Regarding autocompletion

[benbenbang]

Hi there,

Since we are using cobra, any plan to use the completion to help with the flags? I can do a PR if it's ok for you :)

Thanks

[kardolus]

Hi @benbenbang.

I think that's a great idea. I have been struggling with auto completion. Please take a look at this branch (and feel free to make adjustments, or to submit your own PR):

https://github.com/kardolus/chatgpt-cli/blob/d2f1804930d0587563354a7d71f5ea612dd37dfc/cmd/chatgpt/main.go#L59

The problem I am having is that once I add 1 command (completion), chatgpt no longer works as intended:

 2024-02-20 09:12:21 ⌚  Guillermos-MacBook-Pro in ~/workspace/chatgpt-cli
± |main {4} U:1 ✗| → ./bin/chatgpt what is the capital of denmark
unknown command "what" for "chatgpt"

 2024-02-20 09:12:52 ⌚  Guillermos-MacBook-Pro in ~/workspace/chatgpt-cli
± |main {4} U:1 ✗| → chatgpt what is the capital of denmark
The capital of Denmark is Copenhagen (Danish: København).

I love Cobra/Viper but also really like to use the CLI without commands :) -- what would be a good workaround for this?

[benbenbang]

Hello @kardolus,

Sure, let me take a look tmw morning

[kardolus]

If we can't get the command to work I suppose we could always set up a script ourselves.

[kardolus]

Here's a completion script just for flag completion: https://github.com/kardolus/chatgpt-cli/blob/completion-script/scripts/completion.sh -- not sure if it works across different shells though. I can confirm it works on GNU bash, version 3.2.57(1)-release (arm64-apple-darwin23).

[kardolus]

Merged it into the main branch: https://github.com/kardolus/chatgpt-cli?tab=readme-ov-file#command-line-autocompletion

I am open to suggestions for something better though :).

[benbenbang]

Long day :D Anyways, we can make it simple or make it "complex" Maybe I am wrong, or say there's no one way to do it. But I can see from the current implementation that you are trying to circuit the command by flags. Hence, the command is not that "official" structure.

I have a proposal, but it might become a project instead of an issue 😅

the idea is mainly to

• convert some flags to commands and extend the utilities
• group the subcommands
• set prompt, for example, as a subcommand
• regarding the prompt part, two choices:
  • exec the root command directly, if fail, pass directly to default command, which is prompt
  • pre-parse the args and flags in the root, if there's no "arg" in the implementation (could be a list of string), then we fall over the command to prompt
  • either way, we can keep it as "it is", i.e., we don't need to use command to invoke prompt

and the grouping can be something like

$ chatgpt
chatgpt-cli is a cli to...

some other desc 

Enjoy! :)

Chat
  prompt  Run prompt command

Settings Commands
  completion    Generate completion script
  confg  Run config settings command

Other Commands:
  update        Update your chatgpt-cli version
  version       Check your chatgpt-cli version

Use "chatgpt <command> --help" for more information about a given command.

like so

And for the config subcommand, I am thinking we can extend it to

• generate default config template
• verify config also, we can keep the flag to take different config paths in the runtime

And so on.

I'd like to stop here first to hear your feedback. Otherwise, I still have some idea can be implemented 😆

[kardolus]

Thanks @benbenbang! I think the way you are describing the use of cobra/viper makes total sense. However, I really like the current UX (having the simplicity of chatgpt <message> without a command). Maybe the current flag completion script is the way to go for now? It's not perfect but it works. I have been enjoying flag completion since yesterday :). Great idea! Thank you.

[benbenbang]

Alright, nvm, then.

I will create a PR and keep it as it is, and just add a flag to generate a completion script

[benbenbang]

If you don't mind, I can create another PR to add a makefile to run the test / build the binaries?

[benbenbang]

Thanks @benbenbang! I think the way you are describing the use of cobra/viper makes total sense. However, I really like the current UX (having the simplicity of chatgpt <message> without a command). Maybe the current flag completion script is the way to go for now? It's not perfect but it works. I have been enjoying flag completion since yesterday :). Great idea! Thank you.

My suggestion won't change how you prompt since the default command is the only prompt command. It will be triggered when encountering execute error or use chatgpt prompt xxx

[kardolus]

Thank you! This is great. I like Makefiles, so that sounds good to me. Looking at your PR.