Support line ranges in @file references

[dangooddd]

What do you want to change?

Currently file reference does not supports line ranges. So the only option is to include file entirely.
For example, mention in github style:

@file#L122-L145

Why?

I am building neovim Pi plugin for personal usage, one of the features converts visual range in neovim to file range and sends to Pi prompt, so agent knows where I am asking. Currently I mention that in prompt without @, so model needs to call read tool first.

How? (optional)

If file reference is text file, parse @file#L{x}-L{y} after file path. That is one possibility to do so, we can support other formats like @file:{x}-{y}

[muyiyr]

I can take this.

Proposed scope:

• Support line ranges only in existing CLI @file arguments, using the exact uppercase syntax #L<start>-L<end>.
• Treat ranges as 1-based and inclusive.
• Resolve the complete argument as a literal path first. Only parse the range suffix when that path does not exist, so existing filenames containing #L... keep working.
• Resolve the base path through the existing path recovery logic (~, Unicode spaces, macOS NFD, and quote variants).
• Support ranges for text files only; reject ranged image references with a clear error.
• Match the read tool's bounds behavior: reject a start beyond EOF, clamp an end beyond EOF, and reject non-positive or reversed ranges.
• Keep the real absolute path in name and expose the effective range separately:
  <file name="/path/to/file" lines="122-145">.
• Leave non-matching suffixes, whole-file references, image handling, and the existing fatal CLI error channel unchanged.

Regression coverage will include inclusive selection, literal-path precedence, effective EOF clamping metadata, invalid ranges, ranged images, CRLF input, and unchanged whole-file behavior.

The implementation should stay contained to packages/coding-agent/src/cli/file-processor.ts, an issue-specific regression test, and the coding-agent changelog.

[xl0]

@dangooddd Pi does not parse the file name at all (except in cli mode, but that's not what you want). You can inclide the line ranges, and the model will interpret them to the best of its ability. Often the models will mis-count the lines though.

If you want to include part of the file in the context explicitly, you will have to do it yourself from the extension. And it's very doable.

Check out https://github.com/xl0/pi-lovely-ide

For this extension, the ide sends the selection information that gets injected intot the prompt as a file reference (with line/character ranes), but then more context is injected as follow-up content after the user message, so the LLM often does not need to read the file explicitly, or at the very least, it has an anchor inside the file, so it knows which lines you are referring to:

If you'd like, you can add a neovim plugin to work with the extension, the protocol is IDE-agnostic, or you can use it as a reference and create your own extension.

[dangooddd]

Thank you, I do understand now. Closing this issue