lewisclark / patternscan Goto Github PK

A pattern between the chunk boundary defined by CHUNK_SIZE will not be found. The failing test below demonstrates this:

#[test]
fn find_across_chunk_boundary() {
  let mut bytes = vec![0; super::CHUNK_SIZE - 2];
  bytes.push(0xaa);
  bytes.push(0xbb);
  bytes.push(0xcc);
  bytes.push(0xdd);
  let pattern = "aa bb cc dd";
  
  assert!(crate::scan_first_match(Cursor::new(bytes), &pattern).unwrap().is_some())
}

One thought I've had while working with this crate is that it would be nice if the Pattern and PatternByte APIs were publicly exposed. My reasoning for this is that it would allow developers who know the search pattern at compile time to just pass a Pattern to scan directly, without the need for parsing a pattern string. Furthermore, this goes some way to enforce compile-time checking of known patterns, as if a Pattern is constructed from u8s directly, it must inherently be valid.

I'm happy to put together a pull request for this if you agree it would be useful, just interested to hear your thoughts before I go ahead.

The current implementation of both scan and scan_first_match returns the position of each pattern match within a given chunk, rather than within the overall byte string read. This is because we don't keep track of what chunk we're in, and the loop through the bytes in each chunk starts from i = 0 each time. Here's a quick test case demonstrating the issue:

use patternscan;
use std::io::Cursor;

let mut bytes = vec![0; 0x800];
let mut more_bytes = vec![0x10, 0x20, 0x30, 0x40];
bytes.append(&mut more_bytes);

let pattern = "10 20 30 40";

assert_eq!(patternscan::scan_first_match(Cursor::new(bytes), &pattern)
    .unwrap()
    .unwrap(),
0x800); // Fails: left = 0, right = 2048

I'm currently working on a fix for this and #4, as well as updating the API a bit, so hopefully I'll be able to submit a pull request for that this evening, just need to escape from my family for a while to have the time 😆