Support for Fixed Sized Arrays

[Carter12s]

Description

ROS1 serialization with serde_rosmsg requires that fixed sized arrays be represented not as Vec<> but as [; N]. The two are sent over the wire differently. Most of the support for this was already plumbed through, so this just closed the loop at the end, but is hitting the "const generic" barrier.

Fixes

Closes: #148

Checklist

• Update CHANGELOG.md

[Carter12s]

@ssnover Do you have any good ideas on how we should handle this situtation?

1. Is there a workaround where maybe we can use some other type than [T; N] to represent a fixed sized array?
2. Can we codegen our own trait impls for larger array sizes? (This seems awful)
3. Do we just declare our crate only works with fixed sized arrays <32 elements until const generics land? can we limit that impact to ROS1 usage only? probably not cause it affects codegen...

Am I just missing an obvious solve here.

[ssnover]

We have at least one instance showing in the codegen for a 36 member array, so we probably ought to at least support that as a basic ROS message.

Only alternative I can think of is to represent with a tuple of Vec and the fixed size and then define our own serialization and deserialization for serde_rosmsg?

[ssnover]

Actually, how does serde_rosmsg handle Box<[T]>?

[Carter12s]

I have a little side test setup in my fork of serde_rosmsg, from when the bug was originally reported, where I'm deserializing a NavSatFix.msg with the library. NavSatFix has an [f64; 9] in it.

For Box<[f64]> serde_rosmsg spat out an underflow error at runtime.
For &[f64] we get a compile time "Deserialize<'_> is not implemented"

So don't see a good path there...
I may open an issue on serde_rosmsg and see if they have a solve...

[Carter12s]

https://stackoverflow.com/questions/62665558/how-can-i-implement-serdedeserialize-for-arrays-larger-than-32

Has lead me to: https://crates.io/crates/serde-big-array

Which I'm going to play with and see if it can solve this problem...

See root serde issue here: serde-rs/serde#1937

[Carter12s]

So it looks like a combination of some fuckery with smart_default's _code directive and this serde-big-array crate I may have gotten a working solution.

Biggest problem I see is 1) the code is super messy and could use some unit tests, and 2) we're leaking yet another crate dependency through the codegen interface in not a great way...

I'd love your eyes on this @ssnover

[ssnover]

This is great, thanks for taking the time to figure out the attribute macro magic required to get this functioning and adding tests.

I pointed out the current CI failures in comments.

Actions completed, not sure how to "say done"