⚠️ No Changeset found

Latest commit: 3e22c4d

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

Instead of adding a new tool and having the LLM choose between them, should we just add optional start/end line to the read_file tool along with a setting for the maximum number of lines to read from a file?

Instead of adding a new tool and having the LLM choose between them, should we just add optional start/end line to the read_file tool along with a setting for the maximum number of lines to read from a file?

ok, i will update pr with that approach

@mrubens updated ! :D

Nice! How does it feel?

@mrubens it's work perfectly :D current test with file 333M (2000000 line) from https://www.datablist.com/learn/csv/download-sample-csv-files, and it work ok, i think we can support mention read file in range with syntax like "@test.log::{10,100}"

A couple thoughts on this:

1. I think we should have a default (that can be configured in settings) where if you don't specific the start and end it reads up to X lines of the file. I think that would save a lot of tokens from accidentally reading long files.
2. Along with that, I think we would need to return more information about the file along with the selected range of lines. Specifically how many total lines there are in the file, and ideally information from the tree-sitter about the definitions within the file and which lines they're on (similar to what Codey does).

What do you think?

I guess not exactly the same approach, but this is what I'm thinking of: https://github.com/ccrvlh/codey/blob/f823be88322b7417d75c4db98a0cc224de6ad97b/src/core/tools.ts#L306-L341

A couple thoughts on this:

1. I think we should have a default (that can be configured in settings) where if you don't specific the start and end it reads up to X lines of the file. I think that would save a lot of tokens from accidentally reading long files.
2. Along with that, I think we would need to return more information about the file along with the selected range of lines. Specifically how many total lines there are in the file, and ideally information from the tree-sitter about the definitions within the file and which lines they're on (similar to what Codey does).

What do you think?

1. sound good idea, let me do that
2. let me thinking how can we do it

Great! Maybe we can reuse what that list_code_definitions tool does somehow?

Great! Maybe we can reuse what that list_code_definitions tool does somehow?

yes, but the tool use processing is hard to maintain now, we need to separate into another module to easy maintain later

Does tail behavior work? for example:

Reading the lines >= 1000 of a log file:

<read_file>
<path>logs/application.log</path>
<start_line>1000</start_line>
</read_file>

I think so? But maybe we would actually want to support start_line=-100 for this (100 from the end)?

this approach would allow multiple ranges, slicing:

<range>123:456</range>
<range>777:</range>       # 777 until the end 
<range>:1111</range>     # beginning until 1111 (inclusive) 

also see my related comment here: #1374 (comment)

I think so? But maybe we would actually want to support start_line=-100 for this (100 from the end)?

I like that notation too, perhaps:

<range>:-100</range>

this approach would allow multiple ranges, slicing:

<range>123:456</range>
<range>777:</range>       # 777 until the end 
<range>:1111</range>     # beginning until 1111 (inclusive) 

also see my related comment here: #1374 (comment)

I think so? But maybe we would actually want to support start_line=-100 for this (100 from the end)?

I like that notation too, perhaps:

<range>:-100</range>

@KJ7LNW sound good idea

@mrubens should we allow read multiple file but limit their range ?

@mrubens should we allow read multiple file but limit their range ?

what do you mean by "allow read multiple file" ?

but limit their range

hard-coded limits are a bad idea. let the AI choose, or create a config option for limits

@mrubens should we allow read multiple file but limit their range ?

what do you mean by "allow read multiple file" ?

in single request and response allow user can read multiple file

but limit their range

hard-coded limits are a bad idea. let the AI choose, or create a config option for limits

No we have setting, and already have that in this PR, you can check it

I like the ideas of the support for tailing and multiple file reads, but I think those can be follow-ups - there's already a lot in this PR.

it looks like this PR picked up a huge number of changes that are not related to this feature.

can you rebase on main?

be sure to backup your work first:

• git brach read-file-range-before-rebase
• git fetch origin # assuming roo is origin
• git rebase origin/main

it looks like this PR picked up a huge number of changes that are not related to this feature.

can you rebase on main?

be sure to backup your work first:

• git brach read-file-range-before-rebase

• git fetch origin # assuming roo is origin

• git rebase origin/main

Already rebase on main, you can check the git commit

it looks like this PR picked up a huge number of changes that are not related to this feature.
can you rebase on main?
be sure to backup your work first:

• git brach read-file-range-before-rebase
• git fetch origin # assuming roo is origin
• git rebase origin/main

Already rebase on main, you can check the git commit

that looks much better thank you!

Going to run this as my editor today and see how it goes!

Going to run this as my editor today and see how it goes!

I am curious how the AI utilizes the file ranges!

@samhvw8 looking at the exports of my chats, I don't think it's currently appending the definitions to the tool result when the read is truncated.

@samhvw8 looking at the exports of my chats, I don't think it's currently appending the definitions to the tool result when the read is truncated.

Let me find a way to truncate more 💪🏻

@samhvw8 looking at the exports of my chats, I don't think it's currently appending the definitions to the tool result when the read is truncated.

Let me find a way to truncate more 💪🏻

Oh I think I was confused!