unlink2 / rbrep Goto Github PK

Matching is currently rather slow.
The new tinyvec matcher has a clearer api, but as a result performance has degraded.
In the future this implementation should be as fast as the old callback-based one.

Currently the matcher uses a callback to pass data through the recursive call back up to the caller and simply returns a parsed length. Maybe this should be reconsidered.

Sometimes it may be useful to have a group that is matched with the and operation.

It should be possible to read data past the pre-allocated buffer to allow more flexible matching.

Adding tests will help refactoring later on and will also document all valid inputs.

It should be possible to match 1 to n or 0 to n operators for all match expressions.

Maybe use nn? for 0 to n
and nn+ for 1 to n

The current matcher works well, but is slow. Investigate better matching methods that yield faster results.
Something like Boyer-Moore could work.
This should also go hand in hand with refactoring the reader to be a trait that can read bytes at any index of the file and slide the window.