Panic on find_from_ucs2

[jedel1043]

Probably related to #100.

The fix for that issue introduced Input::CODE_UNITS_ARE_BYTES, which skips some optimizations as invalid for UTF-16 and UCS2, like:

regress/src/cursor.rs

Lines 59 to 78 in ffb373e

	#[inline(always)]
	pub fn next_byte<Input: InputIndexer, Dir: Direction>(
	input: &Input,
	_dir: Dir,
	pos: &mut Input::Position,
	) -> Option<u8> {
	assert!(
	Input::CODE_UNITS_ARE_BYTES,
	"Not implemented for non-byte input"
	);
	let res;
	if Dir::FORWARD {
	res = input.peek_byte_right(*pos);
	*pos += if res.is_some() { 1 } else { 0 };
	} else {
	res = input.peek_byte_left(*pos);
	*pos -= if res.is_some() { 1 } else { 0 };
	}
	res
	}

However, some parts of the classical backtrack engine still use byte matching even for UTF-16:

regress/src/classicalbacktrack.rs

Lines 570 to 578 in ffb373e

	&Insn::ByteSet2(bytes) => {
	next_or_bt!(scm::MatchByteArraySet { bytes }.matches(input, dir, &mut pos))
	}
	&Insn::ByteSet3(bytes) => {
	next_or_bt!(scm::MatchByteArraySet { bytes }.matches(input, dir, &mut pos))
	}
	&Insn::ByteSet4(bytes) => {
	next_or_bt!(scm::MatchByteArraySet { bytes }.matches(input, dir, &mut pos))
	}

Reproducer

use regress::Regex;

fn main() {
    let string: Vec<_> = "football".encode_utf16().collect();
    let res = regress::Regex::with_flags("foo", "i").unwrap();
    let matches: Vec<_> = res.find_from_ucs2(&string, 0).collect();
}

thread 'main' panicked at /home/jedel/.cargo/registry/src/index.crates.io-6f17d22bba15001f/regress-0.10.2/src/cursor.rs:65:5:
Not implemented for non-byte input

[ridiculousfish]

Yikes. On it.