Code Dryer

• Added README
• Added icon and updated package.json with repo and author information

• Added support for comparisons across multiple files
• Fixed a few bugs with normalization:

1. When a literal and a variable was close to each other edits overlapped and one got corrupted
2. Keyword arguments were taken as variables, fixed that.
3. In Python type annotations were also being mistaken for variables, fix that too.

• Added a slider to the similarities view to allow the user to change the cut off similarity.

• Added support for top level code chunk (i.e. not in any function or class). Currently, it is counted as a chunk if, it is between 2 functions/classes or there are one or more new lines between sections.
• Also now we are filtering out low similarities, this should have been added at the beggining.

I think after improving the normalization and adding multi-file support this project will be ready to ship But there is only 17 hours left

• Added code comment removal for further normalization of the code
• In the webview the normalized code was shown for no reason, now it shows the raw source code.
• Created the simplest proxy server for embeddings and deployed it to Railway.

Displaying the similarities

I added a webview that displays the similar functions when you run DRY Code command

Also improved the embeddings similarities logic: no duplicates any more

Embeddings and comparisons

I implemented the the embedding creation, storage, caching and comparison.
Embeddings are calculated thanks to HCAI, they are then stored locally. When something changes or a new function is added they are recalculated (of course only the ones needed).
The problem I have is that I was using a hard-coded key for development. I probably gonna need to make a simple proxy backend too

Parser implementation

In order to compare functions and classes I first need to identify them.
Getting this part to work was much harder than I expected. I first tried doing this with VS Code’s language server but that didn’t go as I expected. So, I switched to tree-sitter. To be honest, I relied a bit more on AI than usual in this part.