The Great Software Stagnation

Jonathan Edwards is one of those people I both profoundly agree with and profoundly disagree with.

In his new year essay The Great Software Stagnation he again complains about the way the computer science field seems to be running out of new ideas.

I agree wholeheartedly with his frustration. And, sure, with a few bright exceptions, innovation in computing seems very poor compared to the enormous strides made in the 60s and 70s.

Of course, some of this is that lower-hanging fruit have been picked. Some is the natural inertia of a large number of users of computers these days.

But I continue to oppose Edwards’ theory, that it’s all the fault of the internet and Silicon Valley luring people away with promises of getting rich. I think (as I’ve outlined in this following Tweetstorm response to Edwards) that it’s all the fault of “direct manipulation” as an ideal for UI designers.

The Great Software Stagnation is real, but we have to understand it to fight it. The CAUSE of the TGSS is not “teh interwebs”. The cause is the “direct manipulation” paradigm : the “worst idea in computer science” \1

Progress in CS comes from discovering ever more abstract and expressive languages to tell the computer to do something. But replacing “tell the computer to do something in language” with “do it yourself using these gestures” halts that progress. \2

Stagnation started in the 1970s after the first GUIs were invented. Every genre of software that gives users a “friendly” GUI interface, effectively freezes progress at that level of abstraction / expressivity. Because we can never abandon old direct manipulation metaphors \3

The 1990s were simply the point when most people in the world finally got access to a personal computer with a GUI. So that’s where we see most of the ideas frozen. \4

It’s no surprise that the improvements @jonathoda cites, that are still taking place are improvements in textual representation : \5

The main innovation in Rails is textual (you rethink the Ruby code that describes an ActiveRecord class as a schema definition for the whole system) \6

DevOps tools take configurations that were done through filling in web-forms or obscure config files and puts them back under control of scripts in a real programming language. \7

React is a new higher level set of linguistic constructs to describe a reactive UI … etc. \8

In other words, progress in CS is still possible, but has slowed down in all the places where people have stopped programming in languages where new ideas can be created through grammatical composition and abstraction \9

And where we try to get by with a fixed repertoire of gestures to manipulate data directly. \10

If you want to fight stagnation, you have to fight the cult of direct manipulation, and fight the expectation that computers should be so easy that users don’t need to think or learn new concepts to be able to do things with them. \end

Now, I don’t expect Edwards to agree with me on this. He’s very keen on direct manipulation in his experiments. And I understand that. Direct manipulation is a wonderful thing when your gestures are sitting at the top of a tower of linguistic abstractions. Mapping that ultimate level of power directly to your body and visual system gives you remarkable power very easily.

The problem is that we don’t know how to grammatically recombine and build further abstractions in the gestural domain, so however great that power is, a direct manipulation approach will always freeze you at that point.

Sticking with text and language feels more artificial and clunky, but if you do stick with it, then you at least have a chance that you’ll later on invent a higher level language to go beyond your current state.

A Test of Mind Traffic Control

Today I managed to screw up my todo.txt file that I use with Mind Traffic Control.
I had to recreate it from some backups that weren’t quite as up-to-date as I needed them to be.
And I got a lot of duplication in the file. And obviously items that had been done are now back in it. (Remember, I don’t keep “checked off” items)
So how to fix?
Diffing MTC is complicated. Because you do a lot of re-ordering of the queue in MTC, two todo.txt files can be almost identical, but the diff of them is huge. And it’s hard to understand how similar they really are.
Fixing that problem is kind of easy : you sort both copies of the file into alphabetical order, diff them again, and now it’s much easier to see and deal with how much difference there really is.
So I did that. On Linux simply

cat todo.txt | sort  > sorted_todo.txt.

