> Warren Teitelman originally wrote DWIM to fix his typos and spelling errors, so it was somewhat idiosyncratic to his style, and would often make hash of anyone else's typos if they were stylistically different. Some victims of DWIM thus claimed that the acronym stood for 'Damn Warren’s Infernal Machine!'.
> In one notorious incident, Warren added a DWIM feature to the command interpreter used at Xerox PARC. One day another hacker there typed delete * $ to free up some disk space. (The editor there named backup files by appending $ to the original file name, so he was trying to delete any backup files left over from old editing sessions.) It happened that there weren't any editor backup files, so DWIM helpfully reported * $ not found, assuming you meant 'delete * '. It then started to delete all the files on the disk! The hacker managed to stop it with a Vulcan nerve pinch after only a half dozen or so files were lost.
> The disgruntled victim later said he had been sorely tempted to go to Warren's office, tie Warren down in his chair in front of his workstation, and then type delete * $ twice.
The problem is that shell is so sloppy it's hard to know what will happen. Of course, that's not unique to a new mechanism for navigating directories. It's scary stuff no matter what.
It feels far more intuitive and it's so incredibly convenient to be able to jump between folders I'm working in without having to traverse an entire tree. The shell used to feel so constraining for me, but tools like this are making me enjoy it so much more.
Most shells offer a convenient syntax around a lot of (fantastic) utilities that often mirror C libraries (notably posix). When we were working in C, having our shell mirror parts of that at times while being a very dynamic language at other times sounds great.
But we have a much more varied landscape of language and library options now, and many developers haven’t ever used C or posix.
So my thought is: what if we had a shell that was designed to interop with JS (or python, or ideally a language the user knows well), instead of C/posix. What if, instead of having to remember bash’s for loop syntax (or xargs), you could just write .map() and iterate over lines/execute commands that way?
Transforming the shell is a great opportunity for a new project, and as you said should probably should be done in Python or JS. Lua is a good option as well for its elegance, though it's a bit less popular than the others.
As for Z.lua, this is an interesting project and I really like seeing modern languages enriching what we do every day on the command line. Very cool!
Why Create a New Unix Shell? http://www.oilshell.org/blog/2018/01/28.html
I took a very long route to doing it though, first implementing OSH (which will run existing shell scripts, which is close to done, but slow) and then Oil (a new language, not implemented).
This is because POSIX shell/bash has a really strong network effect. zsh and fish are better in many respects, but people are locked into bash because that's what "everybody else uses".
It's analogous to how C has a really strong network effect because e.g. every kernel is written in C. So C++ became a popular language in a large part because it's compatible with C. (Though the analogy breaks down because the Oil language isn't a superset of bash, so we can break free of legacy. The project is though.)
In my mind, I think of Oil as a mix between Python and shell, with some elements of Ruby and JS (blocks, function literals, etc.)
And yes, network effects are so important—your C++ example is spot on. One recent example that comes to mind is the adoption from python2 to python3, showing just how big a barrier those network effects are to adopting something incompatible.
It exposed the slowness issue though. I have a plan for that, but it's a bunch of work.
It will help me if people try it and give feedback -- I'll give details in the post.
I definitely agree about Python 2/3. I think you basically can't overestimate the amount of inertia a popular language has. I'm sure there will be patches floating around for Python 2 for a long time after the core team stops maintaining it next year.
Another example is all the compile-to-JS languages. I think TypeScript is the most popular in part because its transition path is the easiest, i.e. it's a superset of JS.
Or you can subscribe in one of these places :)
The number of times I've used a for loop in shell is roughly equal to the number of times I've had to look up how to use a for loop in shell. Relatedly, I generally switch to Ruby when my scripts get complex enough to require control flow.
This is my attempt so solve a small personal pain point in the shell while trying to keep track of various commands: https://github.com/bonidjukic/aka
It definitely comes nothing close to the Z.lua, but the enjoyment generated from getting to write the thing which optimizes even the tiniest parts of your daily tasks is quite wonderful.
But I expect for adoption it'll need to be a more widely-used langauge
They could have just used F# as the new shell language, the interpreter already exists, the syntax is clean and concise, etc.
But you can! The dot is spelled '|' of course, and 'map' for historical reason is called 'xarg'. There is no real lambda syntax nor closures though, everything is stringly typed.
But it’s fun to use, still. I might come back to it a few versions down the road.
