I've been posting elsewhere about my recent adventures in setting up dictionaries on my computer.

On my computer I keep 21 dictionaries that can be queried using the command "dict". I could simply use the freely available ones at dict.org, by either accessing the website or using my dict command on my machine to to query their online dictionaries, but in the past my internet would often go down, which was particularly annoying if I was writing at the time and needed to check a dictionary, so I like to keep them locally on my machine, so I can still look up my dictionaries whether I have net connection or not.

Last night I was looking up the spelling of an unusual word which didn't appear in any of my local dictionaries, so I figured I'd see if it was in any of the 166 dictionaries at dict.org, so I used their pseudo-dictionary "all" to force it to search all their dictionaries, including all their English translation dictionaries. (It was found in the English-Finnish translation dictionary. I hadn't bothered to install any translation dictionaries.)

One thing I don't have is the ability to use a single query to check all my local dictionaries (those on my computer) for a word. I must find out how to do that. It is really convenient.

Oh, and you're probably wondering why I keep 21 dictionaries on my computer. Well, most of them are specialised. Here is the list of my local dictionaries:

bouvier - Bouvier's Law Dictionary, Revised 6th Ed (1856)
devils - The Devil's Dictionary ((C)1911 Released April 15 1993)
easton - Easton's 1897 Bible Dictionary
elements - Elements database 20001107
foldoc - The Free On-line Dictionary of Computing (27 SEP 03)
gazetteer2k - counties U.S. Gazetteer Counties (2000)
gazetteer2k - places U.S. Gazetteer Places (2000)
gazetteer2k - zips U.S. Gazetteer Zip Code Tabulation Areas (2000)
gazetteer - U.S. Gazetteer (1990)
gcide - The Collaborative International Dictionary of English v.0.48
hitchcock - Hitchcock's Bible Names Dictionary (late 1800's)
jargon - Jargon File (4.3.1, 29 Jun 2001)
journo - Free journalism dictionary, version 1.1
minimal - Minimal Dictionary (I created this)
moby-thesaurus - Moby Thesaurus II by Grady Ward, 1.0
vera - Virtual Entity of Relevant Acronyms (Version 1.9, June 2002)
web1913 - Webster's Revised Unabridged Dictionary (1913)
wn - WordNet (r) 2.0
world02 - CIA World Factbook 2002
world95 - The CIA World Factbook (1995)
www - Who Was Who 5000BC to Date

If you have "dict" installed on your computer and you list the dictionaries available at dict.org then it will list 166 of them. The first 15 will be various specialised dictionaries (law, computing terms, thesaurus, English dictionaries, USA places, etc). 148 will be translation dictionaries (e.g. Croation-English, suomi-português, etc.) but the last 3 dictionaries are special:

 english        English Monolingual Dictionaries
 trans          Translating Dictionaries
 all            All Dictionaries (English-Only and Translating)

If you check their details these last three are zero bytes in size. I wanted to find out how to set up these kinds of pseudo dictionaries so I could search all my local dictionaries at once for a word. A google search didn't turn anything up, and rummaging around in the documentation at dict.org turned up little (I remembered years ago that it took me a lot of effort to find the documentation at dict.org).

So I asked ChatGPT-4o. It was eager to help and confidently told me how to do what I needed... except it was completely wrong. I spent much of the day trying its various suggestions until I finally decided to cut my losses and give up on the AI.

Somehow, I had forgotten that when I'd last looked into dict a while back I'd saved a whole lot of data that I'd found online. (Thank you to my obsessive nature.) So I spent some hours reading through the documentation, which I should have done right at the beginning. And I found the solution. It is unbelieveably simple. Add to the list of databases -- mine is in /var/lib/dictd/bd.list -- a slightly different entry.

Most of the entries are like this:

database foldoc {
  data  /opt/dict-dbs/foldoc.dict.dz
  index /opt/dict-dbs/foldoc.index
}

"foldoc" is the name of that dictionary of computing terms. The word data file and the index to it are kept locally on my machine at /opt/dict-dbs. I have 21 of these kinds of entries.

