ExtendInto<Extender=Vec<u8>> appears to have no useful implementors

[ExpHP]

It is unclear to me how to use bytes::complete::escape_transform on bytestrings, because the only implementors of ExtendInto<Extender=Vec<u8>> are [u8] and &[u8], neither of which can be returned without borrowing from the input.

• Rust version : rustc 1.46.0
• nom version : 5.0.0
• nom compilation features used:

Test case

use nom::{
    bytes::complete::{escaped_transform, take, is_not},
    combinator::{map},
};

fn main() {
    let res: nom::IResult<_, _> = escaped_transform(  // Error: ExtendInto is not implemented for u8
        is_not(r#"\""#),
        '\\',
        map(take(1u32), |bytes: &[u8]| match bytes[0] {
            b'n' => b'\n',
            b => b,
        }),
    )(b"sfj\\n\\isoeg");
    assert_eq!(res, Ok((&b""[..], b"sfj\nisoeg".to_vec())))
}

[Geal]

you could replace the closure like this:

|bytes: &[u8]| match bytes[0] {
            b'n' => &b"\n"[..],
            _ => &bytes[..1],
}

[ExpHP]

Okay, this is true, you can use static references if there are only a limited number of escapes.

Is there any particular reason why u8 and Vec<u8> do not implement the trait? They would provide additional power by allowing the values to be constructed at runtime, making it easier to support escapes like "\xXX" for an arbitrary byte. (I think this can be done with static refs but it's a bit janky, as it requires having a static string with all byte values like b"\x00\x01\x02\x03...\xfe\xff" to take slices from)

[klardotsh]

I ran into a related-but-different variant of this today with string escape sequences. Definitely frustrating to waste a few hours digging into this and trying just about every workaround I could possibly think of and getting almost nowhere useful.

My usecase is parsing string literals that work more or less like Python's raw string literals: \" will output a bare double quote, anything else after a backslash is passed through verbatim (including the backslash). ExtendInto is not implemented for String, so returning a variable length replacement sequence appears to not be doable. Bizarrely it's implemented for str, which I have no clue how to make use of in this case, and of course it's implemented for &str which is great for 'statics, but not so great for dynamic variables I'm creating on the fly.

I wonder if a PR to add this String implementation would be accepted. It'd be quite little code, and would clean up some otherwise pretty gross hackarounds necessary to support this type of string literal parsing with escaped_transform.