And then

meld sorted_todo.txt sorted_backup.txt

And that was fine.
Deduping a sorted file is easy. I’m sure there’s a command line tool for it already but I wrote a 4 line Python script.
So now I have as up-to-date a version of the todo.txt as it’s possible to have. And it’s deduped.
it’s now sorted in alphabetical order. And has 1506 items in it.
This is going to be interesting.
So the philosophy of MTC is precisely that the order / structure of the overall queue isn’t really that important. It obviously IS important because your “Next Item” is what’s at the front of the queue.
But most of the time you deal with a few items in the front. And then searching / pulling / pushing “tags” to estimated distances into the future.
That’s worked fine. But I’ve never done a mass (and arbitrary) reordering of the queue before.
So this is going to be a big test. Are the facilities of MTC, plus my own previous tagging discipline and ability to remember what to search for, sufficient to the job of managing a list of 1500 items now that they are out of order?
In some ways, I’m happy to have to face this challenge. It’s going to be a very interesting test. I’m optimistic that MTC is up to it. But I’m keeping an open mind. I may find some issues. And have to write some extra functionality to resolve it.

Cardigan Bay Screenshots

I’m going to do some video tutorials shortly.

Just need to find a decent mic and a quiet space.

In the meantime, some recent screenshots of Cardigan Bay

How the wiki Looks
Here you can see multiple cards. The second is an automatically generated “Backlinks” card.
Recent Changes
Here I am looking at a page on IndieWeb. Note that my mouse is hovering over the “Bookmark to 4545” bookmarklet in the browser.
Having clicked the bookmarklet, we are taken to the InQueue page of the wiki. And you’ll see that a new card has been created with the link to that page.
Here I have opened up the “card bar” at the bottom of the card with the bookmark. And I can type in the name of another page to send that card to.
Clicking the Send button creates that new “IndieWeb” page, and moves the card with the link to it.
Here’s something different. We’re editing a card with some short Clojure examples. Nothing fancy, just filtering some numbers. Cards are separated with — and types like :evalraw and :evalmd (eval and render as markdown).
Save the edit and here’s the results. The first example rendered as plain text, the second was treated as markdown rendered as an html bullet list.
And here you’ll see that the results of the last Clojure snippet are in a nice table, thanks to our “double comma” table convention.
Here’s an example of “transclusion”. The CardiganBay page we saw above, has been transcluded into a single card within another page.
Finally some embedding … a bit too big for the screenshot, but here you see YouTube, SoundCloud and BandCamp embed within the wiki
And here was the source-code in the page for that.

Fixing the Lower-Case Mistake

After writing The Lower-Case Mistake this morning, that basically convinced me that I had to move back to having CamelCase file-names for Cardigan Bay.
So … a burst of activity this afternoon, and the automatic lower-casing of file-names is gone.
If you create a page called HelloWorld, it will be in a file called
The GitHub / GitLab repos are now updated and this is now the official behaviour of Cardigan Bay.
Obviously it’s a breaking change. Because if you were using it yesterday and your HelloWorld page was in file called the new version of the code can’t see it.
The default Bedrock Wiki pages that come with Cardigan Bay have now been fixed. But I’m afraid you’ll have to change any other page names yourself.
I wrote a simple script to help me convert the ThoughtStorms page-base (of over 5000 pages)
It wasn’t perfect, but it did enough of them that I’m happy to commit to making the change. (I’ll clean up the problem cases by hand)
Here’s the code to fix the names. It basically searches all the links in pages to find CamelCase names. Then prints a list of “git mv” commands that would turn the lower-cased version of those named files back into CamelCased ones. You just have to pipe that output to a shell script file and run it and it will do the changes.
Like I say, not perfect, but took a couple of minutes to write, and does the bulk of the job.

from fsquery import FSQuery
import re

fsq = FSQuery("pages/").Ext("md").NoFollow(".work").FileOnly()
d = {}

def grepFiles() :
    for f in fsq :
        for l in f.open_file() :
            l =l.strip()
            sqrwiki = re.compile("(\[\[(\S+?)\]\])")
            allnames = [x[1] for x in sqrwiki.findall(l)]

            for n in allnames :
                if ":" in n : continue
                if ">" in n : continue
                #if not n.islower() : continue
                d[n.lower()] = n
                for k,v in d.items() :
                    print("git mv '' ''" % (k,v))


You’ll have to fiddle with it a bit, and make sure you have installed. But like I say, it does the bulk of the work.
If you have problems (eg. a big investment in Cardigan Bay data but can’t do this migration yourself), leave me a message and I’ll see what I can do to help.

