Supporting shell completions (spec)

[aDotInTheVoid]

Is your feature request related to a problem? Please describe.
Ex. I'm always frustrated when autocomplete doesnt work.

Eg:

• touch TAB doesnt work
• ~/pre TAB doesnt work (should expand to ~/prefix
• ninja ins TAB completes to ninja install_manifest.txt, not ninja install
• cd rustd TAB completes to cd rustdoc/, not cd rustdoc, where I can then specify I wanted to go to rustdoc-json, (in rustc src/test dir
• man cla TAB doesn't comlete (zsh shows clang clang++ clang++-12 clang-12 classes.conf)

Describe the solution you'd like

I'd like to be able to deligate to the underlying shell to provide autocompletion.

[zachlloyd]

Makes sense as a request.

Our current approach is to try move this logic into Warp by building out our own completion grammar and library, with the rationale being:

1. Shell autocomplete is pretty hard to configure
2. You lose most of your completions when you go to remote boxes unless you've also configured completions on them
3. The current shell completion specs are (in our opinion) pretty hard to write new completions for

That said, we are missing a bunch of context right now from the shell which is making our completions worse than they should be. And we just have a number of bugs. I think we will see how far we can get without delegating to the shell, and add that as a fallback if we can't make it great.

Thanks for the report.

[d12bb]

Points 1 and 3 just aren't true for fish, writing new completions is pretty straightforward. And 2 is easily solved by syncing one's config, which is often already set up.

Rationale for the opposite however: Completion for programs is often shipped with the program for major shells, but probably won't catch up for a Warp until it has a laaaarge userbase.

In my book, this is exactly what #2: Keep the power, but fix the UI. means. Why reimplement or worse cut (in case of missing completions) the power here instead of relying on the shell and just fixing the UI?

Btw, is there already a way for the user to add completions to Warp?

[rocketraman]

In addition, many if not most commands in Linux come with completions out of the box. I rarely configure my own completions, and yet my experience on Linux with completions on my usual terminal and shell (Yakuake and fish) is far better than my experience with completions on Warp. This contradicts all 3 of the points made by @zachlloyd .

[aDotInTheVoid]

Maybe a hybrid aproach works best, where if you know about a tool (npm, etc) you provide native support, but if not, you can delegate to the underlying shell (ninja, r2, rg etc).

Some of the more basic stuff (starting with ~ or /, completing touch) should definatly be fixed in your layer

[ToontjeM]

Shall-defined functions should also be included in the auto-completion. If i say
myfunction () { bla bla }

The myfunction should be available in auto-complete. Just like normal zsh.

[zachlloyd]

Yup, we will explore that. We actually started with an approach of delegating the shell, but it did introduce extra latency. As a fallback though it might be acceptable. Appreciate the feedback.

[simjnd]

Is your feature request related to a problem? Please describe.
Ex. I'm always frustrated when autocomplete doesnt work.

Eg:

• touch TAB doesnt work
• ~/pre TAB doesnt work (should expand to ~/prefix
• ninja ins TAB completes to ninja install_manifest.txt, not ninja install
• cd rustd TAB completes to cd rustdoc/, not cd rustdoc, where I can then specify I wanted to go to rustdoc-json, (in rustc src/test dir
• man cla TAB doesn't comlete (zsh shows clang clang++ clang++-12 clang-12 classes.conf)

Describe the solution you'd like

I'd like to be able to deligate to the underlying shell to provide autocompletion.

Completion using variables is also broken, e.g. ls $HOME/Dow expands the variable and autocompletes to ls /Users/simjnd/Downloads but it doesn't work in Warp

[jklein24]

Throwing my +1 into the mix on this that the autocomplete issues definitely hinder my ability to use Warp as a primary driver right now. I also source an open source git_completion.sh script to get autocomplete for git branches and commands, but that doesn't seem to work with warp either. Would definitely like a fallback to native autocompletion when warp has no suggestions.

