not API changed on v1.0.0-alpha.4

[soywod]

Hi,

I wanted to try the v1.0.0-alpha.4 (I come from the v0.9.2) and I noticed that .not() is not acting the same. It used to return the current output (let's say Vec<char>) but now it seems to returns the unit. How can I have the same behaviour as before?

pub(crate) fn text_plain_part<'a>() -> impl Parser<'a, &'a str, String, extra::Err<Rich<'a, char>>>
{
    choice((
        just(SINGLE_PART_BEGIN),
        just(SINGLE_PART_END),
        just(MULTI_PART_BEGIN),
        just(MULTI_PART_END),
    ))
    // FIXME: not() returns unit instead
    .not()
    .repeated()
    .at_least(1)
    .collect::<String>()
}

Thank you!

[Zij-IT]

In this case to get the desired behavior you use something like (map_)slice:

pub(crate) fn text_plain_part<'a>() -> impl Parser<'a, &'a str, String, extra::Err<Rich<'a, char>>> {
    any()
        .and_is(
            choice((
                just(SINGLE_PART_BEGIN),
                just(SINGLE_PART_END),
                just(MULTI_PART_BEGIN),
                just(MULTI_PART_END),
            ))
            .not()
        )
        .repeated()
        .at_least(1)
        .map_slice(ToString::to_string)
}

The any().and_is(/* snip */) is required because an empty parse is valid for something.not() and therefore will infinite loop without progress when used in combination with repeated. By using any().and_is I am forcing that at least one token be parsed per iteration of "repeated"

[soywod]

It worked like a charm, thank you. I just needed to replace .map_slice(ToString::to_string) by .collect::<String>().

[zesterer]

Sorry about this change, I'll be documenting this in the eventual migration guide. The new behaviour is strictly more general than the old behaviour, enabling a bunch of more interesting patterns, but obviously makes this specific case a little bit more verbose.

[zesterer]

I just needed to replace .map_slice(ToString::to_string) by .collect::().

As an additional point: the new zero-copy capabilities in 1.0 allow you to avoid allocating in these circumstances by just using .slice() and having the parser return &'a str as its output type.

Even if you still want to return a String, .map_slice(ToString::to_string) will be more efficient since it's directly converting between a &str to a String, avoiding the reallocations that would be needed when collecting up individual chars into a String with the old approach.

[soywod]

Don't worry, I am aware that it is an alpha, I just wanted to try it out. Thank you for the tips!