Open tpaulshippy opened 3 months ago
I'm interested in working on this if no one else gets to it first.
ah yes! would make a great contribution. maybe as a config?
Taking another look...looks like this flag already provides some support for this --
-f, --test-file
I guess the config would be to change the behavior so that files generated from the tool without any flags provided would be structured properly.
And make the interactive mode work for these kinds of projects.
a potentially even better approach - let the llm figure that out. there is a part where we pass in a file tree, so maybe theres a world where it can look at your existing files and determine what the pattern should be
this may be hard for large projects with too many files to include, but could be worth exploring
Interesting. I'll play with the tool more and see if this could work.
Would love to use the tool to make this change on itself.
Oh wow ya! The tough part would probably be is we want to use an LLM for part of the solution then that would be costly to run in a unit test - though if we add ollama ( https://github.com/BuilderIO/micro-agent/issues/12 ) support that would solve that and allow the LLM to run in the unit tests locally which would be pretty interesting
On Fri, Jun 07, 2024 at 9:24 AM, Paul Shippy < @.*** > wrote:
Interesting. I'll play with the tool more and see if this could work.
Would love to use the tool to make this change on itself.
— Reply to this email directly, view it on GitHub ( https://github.com/BuilderIO/micro-agent/issues/13#issuecomment-2155157386 ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAGOEA4YGXBRSHIT4FGDSETZGHNDVAVCNFSM6AAAAABI64DWOWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNJVGE2TOMZYGY ). You are receiving this because you commented. Message ID: <BuilderIO/micro-agent/issues/13/2155157386 @ github. com>
True or we could mock the AI and just make assertions on the prompts.
a potentially even better approach - let the llm figure that out. there is a part where we pass in a file tree, so maybe theres a world where it can look at your existing files and determine what the pattern should be
Reading more of the code I'm seeing that it should already be doing this. With the limitation of 100 files. Going to try this on my rails project and open a separate issue if there are things that aren't working.
Cancel that. The code I saw was https://github.com/BuilderIO/micro-agent/blob/05070973b9c71bd7f3b6f24e203d67314954f5ad/src/helpers/interactive-mode.ts#L58 but this is for the implementation file, not the test file. Looks like we need a recommendedTestFilePath
with the option to confirm.
I wonder if another way to improve the 100 file limitation would be to just grab the first file from each folder. This way your sampling would have a better chance of covering all the folders.
Oh yeah that’s an interesting idea. Or max one of each suffix (so we can see at least one example of the potential patterns used like .js or .test.ts or .py etc)
I agree that more than one example file name of each suffix
On Sat, Jun 8 2024 at 1:42 PM, Paul Shippy < @.*** > wrote:
I wonder if another way to improve the 100 file limitation would be to just grab the first file from each folder. This way your sampling would have a better chance of covering all the folders.
— Reply to this email directly, view it on GitHub ( https://github.com/BuilderIO/micro-agent/issues/13#issuecomment-2156176297 ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAGOEAY4RZUMDWJS6J72ZXDZGNUDBAVCNFSM6AAAAABI64DWOWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNJWGE3TMMRZG4 ). You are receiving this because you commented. Message ID: <BuilderIO/micro-agent/issues/13/2156176297 @ github. com>
In many frameworks, rails for example, the tests or specs are placed in a separate folder from the implementation. The folder structure of the tests often mirrors that of the implementation. Would like this project to support that.