Simplify replace_bytes in line_buffer

[riquito]

This PR removes an unnecessary loop from replace_bytes in line_buffer.

I also added tests for some cases not covered (empty string, same byte replaced, nothing found in a non-empty string).

Unless it's an optimization that I fail to grasp, it should be an oversight from last refactoring
f7ff34f

[BurntSushi]

f7ff34f did not introduce the loop though.

I was curious, so I went back through the history, and this inner loop has been present since it was written initially in 2016:

ripgrep/src/search.rs

Lines 555 to 569 in 812cdb1

	/// Replaces a with b in buf.
	fn replace_buf(buf: &mut [u8], a: u8, b: u8) {
	if a == b {
	return;
	}
	let mut pos = 0;
	while let Some(i) = memchr(a, &buf[pos..]).map(|i| pos + i) {
	buf[i] = b;
	pos = i + 1;
	while buf.get(pos) == Some(&a) {
	buf[pos] = b;
	pos += 1;
	}
	}
	}

To be clear, it wasn't an oversight. That inner loop is indeed an optimization. It works well because binary data tends to have long runs of NUL terminators, and it is slower to stop and start memchr for every byte in a sequence. Its latency is worse than just comparing one-byte-at-a-time.

The extra tests are nice. And you're right that from a correctness perspective, the extra loop is superfluous. So that means a comment saying why it's there would be great to add. Would you like to do that?

[riquito]

Thanks, I learned something