Revert text setter ergonomics pessification; add raw_set methods about capnproto-rust HOT 6 CLOSED

Note that instead of set_foo(x.as_str().into()), you can do set_foo(x[..].into()), which is a bit shorter.

from capnproto-rust.

Adding separate setter methods for capnp::Text::Reader and str could maybe work. We would need to figure out:

1. naming -- I don't like raw_ because that already means "raw pointer" elsewhere
2. how it would interact with generics -- if a field has a generic type T, then does that field now need two different setters, because T might be instantiated to text::Owned?

Alternatively, we could consider reviving #421, which would make pointer setters accept any type that implements SetPointerBuilder.

I'm reluctant to do either of these because they add more complexity, but maybe the potential ergonomic gains for downstream users outweigh the cost.

from capnproto-rust.

Alternative naming suggestion: for a text field foo and text reader r, fd_foo().set(r). fd_foo returns a TextField.

This aligns with another proposal I was planning to make, which is to reduce the boilerplate of building lists of structs, by enabling writing, for a field bars of Capnp type List(Bar) and some v of native Rust type Vec<MyBar>, something like fd_bars().from_iterable(v, |bar_cpn, bar| { /* Initialize the bar_cpn from the bar */ }). The essence of a Field is that, given a length, it can return a Builder: fd_bars().init(len) would return a struct_list::Builder<bar::Owned>. Conveniences such as from_iterable can be built on top of that.

from capnproto-rust.

FYI: I decided to vendor capnproto-rust, so I am not blocked on this issue being resolved here. I reverted the text setters to taking &str verifast@01871b5 . I also changed the init_foo methods for initializing lists to taking a usize length instead of u32, and the list get method to taking a usize index. Furthermore, I generate a fill_foo method for each field foo of type list-of-struct, that looks like this:

#[inline]
pub fn fill_args<E, F>(&mut self, elems: E, body: F)
    where F: FnMut(crate::vf_mir_capnp::body::basic_block::operand::Builder<'_>, E::Item), E: IntoIterator, E::IntoIter: ExactSizeIterator {
  <::capnp::struct_list::Builder<'_,crate::vf_mir_capnp::body::basic_block::operand::Owned>>::fill(self.builder.reborrow().get_pointer_field(1), elems, body)
}

which uses a new fill associated function in struct_list::Builder:

impl<'a, T> Builder<'a, T>
where
    T: crate::traits::OwnedStruct + 'a,
{
    pub fn fill<E, F>(builder: PointerBuilder<'a>, elems: E, mut body: F)
        where F: FnMut(T::Builder<'_>, E::Item), E: IntoIterator, E::IntoIter: ExactSizeIterator
    {
        let iter = elems.into_iter();
        let mut list_builder = Self::init_pointer_with_usize_length(builder, iter.len());
        for (idx, elem) in iter.enumerate() {
            let elem_cpn = list_builder.reborrow().get(idx);
            body(elem_cpn, elem);
        }
    }
}

Commit: verifast/verifast@60e9ca4#diff-fe25a34e136e01802f5f365abad8bf12d11c18a922ffd283bed2c3af52eb1e35

I also did something similar for text lists: I generate a method fill_foo for each field of type List(Text):

#[inline]
pub fn fill_arg_names<E>(&mut self, elems: E)
    where E: IntoIterator, E::IntoIter: ExactSizeIterator, E::Item: AsRef<str> {
  <::capnp::text_list::Builder<'_>>::fill(self.builder.reborrow().get_pointer_field(3), elems)
}

which uses a new fill associated function on text_list::Builder:

impl<'a> Builder<'a> {
    pub fn fill<E>(builder: PointerBuilder<'a>, elems: E)
        where E: IntoIterator, E::IntoIter: ExactSizeIterator, E::Item: AsRef<str>
    {
        let iter = elems.into_iter();
        let mut list_builder = Self::init_pointer_with_usize_length(builder, iter.len());
        for (idx, elem) in iter.enumerate() {
            list_builder.set(idx, elem.as_ref());
        }
    }
}

Commit: verifast/verifast@68b70c6

from capnproto-rust.

I revived #421 as #472. I think this works well, but unfortunately would re-break the code of everyone who has already updated to capnp-v0.18.

from capnproto-rust.

Closing, because I've landed #472.
Thanks for pushing on this!

from capnproto-rust.