Re: Proposal: AI-Based Code Review for DPDK

[Aaron Conole <aconole@redhat.com>]

Aaron Conole <aconole@redhat.com> writes:

> Jerin Jacob <jerinjacobk@gmail.com> writes:
>
>> On Thu, Jun 19, 2025 at 7:42 PM Ori Kam <orika@nvidia.com> wrote:
>>>
>>>
>>> Nice idea,
>>> Some comments:
>>>
>>>
>>> 1. The first link is broken for me.
>>
>> Yes.  https://github.com/DPDK/dpdk-ai-code-review is the proposed name
>> of the repository. It is not created yet.
>
> I guess we need this created by someone (but I don't know who should be
> the contact).

Cc'd David

Is it possible to create this repository?  I guess this is the first step.

>>>
>>> 2. I like the idea that there is a separate GitHub repository. If I
>>>     understand correctly the new repository is a clone of the main one,
>>>     and only used by the bot is that correct?
>>
>> Yes.
>>
>>>
>>> 3. I think main usage is also for the reviewer, just like in your example.
>>
>> Yes
>>
>>>
>>> 4. The only thing I’m missing is that there is no “personal touch”
>>>     meaning that each reviewer puts emphasis on different things. It will
>>>     be great if the AI can learn from previous reviews of the reviewer
>>>     and add review accordingly.
>>
>> Yes. However, It looks like https://docs.sourcery.ai/Code-Review/ dont
>> have that feature.
>>
>>>
>>> 5. It will also be great if the AI could target people who may be
>>>     interested in this patch.  And send them some notification or tell
>>>     the developer to add them in the to
>
> Yes, well - not everyone will be on GitHub.  But the idea / plan from my
> side was to try and get the comments associated with the specific hunks
> and reply to the patch (and all the folks CC'd) with the comments.  At
> the very least, we can send the pull request URL and reply with that.
>
>> Good input.
>>
>>
>> Adding @Aaron Conole . He has some plans to implement this workflow.
>
> Yes - the pull-request submission side is probably fairly simple to
> implement.
>
> But I need to look at monitoring the pull requests.

I looked briefly at this, and it seems like we should be able to do
something with a JS like (just mocking it up):

  curl -L -X POST \
       <GITHUB HEADERS>
       https://api.github.com/repos/DPDK/dpdk-ai-code-review/pulls \
       '{"title":"Series AI review - <SERIES DETAILS>", "body":"<cover
       letter / commit message>",
       "head":"ovsrobot/dpdk:series_SERIESID",
       "base":"main"
       }'

That tells us the pull request ID in the response, and we can then
monitor it and pull the review comments.  I haven't written up a script
suite for it, but it looks fairly straightforward.

>>>
>>> Best,
>>> Ori
>>>
>>> On 13/06/2025 13:03, Jerin Jacob wrote:
>>> >
>>> > Goals
>>> > -------
>>> >
>>> > # Leverage modern AI-powered code review tools to accelerate and
>>> > supplement human review efforts.
>>> > # Prevent AI-generated comments from cluttering the mailing list
>>> > or polluting the main DPDK GitHub repository.
>>> >
>>> > Setup
>>> > -------
>>> >
>>> > # Create a dedicated repository:
>>> > Set up a new repository, for example:
>>> > https://github.com/DPDK/dpdk-ai-code-review, under the official
>>> > DPDK GitHub organization.
>>> > This ensures that AI-generated pull requests do not interfere with
>>> > the main repository (https://github.com/DPDK/dpdk).
>>> >
>>> > # Select an AI review tool:
>>> > Use a free, open-source-friendly AI code review service.
>>> > One potential candidate is Sourcery: https://sourcery.ai/pricing
>>> >
>>> > # Connect the tool:
>>> > Integrate Sourcery with the new repository via:
>>> > https://app.sourcery.ai/?product=prod_opensource
>>> >
>>> > # Trigger AI reviews:
>>> > Once integrated, any pull request to this repository will
>>> > automatically trigger a code review by the AI tool.
>>> > Example of a test PR reviewed by the AI bot:
>>> > https://github.com/MarvellEmbeddedProcessors/dao/pull/2
>>> >
>>> > Workflow
>>> > -------------
>>> >
>>> > # When a patch is sent to the DPDK mailing list, a bot will
>>> > convert it into a GitHub pull request on the dpdk-ai-code-review
>>> > repository.
>>> > # The AI tool will review the pull request and leave comments
>>> > directly on the PR.
>>> > # Once the AI review is complete, the bot will close the PR and
>>> > share the PR URL:
>>> > - As a comment in Patchwork URL for the given patch
>>> > - With the patch author directly.
>>> > # The author can review the AI-generated comments by visiting the
>>> > PR URL, if they choose to do so.
>>> >
>>> >
>>>