Feature Request: Optional Inclusion of Jupyter Cell Outputs

[filipchristiansen]

We are considering a feature to optionally include the outputs of Jupyter code cells. When enabled, the tool would append the last output of each code cell (if any) directly below the cell's source code as a comment.

Proposed Solution:

• Introduce a command-line flag for the terminal client to toggle this behavior.
• Add a similar toggle option in the web client's "Advanced Settings" (in a future release) to enable optional inclusion of cell outputs.

Request for Feedback:

Please share your thoughts on the utility of this feature. Your feedback will help us prioritize new features based on actual user needs. Thank you!

[IsNoobgrammer]

Yes!! , It would help in give llms or users more context on how that "notebook" should work in first place,

as when we see notebook , we see not only its code , but its outputs too , so I think we can try adding it , however I think it should be by default @cyclotruc , Because thats what Notebook are for (Better Understanding and working of code) !!!

[IsNoobgrammer]

To provide further context
outputs can be of different types like and i will provide my suggestion for every type

__OUTPUT__TYPE__ = "execute_result"
"outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x24ee8f511d0>]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }

So in this case , execute_result should be parsed output= execute_result : [<matplotlib.lines.Line2D at 0x24ee8f511d0>]

__OUTPUT__TYPE__ = "display_data"
   "outputs": [
    {
     "data": {
      "image/png": "B64 IMAGE DATA",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "<Figure size 640x480 with 1 Axes>"
    }

so in this case We don't need that image b64 value output=display_data : <Figure size 640x480 with 1 Axes>

__OUTPUT__TYPE__ = "stream"
"outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "hello\n"
     ]
    }
   ]

from this output = hello\n

__OUTPUT__TYPE__ = "error"
"outputs": [
  {
   "ename": "NameError",
   "evalue": "name 'oxe' is not defined",
   "output_type": "error",
   "traceback": [
    "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
    "\u001b[1;31mNameError\u001b[0m                                 Traceback (most recent call last)",
    "Cell \u001b[1;32mIn[9], line 1\u001b[0m\n\u001b[1;32m----> 1\u001b[0m \u001b[38;5;28mprint\u001b[39m(oxe)\n",
    "\u001b[1;31mNameError\u001b[0m: name 'oxe' is not defined"
   ]
  }
 ],

we need output as NameError : name 'oxe' is not defined since in jupyter error is very verbose sometimes

Note: Outputs : List[dict | None] is the output of the cell in all formats , so , earlier I mentioned to parse only outputs[-1] , but that approach was wrong , the better approach would be parse all dicts from outputs according to the proposed format and then write that whole parsed_output as comment

for example

"outputs": [
    {
     "ename": "ValueError",
     "evalue": "'text' is not a valid format string (unrecognized character 't')",
     "output_type": "error",
     "traceback": ["VERY","Verbose","traceback"]
    },
    {
     "data": {
      "image/png": "FILLER b64",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ]

so the expected output from this cell should be

# * output *
#  ValueError : 'text' is not a valid format string (unrecognized character 't')
#  display_data : <Figure size 640x480 with 1 Axes>

[filipchristiansen]

I will leave it up to @cyclotruc to decide about the default behaviour.

@IsNoobgrammer: I am working on a PR.

[cyclotruc]

Yes!! , It would help in give llms or users more context on how that "notebook" should work in first place,

as when we see notebook , we see not only its code , but its outputs too , so I think we can try adding it , however I think it should be by default @cyclotruc , Because thats what Notebook are for (Better Understanding and working of code) !!!

@IsNoobgrammer Hi thank your for your input

I'm not sure I understand right, I am not very familiar with Notebooks, but are the output values sometimes pre-loaded in the notebook's source code?

Because if not, that would mean we would have to run the notebooks to be able to render their outputs, which is not really doable within the scope of gitingest

But I probably misunderstand what you meant, could you elaborate?

[IsNoobgrammer]

Yes!! , It would help in give llms or users more context on how that "notebook" should work in first place,
as when we see notebook , we see not only its code , but its outputs too , so I think we can try adding it , however I think it should be by default @cyclotruc , Because thats what Notebook are for (Better Understanding and working of code) !!!

@IsNoobgrammer Hi thank your for your input

I'm not sure I understand right, I am not very familiar with Notebooks, but are the output values sometimes pre-loaded in the notebook's source code?

Because if not, that would mean we would have to run the notebooks to be able to render their outputs, which is not really doable within the scope of gitingest

But I probably misunderstand what you meant, could you elaborate?

Yes @cyclotruc , When outputs are "cached" in the notebook , then we can also parse output from notebook without running it , it will make the parsed data more like a notebook-type (Interactive) than python script type ,
for example see this sample notebook
HuggingFace - Example
So , if you see the notebook has output also stored , so what i meant was it would be better if we parse output(if there is any) too , since they are integral part of notebook and comment it below each cell , by default

[cyclotruc]

Oh ok I see! Thanks for the explanation

Then it seems like a good idea to include them by default if it doesn't add too much complexity, it's more context for the LLM anyway

[filipchristiansen]

As promised, here’s the pull request for the notebook output inclusion: #128.

@IsNoobgrammer