To get a pseudo-database I simply need to add:

database_virtual comp {
  database_list "foldoc,jargon"
}

In this case I'm defining a virtual database which I name "comp" that I can search for computer terms. It contains a list of other databases to be searched. In this case they are foldoc and jargon -- the two computing dictionaries I have. The list of databases to be searched must be separated by commas and be enclosed in double quotes.

Now if I want to look up a computing term (like "fork") I can use:

dict -d comp fork

Which will search both the foldoc (Free On-line Dictionary of Computing) and jargon (The Jargon File) for their entry on fork.

So simple.

It's a pity I spent so much time on the AI's wild goose chases. I should have referred first to my enormous store of documentation. It's slow and frustrating to read documentation, but it would have ended up saving me many hours of time.

Nevertheless, I learned in my failed explorations with the AI that I can use various search strategies with dict, such as regular expressions and soundex. That's exciting!

-----

Again, I'm such a dummy. Turns out I didn't need to spend all that time to make virtual databases for dict just simply to be able to search all 21 of my local databases for some word. All I needed to do was just not specify a database. 😃

For example if I want to search for "fork" in WordNet I give this command:

dict -d wn fork

But if I want to search in all my databases I just have to give:

dict fork

Interestingly, it doesn't work exactly the same in the online databases. That simpler command just searches the all the conventional English databases and ignores the translation databases. In that case,

dict -d all fork

will search for "fork" in all the online databases.

But my effort wasn't entirely wasted. Apart from learning a lot, I created a whole bunch of virtual databases that let me restrict a search for words to my 2 dictionaries of computer terms, or my several databases of geographical places, or one of several other virtual databases.

So, yay! 🙂

What a relief! I was starting to panic. Sound had ceased to work in my computer. I've really needed a new computer for a while, but really can't afford one, and dreaded having to find the money (that is, borrow it) to buy one.

I keep several operating systems on my computer, so I booted into another one, to check. Gratifying surprise! The sound worked. So it was not a hardware problem... thank goodness.

So back to my preferred operating system. Why isn't sound working in it? I tried everything I could think of, but was reluctant to spend hours on the net searching, so I began with asking ChatGPT-4o if it had seen anything on the net about this particular problem. It spent a little while searching the net and came back with 8 suggestions gleaned from its search.

The first suggestion I'd already tried.

The second suggestion (to check that alsamixer was not muted) solved the problem. How it had got muted I have no idea, but now everything is okay again.

Whew! My relief is overwhelming. :(

One of my all-time favorite cartoons:
Eek & Meek by Howie Schneider.
It is astonishingly difficult to find his cartoons online.

Last night, after watching Robert Murray-Smith's wonderful short video about the Accutron mechanism, in which he mentions an amazing book, "1800 Mechanical Movements Devices and Appliances" by Gardner D Hiscox published in 1921, I went ahead and downloaded it:
https://archive.org/details/1800-mechanical-movements-devices-and-appliances_202005

Depending on the kind of book I'm fetching I'll get one format or another. The InternetArchive almost always has a link to "SINGLE PAGE PROCESSED JP2 ZIP". I generally don't use that link, even though it is a zip file of the original page image scans in extremely high quality JPEG2000 format, because I have almost nothing to view the images or convert and resize them to something smaller and more usable. ImageMagick needs a special "delegate" converter which I haven't been able to get working. Netpbmtools' converter fails to even read the images. G'MIC needs a special computer vision library which I'm reluctant to add to my computer. And none of the dedicated viewer programs I have will even display it.

Just two programs will load jpeg2000 images: Gimp, and mtpaint.

I'd kinda expected Gimp would be able to. It is big and extremely capable -- able to do much, much more than I could learn in one lifetime.

The great surprise was mtpaint -- a tiny, exceedingly fast, deceptively simple, paint program. And I dimly recalled from somewhere that mtpaint could be used from the commandline without opening the GUI (graphical user interface). So I typed into the terminal:
mtpaint --help

It responded with:

mtPaint 3.49.13

Usage: mtpaint [option] [imagefile ... ]

Options:
  --help          Output this help
  --version       Output version information
  --cmd           Commandline scripting mode, no GUI
  -s              Grab screenshot
  -v              Start in viewer mode
  --              End of options

There it was, commandline scripting mode! I'd been greatly underestimating little mtpaint.

