impl Debug for Readers

[as-com]

This generates implementations of Debug for Readers for easier debugging. (I did not implement it for Builders because of some messy lifetimes issues.)

To make this work, I changed the getters on Readers to borrow instead of consume (see: #241 (comment)). This should be a safe and backwards-compatible change but I am not completely sure. I had to do this because reborrow throws away lifetime information that is necessary for the Debug implementations to pass the borrow checker.

[dwrensha]

Thanks! I agree that there should be an easy way to print out a textual representation of a reader.
To achieve that, what I really want to do is to port capnproto-c++'s DynamicValue to Rust. That would let us write a Debug impl that require much less generated code -- all readers would just delegate to the same stringification code. I've made some initial small steps in #392 and I intend to continue exploring that.

[as-com]

Something that I considered was implementing LabelledGeneric where Readers' implementations produce an HList of something like Field<(f,i,e,l,d,__,n,a,m,e), PointerField<type::Reader<'_>, Text>> for a sort of type-safe reflection. This could result in lots of messy code requiring specialization though.

Should I separate out changing the getters' self parameter into a separate PR?

[dwrensha]

Maybe instead of doing this we could add the Debug bound in the definition of the Owned trait.

[as-com]

Adding a Debug bound to the Owned trait required adding more constraints to a lot of other Readers. I've put this in a separate commit - which way do you think is cleaner?

[dwrensha]

Should I separate out changing the getters' self parameter into a separate PR?

How did changing the self parameter help you? My understanding is that you should be able to insert reborrow() calls to make things work, but also in the Reader case you shouldn't need the reborrow() calls. Does your issue have to do with some interaction with generics?

[as-com]

How did changing the self parameter help you?

The Debug trait's fmt method uses &self, which means if the getter calls consume self, a reborrow is necessary. The return value of a struct getter after a reborrow has some anonymous lifetime that does not outlive the function, but the returned struct must be passed into the provided Formatter<'_> and must outlive the function call. The reborrow erases the longer lifetime of &self, causing the borrow checker to fail.

[as-com]

That being said, it appears that changing the generated reborrow function to have a signature reborrow(&self) -> Reader<'a, ...> (instead of -> Reader<'_, ...>) allows the reborrows to work as expected (not completely sure of the safety though). This is still less ergonomic than not having to reborrow in the first place though.

[dwrensha]

Ah, the problem is that generic readers were not Copy. I believe that after #393 your self -> &self change is no longer needed.

[as-com]

Ah, the problem is that generic readers were not Copy. I believe that after #393 your self -> &self change is no longer needed.

Yep, that was the issue. I removed the &self change.

[dwrensha]

I've made significant progress on #392, and I'm now convinced that approach is going to work. It should generalize to Builders too.

My plan is to get it into a releasable state and to ship it (including Debug impls for readers) within a small number of weeks from now.

[dwrensha]

@as-com: #392 is almost ready. I would welcome any feedback on it. I intend to write a blog post about it and release capnproto-rust v0.17.0 soon.

[dwrensha]

Closing in favor of #392.
Thanks for the PR!