The Lower-Case Mistake

I’m doing a lot of of work on Cardigan Bay at the moment.

And it’s getting pretty good.

I wish I’d done all this sooner. I spent years with wikis, not writing my own wiki-engine because I thought that would be pointless reinventing the wheel. Then when I did, I realized I could make wikis do all the things I wanted from them.

Then I loved Clojure but kept writing my engine in Python because I thought it would be more straight-forward, a known art, and just quicker and easier.

Now I’ve bitten the bullet and moved to writing my wiki in Clojure (something that seemed yet another round of self-indulgence when I started it at the very end of last year) I realize that it’s exactly right, and Clojure is a super-powerful and easy language to do it in.

I’m making great progress and loving Cardigan Bay … (and very soon there’ll be a new release with lots of nice new features … watch this space.)


somewhere along the line I have made a huge, huge mistake.

This goes back to my earlier engine, Project ThoughtStorms, I think.

Or maybe I inherited it from somewhere else. I can’t even remember.

The mistake is that wiki page names are obviously CamelCased for readability. A convention I like very much and continue to use.

But I decided to make the file-names all lower-case. In other words, I throw away the capitalization of the name.

The only place that the capitalization exists, is in the link that leads to a page. So links look OK. And from following a link, we can deduce the capitalization we show the next page.

Which works reasonably. But you’ll notice that in RecentChanges and other queries for OrphanPages etc. the pagenames are all lower-case. They have to be, the file system has no knowledge of the capitalization.

That was a pain, but I could live with it.

Now, though, I’m working on the “export this wiki as flat files” functionality. (Which is going to be fundamental to my new thinking … more on which later)

And I have same problem. Running through the directory of pages, I can render them beautifully, but I have no information as to how the pagename should be capitalized. And so can’t give the static pages a reasonable looking title.

That is an absurd restriction which I certainly can’t live with. (And nor will anyone else)

So what is to be done?

I could decide to store the correct page name somewhere else. Either in the page files themselves, or another database. But that breaks the convention that pages are pure markdown files (or slightly extended markdown files). I don’t want to add my own special header convention and force people to deal with that when migrating into or out of Cardigan Bay. It should be possible to work on CB wikis in your own preferred text editor if you choose.

Similarly, I don’t want to depend on another database.

Or I can try to capture the capitalized version of the names when I scan the links into the temporary database at the beginning of each session. And use that when exporting. Which will work for pages which aren’t orphans.

Or I do an export the way I used to back in SdiDesk, not by grabbing everything in the page-store, but by crawling it via the links. Which would give me the capitalized version of the pagenames.

Or maybe I need to change the convention so we don’t lower-case the file-names. But keep the capitalization. I’m tending towards this. But I have 5000 pages of ThoughtStorms in files whose names aren’t capitalized. And another 4000 in my personal wikis. That’s a lot of work to fix.

Anyway, thanks for listening dear reader … just writing this has been useful for me.

Command line HTTP apps.

I just rediscovered a question I wrote on Stack Exchange. I can’t remember much about the context, but the idea is to have an ordinary “web server” application accessed through http, but simply sending plain-text backwards and forwards to a command-line based client.

command line – Is there a “terminal” style program that talks to a server over http? – Unix & Linux Stack Exchange

The question was :

I’m looking for something that acts like a terminal but lets me have a “dialogue” with a server over http. Something like this :

$ connect
Welcome to
A - Fribble the obsticator
B - List Frogits
C - Show the log
Q - Quit
$ A
Obsticator fribbled
blah blah
$ C
Log file
$ Q

Http is a useful protocol and it goes everywhere, it’s easy to write a server that handles it (thanks to libraries). But you aren’t always in a place where you have a browser or could run one. Sure there are browsers like lynx etc. that run in the terminal, but they aren’t particularly easy. Sometimes you just want something that’s more like the old Gopher etc. systems. But with the benefits of modern server technology.