Next, I quickly looked through its manual to see what I could learn about its scripting mode, but didn't look closely enough so I missed it.

It occurred to me to ask ChatGPT-4o. It seemed very helpful, giving lengthy descriptions of how to use mtpaint scripting. So then I sat down to try its examples... but none of them worked.

So now I examined the manual more closely and found chapter 10 discussed scripting. It turns out EVERY example GPT gave was wrong, and not just in minor ways, but in every way. That really surprised me.

The only thing left was to patiently read through the manual's chapter on the topic.

I'm quite pleased at its scripting language. It uses the GUI as its reference. Commands are based upon it. So if I want to know the way a command should be written I simply find it in the program's GUI and that will define it for me. Very smart. For example if I want to load a jp2 image, scale it to 22%, with proportions fixed so height and width scale together, then save as an ordinary jpg image at 90% quality, then I write this on the commandline:

mtpaint --cmd -file/open="00.jp2" -image/scale w=22% fix=1 -file/as="00_22.jpg" f=jpeg q=90

The "file" menu in the GUI contains the "open" and "save as" commands, and the "image" menu contains the "scale" command. The other settings are part of those commands in the GUI, so immediately follow those commands with an unambiguous key (for example "width" or "w" both work fine).

Using mtpaint on the commandline I can quickly and easily convert and rescale hundreds of jp2 images to usable size and format using a simple looping command. Yay!

One last thing. I mentioned that I didn't have a program that would simply display jpeg2000 images. I was wrong. I noticed mtpaint has another little surprise. If I make the mimetype command for all jp2 images
mtpaint -v "$@"
then mtpaint displays them in its own fast, simple picture viewer.

I have great new respect for mtpaint. I'd been underestimating it awfully.

One of the disadvantages of a small language like awk is the lack of some functions. It's also a great advantage as it is not a gigantic, sprawling language the way many have become nowadays.

Today I encountered the disadvantage, in that I needed something that worked like the eval() function many languages have, that would let me read in a string and execute it. In this case I had a long list of variables and the values I wanted stored into them. I needed to be able read them in to my program.

I tried a few ideas, including a few proposed by ChatGPT-4o. None worked... until I came up with a solution that even GPT-4 hadn't heard of. I made the settings file (named ".canvas") into a fake awk file with
function settings(){
at the top, then a list of the variable assignments, and the last line the closing curly brace.

In my main awk file I have
@include "./.canvas"
near the top of the file and whenever I want to read the variables into my program I simply have
settings()
in my program. Easy peasy. Yay! :D

For a couple of weeks I've been trying to work out how I might create my own procedural textures like those in POV-Ray. Partly this is an intellectual exercise, but partly it's because I'd like the versatility of being able to control textures myself when I make virtual worlds.

Today I think I've solved a crucial part of the problem. I used a simple pattern of waves and distorted them in a way that if taken further can create marble and granite textures.

Here is how I did it:

I began by making a very orderly heightfield of straight, parallel, sawtooth waves (so named because they ramp up from black to white in a linear gradient, then down again in a linear gradient, which if you graph their values look like the teeth of a saw). I made that using ImageMagick. Here's how...

Make a gradient:

convert -size 512x32 gradient: -rotate 90 ramp.png



Duplicate it reversed horizontally to make a single full sawtooth wave:

convert ramp.png \( +clone -flop \) +append sawtooth_1.png



Duplicate it multiple times to make 8 waves:

convert sawtooth_1.png +clone +append +clone +append +clone +append sawtooth_8.png



Okay. That's the nice orderly image that I want to mess up and distort. To do the next step I need a displacement map to tell G'MIC how much to displace each pixel in the image I want to distort. G'MIC can produce a number of noise images. Its "turbulence" command, is perfect for making a nice displacement map.

Generate a blank canvas of 255x255x1 (X,Y,Z) with one color channel so that it is grayscale. (3 channels would be 3 color rgb) Then add turbulent noise with a radius of 80 pixels, so that large structures dominate. All the other parameters, leave at their default settings.

gmic 512,512,1,1 turbulence 80 -o turbulence_80.png



Now we warp the orderly image with the turbulent noise image :

gmic turbulence_80.png sawtooth_8.png -warp[-1] [-2],1,2,2,1 -n[-1] 0,255 -o sawtooth_turbulence_80_warp.png



Ta-da!!!!

