Upterm – A terminal emulator and interactive shell based on Electron
github.com
github.com
Author: I wrote something for people who care more about features than performance.
People who care more about performance than features: I am personally offended that you created this.
But I do notice latency in hyper.js, and it just makes it feel very "dull", and that's what I think a lot of people know of electron terminals
plus the competition from native apps is a bit absurd in terms of performance, with most of the serious ones either rendering asap or on the next frame (when vsync'd)
I would of tested this one, but no windows build, and I'm not using a linux install atm
editors don't roundtrip to a separate application through multiple buffers (bash, zsh, fish, or whatever) to decide if they need to show a color or letter or nothing
For some reason, there aren't so many, except - 9term from plan9, - tilix or tylix, which has customisable actions for regex patterns - URL, mostly http, in vte based terminal
I wrote a simple patch for st that process some escape sequence from the shell to have the terminal aware of its shell current working dir (ocs7). Using this simple hack, one can just cd & ls & select "foo.pdf" & right click to have it handled by one ´s plumber (open, xdg-open, plumb ...) or even right click "path/to/file:line:column" from a log to the editor.
This feature is so simple, so addictive in plan9/acme, that I don't really get why it isn't offered by most terminals.
It sure results in exciting and interesting bugs, if you look at upterm's issue tracker.
Except, you know, all of the folks that say that they like Atom or Visual Studio Code: https://insights.stackoverflow.com/survey/2017#technology-mo...
(Atom and VS Code were specified by 20% of voters in each category).
https://danluu.com/term-latency/ (Previous discussion: https://news.ycombinator.com/item?id=14798211)
Recently there was this terminal performance test https://danluu.com/term-latency/
The author says that the most important things are typing delay and how quickly you can Ctrl+C a command, not how much text it prints per second. Upterm is good at both.
I dropped those because they're magnitudes slower than urxvt. You managed to make a terminal emulator that's even worse?
Not every tool fits everyone’s needs. Some tools might only be useful to a few people.
No need to be rude about it.
Considering I dropped those because I consider them slow and bloated, good luck with this Electron monstrosity.
You can have my RXVT when you pry it from my cold, dead hands.
There are other aspects of terminal performance that are very important, and you may have meant to point towards those, but as it is this comment just reads as bikeshedding. "Performance is (a|the most important) feature" refers to performance in areas that matter.
* And yes, I know that yes(1) can be used for performance/stress testing of systems by outputting endlessly. Doing that to a human-consumed TTY seems like an example of spacebar heating: https://xkcd.com/1172/
I don't want OOM killer to murder everything just because I accidentally catted a 1 GB file to stdout rather than another file. Or accidentally catted the .iso rather than the .iso.asc. Nor do I want to wait 5 minutes for the terminal to render it all.
Sure, accidents don't happen every day, but it's a massive inconvenience if it does. It was enough to make me abandon all feature-rich VTE based terminals in favour of rvxt.
This isn't the same as crashing/hanging your terminal's command-output parser (depressingly easy in Upterm) due to volume; when that happens, the process writing tons of data (cat my_huge_file.dat) will fill up its STDOUT pipe and usually block until the situation is resolved. I think that stability-in-the-face-of-large-output behavior is what you're referring to, rather than OOM killing, in which case I agree wholeheartedly that the behavior of terminal systems (and remote shell systems, etc.) is distressingly bad in the face of large-data influx situations.
The only caveat is that I'm not sure I want to have another electron app chewing through 15% of my CPU...
i.e.
`cd $THAT_ENV_DIRECTORY`
however I always get
`The directory "$THAT_ENV_DIRECTORY" doesn't exist. `
although clearly it does as it is autocompletion suggested
My differences: (1) The shell is in pure js. It's a thin wrapper around shelljs but I had plans to move to a builder type syntax, eg. fs.cp().f("/tmp/foo").t("/tmp/bar").x();
(2) I made an api so people could quickly code up graphical output for commands.
Just wondering if you'd be interested in integrating a feature like this. Do you have a chat channel?
First of, I really like the idea of this project despite normally being a vocal critic against Electron. However I think you bit off too much trying to write your own $SHELL as well as terminal emulator. The result is rather uncomfortable to use because there are shell bugs and incompatibilities compounding the usual expected bugs with any new terminal emulator. For example it took me 5 minutes to just cd into a directory because env vars weren't being recalled, the ~ variable wasn't being recalled, and auto-complete wasn't keyboard optimised (I'd arrow down and hit enter but it would crop the line rather than select the auto-complete entry).