No-one seems to have written one, so I mocked up a client in Python using the Requests ( library :

import requests
import sys
url = sys.argv[1]
print "Connecting ", url
r = requests.get(url)
print r.text
flag = True
while flag :
    s = raw_input()
    data = {"opt": s}
    r = requests.get(url, params=data)
    print r.text

A server now just has to respond to messages passed as “opt” on a particular URL, and to respond in plain-text.

Obviously, something like file-downloads etc. would also be useful.

Cardigan-Bay Early 0.0.3

Cardigan Bay development is slow, but continuing.
It’s starting to look a bit more respectable (I’m so NOT a CSS programmer, but I have to do something)

Get the latest : Release Cardigan-Bay Early 0.0.3 · interstar/cardigan-bay
This release is largely just bug fixes and cleaning up the UI etc.
But that makes a massive difference in usability on tablets and even my phone. It’s still not a mobile friendly UI, but it is at least readable and usable through the phone.

Carp: A statically typed Lisp, without a GC, for real-time applications.

This looks rather good.
Carp: A statically typed lisp, without a GC, for real-time applications.
It is what it says. It’s a Lisp, with some inspiration from Clojure in terms of syntax, that’s designed for writing fast native code like games and … (w00t!) audio applications. Etc.
It actually compiles into C.
It takes some syntactic inspiration from Clojure (so it looks like Clojure and Clojure editors should handle it fine)
But it also takes inspiration from ML.
It has an ML-style type-system with type-inference. And the best way of representing type information in a Lisp program that I’ve seen.
Using a form called “the“.
So you write something like

(the Int sexp)

This declares that the sexp needs to evaluate to the type Int.
Because it has type-inference, the type of much of the rest of the code can be deduced just from a few of these explicit specifications.
Carp also (allegedly) has an ML style module system. I have no idea what that means because I’ve not used ML. But it is something I hear a lot of people raving about.
However, Carp is a low-level language designed as a C-level language. It has the usual Lisp-style dynamic lists, but only for use at compile-time. It also has macros executed at compile-time.
At run-time code, you only have more primitive types : Ints, Floats, etc. Plus data-structures including Arrays, some kind of map / dictionary, structs and Sum Types. So you get the power and type-safety of ML / Haskell algebraic data-types but at a C-level without depending on heap allocation.
It also has some kind of Rust-like “ownership semantics” for pointers. I haven’t quite understood this yet, but again it looks sensible.
In all, this looks a really tasteful and well thought out combination of good ideas from a number of places (much like Clojure itself) that will add up to a much nicer than C experience for writing the sorts of things that I usually write with C.
All I’ve done so far is installed it and run a couple of examples, but I can well see myself trying to use this to do the coding for some audio / VCVRack stuff.

Tablets for Content Creation

Source: Thread by @interstar: @msimoni @coreload I think the tablet COULD be a perfectly good device for content creation if people would just do the damned UI design wor…
Another part of the thread. Manuel Simoni (@msimoni) says :

But I don’t want a file system, I want Xanadu. But it should still be editable through a WIMP/desktop UI.

I reply :
Sure. We all want / work towards Xanadu 🙂 Though at the very least I want mine to run on a tablet, have swipe-able cards like Material Design, and allow me to write Prolog-like rules for making automatic inferences that generate new information / views.
To which Manuel replies :
Do you think a tablet is a good device for creating content, or are you talking just about browsing?

And my response :

I think the tablet COULD be a perfectly good device for content creation if people would just do the damned UI design work to make it good. Right now, we haven’t figured it out. This is where graphical input could be very useful. \1
This is a whole other story.
It seems that social apps have figured out ways to get users to put a lot of data into their systems. Problem is that the data is just being used for social media type things. \2
Could we make UIs that “afford” input the way that FB, IG etc. do, which actually feed that information into more useful apps? \3
The Twitter “roll-up” type apps. and Telegram bots smell interesting here. Why not have more agents extracting data from streams of short inputs, which can collate them into larger more useful blocks of information? \end