Make crate no_std

[encounter]

The majority of changes are removing reliance on std::io. Instead of using write!, we can use the infallible Vec::extend.

This removed most of the error handling from serialize, with a couple of exceptions:

• The ctor/dtor errors "If there's a ctor/dtor, there should be another name in this sequence". I replaced these with writing [invalid] to the output.
• The conversion from Vec to String. I modified this to just emit lossy UTF-8 if there are any errors. (Though I don't expect this to ever happen in practice, given that parse should have handled any invalid UTF-8 already)

With those refactored, serialize now returns a String instead of Result<String>. This also allowed me to remove the Io error variant entirely. #![no_std] success!

Also included:

• Bump to Rust edition 2021
• Some minor clippy suggestions
• Version bump to 0.11, given the serialize return type is an API breaking change

[mstange]

Looks good to me except for the addition of the format! calls. I think those are new memory allocations that we didn't have with write!. And all of them are just for surrounding a value with fixed strings. Could you replace each format! call with a sequence of three calls to extend?

[encounter]

It looks like all of the format! calls are actually formatting integers, so I'm not sure of a way to avoid format! or .to_string() on those.

[mstange]

Oh I missed that.

I think one option would be to use the itoa crate for that. It seems to be a very lightweight dependency. What do you think?

[encounter]

itoa looks nice! I’ll add that in.

[encounter]

All set, I replaced integer formatting with itoa, and single-character extends with push.

Also, I was able to avoid allocations for nested serialize calls by simply calling self.serialize(val) rather than self.w.extend(serialize(val).as_bytes()).

[mstange]

Excellent, thank you!

[mstange]

Released as 0.11.0.

[mstange]

Out of curiosity, what's your use case for running this in a no_std context?

[encounter]

For my objdiff tool, I've been working on ensuring all dependencies are #![no_std] for the core layer that can compile to WASM. Specifically for reduced binary size and control over WASI imports.