So the next thing I did was to start my preferred shell and I then ran into a whole other series of bugs due to the terminal emulator being tailored for it's own specific shell. In fact the bugs there were so severe that it renders the term virtually unusable on any of the standard shells nor the more esoteric ones I contribute to.
One thing I was pleasantly surprised about though was how well it supported ANSI escape sequences - even ones that aren't part of the formal standard (such as the iTerms true colour sequences). But the bugs surrounding the terms handling of readline (and similiar) really feels like many of the basics are missing despite having some cool advanced features already.
Overall, as much as I'm ideologically against Electron, I can see the advantages of using it in this domain because it allows the ability to include rich content in the command line. It might be nice to get to the point where the command line can offer as much rich content as a web page (something I'm working towards with my own project in fact) - in which case Electron (or similar tech) seems the obvious choice for building a terminal emulator. However upterm's dependance on using your own implementation of Bash (I think it's a bit disingenuous calling your shell "Bash" but that's your prerogative) makes this highly frustrating compared to other rich terminal emulators already out there. And the inability to run your own shell within upterm makes this inflexible for anyone wanting to do something a little more indepth than just cat'ing a JSON file (even running a DB command line tool like mysql or redis-cli would be impractical in upterm in it's current state. And while I didn't try SSHing into remote boxes I can see the same issues happening there too).
I don't know the age of this project but I think it does have real potential. However if I was to ask for anything - and I appreciate this is your baby so I have no entitlement what-so-ever - it would be if you could offer a $SHELL'less version for power users.
edit: Also, I've noticed commands that fail don't appear in the upterm's history. Was this an intentional decision? I appreciate there are arguments for and against only including successful commands but sometimes you'd typo something and the inability to up-array and edit my failed commend added to my frustration. If this was an intentional design decision then is it worth including the last command regardless of success and then excluding it from all subsequent placements in the history?
We do utilize a login shell for executing commands, but we only support Bash and ZSH. So, it's a real Bash. However, there are some complications with environment interpolation and built-in commands; it's a known bug and we're working on it.
Unfortunately, I don't see how to allow using custom shells while providing a better command entering experience (i.e. zle or readline replacement). The core issue is that shells' interface is too primitive: it only consists of three streams with no per-job separation or anything; I wish Bash had a programmatic or REST-like interface.
With regards to the shell input: I don't know how you are currently detecting the $SHELL prompt (and thus when you can do your auto-completion) but maybe you could allow users to specify hooks for what to watch for to determine the different between the $SHELL and other tools, based on the $SHELL prompt (possibly using $PS1 env var?)
For example you could define that $SHELL's prompt would be:
SHELLPROMPT>
As a side note, the project I'm currently working on is a $SHELL that supports defining data types to STDOUT and STDERR. The issue I have is that those byte streams are typeless and scripts written in my $SHELL needs to work with existing POSIX utilities and potentially on other $SHELLs with untypes streams. Plus I wanted people to be able to type those streams from other languages as well if they wanted and for my $SHELL to understand that. The solution I've decided upon was for an ASNI escape sequence to surround a string with the typed information and for this to only be a included if an environmental variable was set to the name of my shell. This is how I'm working around limitations with POSIX compatibility while still offering a richer environment to develop in when my $SHELL is running (I hope that makes sense).
In fact it sounds like we are attempting to solve similar problems but from a different angle - with myself targeting the $SHELL while you are targeting terminal emulation (which it probably why I'm warming to your project).
I'm not going to criticize your approach as it totally makes sense in terms of building a minimum viable product but I do think you're going to run into a lot of problems with that approach as your project gains traction as it's very different to how a terminal emulators traditionally work. Plus you'll potentially spend a lot of time reimplementing Bash features like it's builtins, scripting features and the multitude of parsers it runs (depending on if you're interested in having a REPL environment like you would normally would with $TERM + $SHELL).
That all said, it looks like you're handling pseudo TTYs decently, which is an easy thing to get wrong (I say that from experience as I've screwed up PTYs in the past!) but also probably the most important part to get right, in my opinion.
I know there is a lot of clash around electron apps these days, and I'm the first one to dislike the waste of resources. But maybe in 5 years electron and hardware will improve so much it won't matter anymore. So if it allows you to experiment with new stuff, keep going.
And good luck.
I don't know any of the above mentioned languages myself, so beyond that outsider perspective I can't form an opinion on it, I am just curious why this choice was made.
