Repository navigation
Feature Requests: Lua support, and questions of beyond. #64
Replies: 3 comments 1 reply
|
Thanks for opening a discussion! For a more immediate unblock, I believe adding Lua (or any other language support) is not too tricky
export { default as luaQuery } from "./lua"
const languages = [
// ... existing languages
"lua",
]
Do you have bandwidth to take this on? Feel free to send a PR. I am a bit swamped rn but can do this in some time |
|
very good points.
Dirac lazily initializes treesitter parsers so more parsers won't necessarily hurt performance unless there is a project with 10 different languages present and we overload memory loading them all at once, but that's a rare situation so I am not worried |

Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Lua is my current use primary case with Dirac, but it is not supported in the list of langauges, so its support is degraded compared to others.
It is a fairly popular language and has an active listing: https://github.com/tree-sitter-grammars/tree-sitter-lua
Speaking more broadly, there are a lot of other languages such as Zig (sorta niche), Fennel (very niche), etc. that I'd be interested using with Dirac in the future that have treesitter queries. Including some of my own private DSL queries that I'd like dirac to gain awareness for, but couldn't be expected to be added given they'd be constantly moving targets for a userbase of one. Not sure how those can be easily handled without constantly forking? The pace of AI harnesses are blistering, and Dirac is no different with around daily changes, so maintaining a fork would certainly be a hassle. Ideally there would be some way to point Dirac at some queries stored separately so it doesn't need to be constantly updated.
All reactions