[maximkrouk]

I think the lack of make autocompletion is related too.
Otherwise I may open another feature-request to support make 🙂

[davecombs]

While it's a beta (and pretty impressive so far - I find very little of what it actually does that I don't like a lot!), not having some sort of command completion at least for filenames working is really going to make things difficult to continue. Maybe my setup (bash) is interfering, but even just '. ~/.bash_al' doesn't do anything - no list of files, no default pick, nothing. I have no choice but to retype in the file name completely. Anybody reported anything in bash that might interfere with what you're building (so I could at least get the minimal file completion?)

[nesl247]

I definitely think that the shell needs to take precedence here over any autocompletion by the terminal. I have my shell configured exactly how I want it to be, with the autocomplete that I want it to have. Regardless of if I'm using Warp, iTerm2, Terminal.app, or any other terminal, I don't want to have a difference experience. The terminal should enhance the shell, not replace it.

[ned-si]

I definitely think that the shell needs to take precedence here over any autocompletion by the terminal. I have my shell configured exactly how I want it to be, with the autocomplete that I want it to have.

+1 for this!

Having default autocompletion included in the terminal is fine, but I think that users must have the ability to customize it exactly the way they want is very important.

Being able to simply use what was already made at the shell level would be best, but another acceptable possibility, imo, would be to be able to customize it with a config file for Warp.

As someone else mentioned already, this is the single thing that prevents me to daily test Warp.

[dannysteenman]

Same here! This is preventing me from switching to warp from iterm

[ToontjeM]

+1 So far I like Warp a lot, but configurable completions is a must have.

[arunkumar413]

I think a JOSN like command structure would be of great help. More info here: #585

[elviskahoro]

From @arunkumar413

Hi Team,

Most of the Linux commands don't have a common structure. Each command has its own options and flags. I recommend a JSON like structure for the commands and their options.

For example:

{
    commandName: "ls",
    options: {
      all: true,
      author: true,
      humanReadable: true,
      recursive: true
    }
  }

This JSON like structure should be parsed into the regular Linux commands and executed. We'll have a terminal feature like text editor that takes JSON input, interprets the JSON like structure into regular commands and outputs the result.

This feature would make the documentation, linting, code completion easier. Each command option would be self explanatory and easier to understand.

Note: This wont be a complete replace of the existing command line but just an addition to the existing command line which takes a JSON structure as a command.

[neelmehta247]

Had a ticket related to this discussion here: #622.

I was having issues using commands like kubectl and skaffold where the tab autocompletion would simply do nothing (presumably because there wasn't a file in my working dir with the same prefix). I think in cases where the command is unsupported, delegating to shell is a good way to go, even if it adds latency.

For kubectl for instance, the shell completion fetches suggestions from the k8s API I believe to give autocompletions that wouldn't be found on machine. I can see how this would be tricky to build a custom solution for in Warp so falling back to shell could be a good stopgap.

[elviskahoro]

From ChrisVarenChris via Discord: https://discord.com/channels/851854972600451112/852251449068486773/900932342653255700

i assume a shell's native completion system is just incompatible, so you need to write new ones for Warp? At the moment, this 100% what's stopping me from using Warp for the majority of work. A crowd sourced repository of open source completions wouldn't solve it since we have completions for internal tools that are used all the time. Though most of these completions are autogenerated from getopt-style help texts so just the ability to have custom user completions would be enough. (I believe fish shell auto-magically generates completions in a similar way)

[jesus-villalba]

It doesn't autocomplete functions defined .zshrc, e.g.
function rstoserver { rsync $1 [email protected]:$2 }.
Is there a work around?

[elviskahoro]

From blitz2145 via Discord: https://discord.com/channels/851854972600451112/906269689116770304/908912605232529418

Perhaps this one is out of scope for the warp terminal in the short term, but the UI could build on a shell that supported this theoretical completions protocol. How about something like the LSP (Language Server Protocol) but for autocompletions, perhaps cli programs themselves could be invoked to output completions for their own commands, rather than having to build stuff like fig (which is nice, but cli programs themselves should take on the responsibility for completions), etc. The Oil shell actually has a great design/architecture for how to do this, https://github.com/oilshell/oil/wiki/Shellac-Protocol-Proposal

[jpbochi]

Related issues: #2129 and #2173

[jpbochi]

Another one: #1811 [Feature] Bypass/Disable Warp TAB-completion and use shell TAB-completion

[sandzhaj]

Any news from the developers about that? Are they planning to do something with it?

