is_ident_start reads the wrong byte for a hyphen before a backslash

[hey-jj]

When a - is followed by a backslash that begins an escape, the check that decides whether an identifier starts here reads the wrong byte. A backslash right before a newline is an invalid escape and cannot start an identifier, so -\<newline> is mis-tokenized.

Reproducer

cssparser 0.37.0, default features.

use cssparser::{Parser, ParserInput};

fn main() {
    let mut pi = ParserInput::new("1-\\\n");
    let mut p = Parser::new(&mut pi);
    let a = p.next().unwrap().clone();
    // cssparser: Dimension { value: 1.0, int_value: Some(1), unit: "-" }
    println!("{:?}", a);
}

Observed vs expected

Observed for 1-\<newline>: the tokens are Dimension { value: 1.0, unit: "-" } then Delim('\\'). For -\<newline> the first token is Ident("-"). For #-\<newline> the first token is IDHash("-").

Expected per CSS Syntax Level 3, matched by tinycss2: 1-\<newline> gives Number(1), Delim('-'), Delim('\\'). -\<newline> gives Delim('-') then Delim('\\'). #-\<newline> gives a hash token Hash("-") with the unrestricted type flag, not an id hash.

Root cause

src/tokenizer.rs:895. The would-start-an-identifier test reads the byte after the backslash to decide if -\ begins an identifier. A backslash directly before a newline is not a valid escape, but the offset used skips the backslash and reads the newline as though it were escaped.

Scope

A hyphen before an invalid escape produces a unit, an identifier, or an id hash that the input never held. A consumer that keys on token type, such as a linter or a minifier, then acts on tokens the source did not contain.