a sort of web of nodes which you could expand or shrink, and had a natural mapping TO a 2D grid if you wanted it...
the achilles heel is how bound the physical structure of the data is to the representation & interface, but maybe that's what makes it approachable. there has to be some way for a serious db to be as approachable as excel is, but i don't know what it'd be.
I think perhaps copying ideas from SQL for structure and drilldown would be interesting (another funcprog environment with good ideas, but clumsy tooling; SQL feels like funcprog if it were implemented by programmers who only used COBOL, and half the functions are missing - i'm convinced SQL is also why people swear of RDBMSes to the point of using Mongo et al)
those caveats do matter, though. my hunch is that mostly people get lured into garbageware like mongo because sql rdbms systems make manipulating schema a difficult, largely out-of-band sort of task relative to most of their interface.
@brennen @natecull @dredmorbius @pnathan see, I always think talking about "schemaless" is a red herring when it comes to NoSQL - it's really about SQL as a query language & about making the DB just plain ol' serialization of Plain Ol' Objects - this is easy for a programmer, and it's a *schema*, typed language or not
SQL is only lovable if you're a 70s DBA. I'd want a modern RDBMS protocol of sorts, that makes queries and serialization easy - with strict schemas, since the other end has those
@brennen @pnathan @dredmorbius @calvin yeah, a database where defining a schema is an out-of-band operation seems a little like a functional programming language where you can't pass functions to functions.
One of the defining characteristics of data is that it exhibits recursive structure: things inside other things.
RDMBSes in the SQL model, however, are built assuming that there are certain Things (databases, tables) which May Not Ever Be Put Inside Other Things.
@natecull @brennen @pnathan @dredmorbius well, hierarchial DBMSes exist (a lot of NoSQL can be considered as such) but then you lose the advantages of RDBMSes. the problem is that mappings in SQL between tables are clumsy. what if RDBMSes had typed pointers between objects that made it easier to related between objects between tables?
Relational 2D data is convenient. It does /not/ fit all sets of circumstances. Normalising hierarchical data may or may not be possible. And often you're dealing with a cross of "what is present state" and "what is most recently-added transaction" (or "what is the full transaction history"?)
so it's naturally a lot simpler to work with 'an object full of objects full of objects'
'a database, which I gotta get my RDMBS administrator to spec a server and provision for me and define a schema which if I ever then change I gotta file a helpdesk request to do it and butwhooops the user just put a new data type in and crap...'
So it's not /all/ that bad.
And there are any number of CLI / scripting languages that promote Very Bad Habits. PHP is notorious for this. Perl can definitely get you there. I've even heard C occasionally causes problems.
And, for that matter, contrast "X Power Tools" to "Unix Power Tools" The latter was indispensable, and even today remains highly useful despite being first published in 1993 (there have been 3 editions, through 2007). The former was ... dispensable, and seems only to have seen 1 edition.
Since you're dealing with what a person can *see* and *manipulate*, both *directly*, there's only so much bandwidth for information or control.
CLI has a vastly larger interface surface, if you will. The options are far vaster.
I have 1,000s of commands available to me from the OSX or Linux bash prompt.
My OpenWRT router: 660.
@dredmorbius @pnathan @brennen One reason for this is probably that GUI actions tend not to be joinable or linkable; there's no fundamental GUI 'grammar' for anything like the concept of 'two symbols separated by a space' in CLI.
other than 'drag icon1 onto icon2' for a kind of primitive 'apply icon2 as argument to icon1 as function', but it executes immediately, it doesn't give you a third object which is the result of combining the two.
Ironically: if you /start/ with CLI and toss a GUI wrapper around the elements, you can address some of this. I'd argue that a problem (and, more to come, quite probably advantage) is that one GUI tool cannot /easily/ "become a keyboard or mouse" to provide inputs.
First: think of the malware possibilties.
(X11 already had this problem: grabbing mouse/keyboard inputs.)
This is also usually test/QA stuff.
@brennen @pnathan @natecull That is, tools that /can/ automate GUI bits are generally used for testing and/or QA. I've used a few briefly a long time ago -- can't even remember names. OK, Selenium was one I recall.
The other problem is sorting out /what element you want to talk to / hear from/. You've got a bunch of crap on screen, there's no sense of pipelining or stdin/stdout/stderr.
Named sockets /might/ work.
I remember hearing that Amiga, and OS/2, and maybe NeXT offered much of this, but I've never really seen any particularly convincing demonstrations.
(OS/2 seemed to have a sense of "workspaces", though they seemed underwhelming to me.)
There's also the problem of preserving user state *within* GUI apps. *Even if I can fire up all the shit that I had open now, on reboot, I'd lose massive amounts of state.* Which fucking sucks.
(There's also the month+ old problem of a nuked Chrome Browser extension's user state that's been on my to-do list for that month+, though I have to dig into the fucking headspace of Chrome and the extension to see how that happens. The developer of the extension himself *does not know and isn't interested in finding out.* Fucking sigh.)
Follow friends and discover new ones. Publish anything you want: links, pictures, text, video. This server is run by the main developers of the Mastodon project. Everyone is welcome as long as you follow our code of conduct!