[jpbochi]

According recent comments by @elviskahoro on #1811 [Feature] Bypass/Disable Warp TAB-completion and use shell TAB-completion, there will be "EXPLORATIONS as to what we will try to do in order to improve this experience will start probably next quarter."

[c4shdev]

Warp is great, but autocomplete still doesn't work as expected. 2023 is already over and I'm still waiting... 👎

[tybritten]

Yeah this is a huge blocker for me. I was using iterm2 + omz + p10k + Fig and warp could do most of it all in one, and has a lot of other great features, but the lack of completion support is a killer.

[c4shdev]

Warp is great, but autocomplete still doesn't work as expected. 2023 is already over and I'm still waiting... 👎

[nathanblair]

This decision to build a home-brewed autocompletion system is perplexing to me. It's a massive reinvention of multiple wheels, the wheels being bash, zsh, fish, etc.

Maintainers of tools like kubectl, task, and many others, already release their tools with appropriate shell completions. So at best, I would expect Warp is capable of staying on top of every single tool that already releases with its own autocompletions for various shells and fully up-to-date with what those auto-completion definitions are.

...that's just not sustainable. I get the desire to have Warp able to handle auto-suggestions in an ssh session but...this is just not the way.

My only current workaround for this problem (when it irks me enough that I need a solution for it so I can just be productive) is to run my shell as a sub-shell of the Warp terminal shell. And ultimately it looks like there's an endeavor to remove that workaround by having Warp take over sub shells as well. I give up blocks but at least I don't spend 10 extra seconds every minute refreshing myself with $command --helps on the multitude of tools I need to run in a single day.

[aexvir]

this is also my biggest issue with warp, I've spent a lot of time building my own completions, built on top of fish+fzf, which offers me a much better experience than warp, not everyone may like it but I do and it's something I use hundreds of times per day, so I made it work exactly how I want

just please let me disable the tab completion in warp so my shell can handle it, feel free to build your own system as much as you want, but don't force me to use it

[tettoffensive]

For me I've loved transitioning to warp except that in iTerm2 (with oh my zsh) git completes my remote branches when I type git switch or git checkout, but warp only completes local branches.

[ilrudie]

related #3801
TLDR: Some ideas for how this might work

• Option to set Warp's behavior to nothing/passthrough
• Option to set a keybinding which passes down
• Keybinding to toggle behavior between "native" and "warp"

[thedodd]

This sounds optimal.

[ToontjeM]

sudo systemctl status servi<tab> doesn't find servicename.service

[flipphillips]

Good - I thought I was on crazy pills when I first started playing with warp. It surely seems like some enable/disable logic, or prioritizing logic would be useful here. It's the main reason I use zsh

[welcoMattic]

I was super enthusiast about Warp on Linux, I installed it, save one Configuration to have a one-click workspace ready to use, started to type a command, hit TAB and 💥 I missed my ZSH completions.

Instant uninstall of Warp.

Too bad to get rid of a basic standard usage as shell completions.

[flipphillips]

As a very old-time keyboard jockey, I have to say that not just ‘old’ but ‘all’ habits die hard. Lots of people propose ‘paradigm shifting’ UI, especially for typing and other mechanical tasks (musical instruments being an interesting side-study). But if they even -hint- at the support of an older way (e.g., @welcoMattic ’TAB’) and it’s not supported, uninstallation is an almost certain side effect.

Have a look sometime at Jef Raskin’s early stuff like FLOW and friends. Eventually, we got the Macintosh out of it, but it was a long and crazy road.

I ‘grew up’ a Gibsonian, a theoretical position that was heavily adopted by Don Norman. Things in our world behave in particular ways and when they don’t work as expected (black ice) chaos ensues.