Currently shell code is limited to one language. If you have bash, you can only do bash. But we could imagine a shell that understands webassembly, and hence we can use it with many languages. We could even imagine tools to inline webassembly generation, so that you can code directly in your favorite language in the shell, and it gets transparently compiled and executed.
I use a combination of GUI and TUI all the time. I like my mouse. I wish there were a terminal that behaves more like a file browser, with back and next button, an address bar, and a way to temporarily switch the view from text to icons to select files with the mouse then enter a command for thoses. Some file operation should also be available from a right click on a file path.
Not every shell lover wants his hands on the keyboard all the time.
We might even compile directly to native code, nay, automatically identify the format and pass it to the appropriate interpreter, and then provide an interface to call arbitrary modules that conform to a particular interface―
... and then expose it as a library function so the shell can use it, let's call it execve().
Oh man, you are gonna love this. In Plasma 5, hitting F4 in a file browser toggles a pop-up terminal in the folder you're in. As you navigate the directory tree, the shell actually follows you - and vice versa when you type `cd`! Working directory is always kept synchronized. If you split the file manager into a two-pane layout (another great feature), the terminal working directory follows whichever pane has focus. And of course you can drag files into the terminal and it pastes the path.
It's basically exactly what you asked for, and it is indeed fantastic. Plasma 5 is fantastic. Light years ahead.
But unfortunaly they discontinued it, which is a shame because indeed it's a nice feature.
However there is a limit to this conccept:
- no quake mode
- no horizontal split
- no way to manipulate the terminal shape and sessions from another terminal
- keyboard shortcuts and config options are file explorer first, terminal second
- no way to hide the explorer part. I don't want the files to always be here.
- no shell integration. I want some status such as the git repot state or the virtualenv name to appear in the bar, and the shell to remove it automatically from the prompt if I'm in GUI mode.
All in all, I'd prefer a specialized terminal, that have some file explorer features, and a real file explorer on the side, without the terminal features.
But yes, it is a good start.
For a shell as an interactive command interpreter, I don't see the point of it. For the shell as a programming language, you can already use any other language in its place.
> I wish there were a terminal that behaves more like a file browser, with back and next button, an address bar, and a way to temporarily switch the view from text to icons to select files with the mouse then enter a command for thoses. Some file operation should also be available from a right click on a file path.
This sounds more like the domain of a shell than a terminal. A terminal has no concept of what's generating the stream of control codes. A shell on the other hand could easily implement functionality like this.
I also don't really understand why you'd want more than one language in the shell, but I guess if you wa
I've also written an autojumper (in rust) after being frustrated by performance issues in other autojumpers, among other reasons.
For mine, I've gone to the effort to write a test and benchmark harness, and I also ended up with about 10x faster than fasd/autojump.
I'm excited to benchmark pazi against z.lua, and if needed take some ideas to be even faster :)
I already benchmark pazi against autojump by itself, so adding in a benchmark against autojump-rs will basically be drop-in, and I do expect pazi will still be faster, but I'll see.
At a glance, the differences I see are the following:
* pazi has extensive integration tests and benchmark code
* pazi doesn't fork processes into the background like autojump's shell scripts do (and therefore autojump-rs does), which avoids a few races
* pazi's interactive selection (z -i) makes use of the terminal's alternate buffer to avoid blowing away your scrollback, which most jumpers don't bother with.
* pazi has a cooler name in my opinion :)
These differences are not all that major. pazi is definitely short on some features, and I don't blame you if you wish to use one of the others. I'll keep working on pazi because I like using it and I'm having fun working on it / optimizing it / testing it.
The cross platform story for rust isn't as seamless as an interpreted language like lua, but it's still good enough that it's not a huge effort to support another platform or shell.
> cross platform/shell support
Shell support should be fairly similarly easy or hard between most autojump implementations.
Also when I do `z doc` I expect to go to `~/Documents` like in autojumper, but in z.lua it jumps to a subfolder `~/Documents/projects`, which is annoying. Finally it does not seem to cycle through all matches when I repeatedly do `z proj` -- I like that about autojumper.
eval "$(lua /path/to/z.lua --init bash enhanced once)"
I can confirm z.lua feels really snappy compared to autojumper after disabling the git related stuff! :) So now I should find a way to make the git prompt faster.
You want to install something troublesome, and the shell suggests a solution command- or even several solution commands, mechanical turked away from other linux users.