gera2ld / ai.nvim Goto Github PK

When sending multiline prompt to Gemini, the lines are joined without line break.
This is because lines are just joined without appending \n.
Appending \n to each lines would help having a prettier loading_tpl if we include the ${input} in it, and may be, will make the promt more understandable by Gemini.

Hi @gera2ld, I've been trying out your plugin with great results until recently, when all the commands I try to send using gemini are returning with a bad request response:

Gemini

Error: Gemini API responded with the status 400

{
  "error": {
    "code": 400,
    "message": "Invalid JSON payload received. Unknown name \"role\" at 'contents[0].parts': Cannot find field.\nInvalid JSON payload received. Unknown name \"role\" at 'contents[1].parts': Cannot find field.",
    "status": "INVALID_ARGUMENT",
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.BadRequest",
        "fieldViolations": [
          {
            "field": "contents[0].parts",
            "description": "Invalid JSON payload received. Unknown name \"role\" at 'contents[0].parts': Cannot find field."
          },
          {
            "field": "contents[1].parts",
            "description": "Invalid JSON payload received. Unknown name \"role\" at 'contents[1].parts': Cannot find field."
          }
        ]
      }
    ]
  }
}

If it helps, this is the configuration I'm using: https://github.com/davidmh/dot-files/blob/main/nvim/fnl/plugins/gemini.fnl

Hello @gera2ld,
Gemini Pro 1.5 now supports "system instruction", just like ChatGPT does.
We did not leverage it before, but now we can.
I made it here: natixgroup/code-ai.nvim@ec8f0c1
Unfortunately, my fork has too much divergence with here so that it would be too much hard to merge.
I just put this here if you want to implement it.
Regards.

Error executing Lua callback: .../ai.nvim/lua/ai/init.lua:83: Column index is too high
stack traceback:
        [C]: in function 'nvim_buf_get_text'
        .../ai.nvim/lua/ai/init.lua:83: in function 'getSelectedText'

If selecting in simple visual mode, no problem.
This occurs only in visual-line selecting

@gera2ld,
I propose to add chatGPT API to the extension.
Implementation is very similar to Gemini's one, this wont be a lot more lines of code.
We would display Gemini and chatGPT in the pop up, like this:

Gemini says:

Gemini answer
Gemini answer
Gemini answer

ChatGPT says:

ChatGPT answer
ChatGPT answer
ChatGPT answer

There will be feature flags: in the config, users will be able to enable or disable one or another.
Note that it is not a perfect thing, but you know is software development, improvements are incremental ;-)

Processing API response can be a big chunk of code, because the JSON response can have a different structure depending on Gemini feeling.

Let's not concentrate processing that response in the callback lambda function: let's split it out.

how do we resize the window? I don't think we can resize it?

When AI answer contains %, that caracter is silently ignored.

Lets take as exemple this Symfony services.yaml file:

# This is a services.yaml file in a Symfony project.
# In ../.env file, I set MIHAMINA_ID to 1234.
# I want to set the parameter mihamina_id to the value of MIHAMINA_ID.
# Modify the parameters section of this services.yaml file 
# to set the mihamina_id parameter to the value of MIHAMINA_ID.
parameters:

services:
    _defaults:
        autowire: true      
        autoconfigure: true 

    App\:
        resource: '../src/'
        exclude:
            - '../src/DependencyInjection/'
            - '../src/Entity/'
            - '../src/Kernel.php'

If I prompt AI with the whole file, the displayed result is :

parameters:
    mihamina_id: 'env(MIHAMINA_ID)'

services:
    _defaults:
        autowire: true      
        autoconfigure: true 

    App\:
        resource: '../src/'
        exclude:
            - '../src/DependencyInjection/'
            - '../src/Entity/'
            - '../src/Kernel.php'

We notice the % are missing.

It should be:

parameters:
    mihamina_id: '%env(MIHAMINA_ID)%'

services:
    _defaults:
        autowire: true      
        autoconfigure: true 

    App\:
        resource: '../src/'
        exclude:
            - '../src/DependencyInjection/'
            - '../src/Entity/'
            - '../src/Kernel.php'

We must escape the % in the AI response with a kind of :

function query.escapePercent(s)
  return string.gsub(s, "%%", "%%%%")
end

I made it quickly in https://github.com/natixgroup/ai.nvim/commit/0356e9a7ecb19937720d4cd5b8c6239870821fa6 but it very ugly as it needs to be factorized.

The result popup would be better if

• it was a little bigger. 60 x 22 is too small
• it has focus after displaying the results
• has the "markdown" filetype, because Gemini result is always Markdown formatted