So - saying this, does some of the strategy in this terminal program hold some potential for making my daily battle with the keyboard easier / more pleasant / etc? Yes, most certainly. But - does achieving it by ‘breaking’ many people’s expected behavior seem like a good ides? Maybe not.

[ctrlok]

@flipphillips As a person who always tries something new (I switched and learned Emacs and Lisp, vim and vim script, neovim, vscode, sublime, idea and others; I tried every new editor and terminal and shell), I want to add that the problem is not only in unexpected. For guys like me, it's good because it's something new to learn and try.

The main problem is that autocompletion does not work for some new, old, and updated apps.

My kubectl can't autocomplete pod name or container name, my kubie dosen't work, and the newly installed procs app is not supported by Warp, as well as eza and zoxide (which is replacing for cd command) and new flags from gcloud command and I can continue that list, which already pretty big based on a single day usage.

However, a bigger problem is that Warp will not work for any new commands and improvements that may happen in the future.

While I adore Warp blocks, it feels stupid for me to give up on many existing tools and new tools that may be created just because Warp developers think they can do the work of thousands of developers who independently work on shells, tools, and completions.
It's just a bad deal.

So my only option is to wait until Warp developers realize that they are trying to do a job bigger than themselves and will start to utilize the free work of other engineers around the word or when other terminals will provide the blocks feature (and it may happen pretty soon, while blocks are already implemented by IDEA EAP terminal and it's only a matter of time when this unique feature will be not that unique)

[szamanr]

loving warp, but not supporting fish completions is a big issue. for some commands there is no auto-complete (e.g. z from zoxide), for others there is auto-complete, but somehow it's broken (e.g. gt from graphite shows some commands, but not others 😕).

i think an ideal solution would be to both:

1. mark warp auto-complete as experimental/beta and allow disabling it via config - when disabled, the shell's auto-complete would be used.
2. when warp auto-complete is enabled, it should still use the shell's auto-complete as a pass-through in cases where it cannot auto-complete a certain command.

not sure how complex these tasks are, but in my opinion they are urgent. this is on the border of being a deal breaker for me, and based on this thread and common sense, it's a deal breaker for many users.

thanks and keep up the good work!

[szamanr]

i ended up switching to my old terminal due to these issues, but i miss the code blocks and built-in AI. please prioritise this.

[ctrlok]

@szamanr fyi, iTerm has built-in AI integration (not much people know it because of poor marketing in open source)

[szamanr]

thanks but i'm using linux - looks like iTerm is mac-only.

[sandzhaj]

Nobody uses the terminal as often as people who write CLI utilities for their tasks.

If you do this, you're probably spending more time in the terminal than the average developer running mvn build.

Writing completions for your utilities is a very common situation.

It's a shame that the Warp developers have been ignoring this for years. But this is not surprising. I have had a bug since the first days, with long lines that do not fit into the block and cannot be scrolled to the end (just try writing echo CTRL+V (some certificate in base64) | base64 -d in the terminal - after pasting a line from the buffer you won't be able to scroll to the end to see where you'll be typing | ), which was announced in both discord and Issues

No, thanks.

[vishnup-ditto]

How is this not addressed yet. Isn't autocomplete a very fundamental part of CLI experiences now. Of course I am a bash shell user, who likes fish likes suggestion, I mostly expect TAB completions to work. Is there an issue/open pull request to track this. WARP is great but it feels incomplete without suggestions. Is it possible to build something like this as a plugin/integration?

[maciii]

I was thinking that autocomplete do not work by my mistake in some configuration. What a surprise that this is native "feature". What a disappointment - uninstall follows now 👎

[xilopaint]

Yes, it doesn't work by design. It's unbelievable.

[shanness]

Going on 3 years and Nada. It's such bad form the way the dev team seem to write this off and try to convince us it's not such a problem, or that we should use workarounds.

I'm subscribed to the big pile of tickets about this, but until this is fixed, I'm outta here.