search

oceanlewis / uap-rs

User Agent Parser for Rust
MIT License
42 stars 17 forks source link

Switch to or support `regex-lite` #18

Open Dav1dde opened 2 months ago

Dav1dde commented 2 months ago

I haven't run the numbers yet for this crate, but I did for ua-parser which brings down memory consumption significantly.

I was initially just going to PR the change, but since this crate supports full unicode and no unicode modes (is there a usecase for full unicode?), this is not as easy since regex-lite does not support unicode. So it would only be possible to switch the non-unicode codepath to regex-lite. Which means the regex dependency can't be fully replaced.

What would be your preferred approach?. I have a these proposals:

  1. Add regex-lite with a feature flag and only enable it for the non-unicode codepath (might be quite unergonomic code wise since we'll end up with a wrapper type for the regex dispatching to either regex or regex-lite).
  2. Drop unicode support and switch fully to regex-lite
  3. Do nothing

What are your thoughts?

Dav1dde commented 2 months ago

It seems like the performance hit may not be desirable and is much more than I anticipated. It may still make sense as an opt-in for environments where the user wants to prioritize resources constraint environments, e.g. when unicode is disabled and the regex-lite feature flag is enabled.