With zsh's built-in Ctrl+R it's at least possible to reduce list of suggestion and then travel up the history in those suggestions. While the more powerful fzf will give you the list of commands you've run in a selectable dropdown.
Upterm I think could be eliminating entire classes of tools which sometimes are unnecesary complex. Server automation could be part of pre-saved playbooks which are saved to your terminal history and can be easily brought up and run according to rules. Curl arguments could be arranged in a way where they are actually managable. Parsing --help and man pages (which is possible, there is a workable implementation in fish AFAIK). Upterm is in position to take on that functionality, but right now it's not much more than already mentioned hyper, but with custom handled prompt line.
Also, we recently integrated monaco-editor by Microsoft, so editing is much less clunky than it was before.
Autocomplete that was a little bit laggy on zsh simple works super fast on fish. Other features are also very nice. I think people who might be interested in this, might find fish interesting.
Have some patience. It will eventually improve to the point where it is very low overhead and provides much richer experiences than we have in the terminal today.
Nobody is forcing you to use it today. Just be supportive because we want it’s future.
This is a very optimistic of framework/technology performance.
> Nobody is forcing you to use it today. Just be supportive because we want it’s future.
Using web stack for desktop application wasn't a technological choice. It's a much more a reflection of the inability of our industry to cohesively innovate
it would be awesome if you had some way to support more commands - especially if done in a more automatized way.. parsing the manuals (?)
This means that while it is easy enough to pull flags from the man pages, offering usage suggestions is something too tricky to handle in an automated on-demand fashion.
I did toy with the idea of starting a new project porting man pages to a new document format designed specifically for utilities parsing (eg a JSON format with standardised fields). However even that would be a massive job to maintain without the support from other developers upstream.
But I'm open to other suggestions if anyone has any?
Hyper + zsh can.
I found Conemu clunky (maybe I needed to spend more time configuring it, but out of the box it was fairly ugly, and borked out under some curses applications and when using tmux), PuTTY/KiTTY to be lacking some important features (like tabs) -- ditto cmd.exe.
IMO there's a real gap in the Windows market, be that a new terminal emulator entirely, or improvements to Conemu making it more intuitive and user friendly.
However, this is all just my opinion and maybe there will be a load of users, me and people who share my opinions will slowly descend into obsolescence.
* Make a decision you want completion.
* Press tab for it to appear.
* Guess when your token is precise enough to be unambiguous or else cycle through results.
Not to mention fuzzy search.
It may sound as not that much of an overhead, but it's so much slower than what I have in, say, Intellij IDEA.
Shouldn't be too hard to modify it to display completion data instead of entries from your history.
I prefer the other plugin though because it's less obtrusive.
Haven't used a Windows machine for a while but I remember there were bunches of different VS Runtime version installed on my PC before. Some app installers came with a runtime installer, but if the required runtime version was installed it would just use that one. I think this might somewhat reduce the installed app size for Electron apps as at least some stuff got reused.
npm install -g electron
electron ./myapp
npx electron ./myapp
(It still won't share as much runtime memory as you might want, but it would reduce install footprints at least.)
I'm surprised that I haven't yet seen a fork of electron-windows-store [2] that makes use of the UWP webview, as that seems an obvious enhancement to me. The Electron remoting between webviews and Electron code is already setup well enough you should be able to run the Node app as a UWP background app and use UWP remoting.
[1] https://github.com/pojala/electrino [2] https://www.npmjs.com/package/electron-windows-store
as said terminal emulator should do one thing lightly. Tough the nice interface still made me download it.
Try building the same stuff with smtg else than electron ?
No thanks
I'm not against electron app, in fact I think it's a really good idea. But saying it's the only good ui framework is disingenuous.
[...] Electron is the only decent cross-platform UI framework
What about Qt, wxWidgets, Tcl/Tk, or even Kivy?
So, in terms of design, there’s basically no major difference there.
Battle.Net ? https://i.imgur.com/VCNEhgm.jpg
both are made with Qt
Of course it depends on where your interests lie but here's some examples: VLC, Spotify (Linux), old Opera, Krita, Google Earth, Tesla Model S in-car UI, webOS, SailfishOS, AMD's Radeon Software Crimson driver tool, and many others including what's already mentioned, Battle.net.
JavaFX? Also a nice cross-platform UI framework, supports JavaScript, and is also better performant than Electron.
Also, both support HiDPI by default, and can be themed with CSS however you want them (in the exact same way that you could theme Electron apps)