Hard to believe this took me so long to work out. Interestingly, GPT-4 was almost no help at all... other than the fact that it introduced me to G'MIC, which I hadn't heard of before. GPT-4 slowed me down by sending me off on several wild goose chases with its terrible efforts to write code.

Here's a nice trick for Linux users.

When outputting text to the terminal, most commands don't do smart-wordwrapping, however text is generally much easier to read if words are NOT broken at the ends of lines. Here is a way to make any text output have smart-wordwrapping.

In your .bashrc add this line:

alias ww="fmt -s -w $(tput cols) - "

It uses the fmt command to wordwrap text according to how many columns wide (-w) the terminal window is (tput cols). The -s option prevents it joining short lines, so it only splits long lines. The final '-' makes it take its input from the standard input.

Now, when you output any text to the terminal just pipe it through ww and it will display as smart-wordwrapped, for example:

cat test.txt | ww

Don't use it to output to a file though, unless you want the extra linefeeds inserted into your file for some reason.


A big shortcoming of the fmt command is that you can't remove all carriage returns inside paragraphs, if you want to undo its formatting.

I solve this using a surprisingly complicated-looking sed command:

sed -z -r 's/[ \t]*\n[ \t]*/\n/g; s/([^\n])\n([^\n])/\1 \2/g'

It's simpler than it looks though.

The -z option treats any zero bytes as end-of-line markers, so since most text files don't contain any zero bytes it treats the entire file as a single line. This lets us treat newline characters as just any other character.

The -r option lets sed use extended regular expressions. This simplifies things because it means a lot of regular expression characters don't have to be prefixed with the '\' escape character.

The sed command breaks down into 2 instructions. They're both substitute commands: s/search/replace/g
The 's' means substitute. The text to be searched for and the text to replace it are both surrounded by '/' characters. And the 'g' means global replace, so every occurrence is replaced

The first command is:
s/[ \t]*\n[ \t]*/\n/g
It searches for any amount of spaces and/or tabs [ \t]* that immediately precede or follow a newline character and deletes those, leaving just the newline character \n. The asterisk means any number of the preceding pattern [ \t].

The second command does the real work:
s/([^\n])\n([^\n])/\1 \2/g
It is a bit weird. It marks some patterns with parentheses () so that sed can make use of what they match. If the first character of a pattern inside square brackets is '^' then it means the reverse, so [^\n] means any character that is not a newline character. So this search pattern means any newline character with ordinary characters on either side of it.

The replace pattern is a lot simpler. The first pattern character in parentheses is remembered and put back into the output. The newline is replaced by a space. And the second pattern character is remembered and replaced back into the output.

So this replaces with a space all lone newlines at the ends of lines, but doesn't touch paragraph ends that have multiple newline characters.

It took me quite a while to work out what is a pretty simple command. I tried a number of different languages. In the end sed rules them all, despite its dense-looking form.

I should rewrite it to let it ignore short lines, such as those in poetry... but that gets a bit complicated. I would have to find lines that are shorter than "normal" and leave them alone. What is "normal" or average line length for the file? How much shorter qualifies it as verse? Tricky.

I made this meme today:

I've just emerged from a several hours long trip down an interesting rabbithole.

I'm getting close to deaf -- so much so that I can no longer watch videos without subtitles. Unfortunately, about a year ago one of my favorite YouTubers released a 2-hour video without subtitles, and for some reason blocked YouTube from making automated subtitles.

Last night I watched a video by another of my favorite YouTubers that demonstrated that the video editing program "kdenlive" can now automatically subtitle a video from its soundtrack. This was exciting!

This morning I installed kedlive on my computer... but it doesn't have any subtitling functions. The version for my computer is pretty old, created in 2017. So I went online to the kdenlive website and downloaded the latest version and tried it.
https://kdenlive.org

I won't go into why, but suffice to say it doesn't work on my machine. I ended up downloading 6 older versions and none worked on my computer. There are many more older versions, but I didn't bother. The first version to introduce speech-to-text subtitling was one of those I tried (version 22.04).

This was way more complicated than my description indicates. In the end, I'm convinced I need a better computer and a newer Linux.