Add Protobuf.Text

[v0idpwn]

This PR introduces support for encoding Protobuf messages into the textproto format.

Motivation

The primary motivation for this feature is to improve the ability to inspect and visualize Protobuf messages, while still conforming to some standard. The inspect protocol default implementation can be painful to work with large or complex structures. Finding the relevant, non-default fields in a inspected struct is not intuitive and takes an unnecessary amount of time. A textproto representation provides a human-readable alternative that is much closer to the actual data transmitted over the wire (specially when using transformers).

This implementation

• Encoding only: Decoding was deliberately left out as it is less valuable compared to encoding, and has many caveats.
• Emphasis on readability: pretty-printed by default. We do a best effort for breaking lines, and opted for the most readable and elixir-like alternatives in the spec. Max line width and tab size are is configurable.
• Limitations:
  • No support for extensions
  • No support for Google.Protobuf.Any.
  • No type validation is performed.

Future Considerations

I'm open to extending this implementation in the future to include:

1. Adding support for extensions
2. Adding support for Google.Protobuf.Any.
3. Implementing type validation.

[whatyouhide]

Should we look at using Inspect.Algebra to simplify the line splitting? Not sure what the spec of the format is and I don't have the bandwidth to go read it now, but it seems flexible enough that we could break lines in a few different places?

[v0idpwn]

Thank you, @whatyouhide. I'll explore using Inspect.Algebra, it's probably a good fit here. :)

[v0idpwn]

Pushed a new version using Inspect.Algebra. It indeed simplified code a lot, thanks for the suggestion.

Some limitations/changes:

• I wasn't able to have different separators on container_doc for single line and multi-line. I'm sure it's achievable if I use the lower level functions, but I think having a , after every line isn't that bad (it's optional in the spec), and I like the simplicity of container_doc.
• I wasn't able to define :pad_size as I was in the hand-crafted implementation, so it's a fixed 2.

I'll address the other comments over the week :)

[whatyouhide]

@v0idpwn OK sounds good, lmk when this is gonna be ready for re-review. 🙃

[v0idpwn]

I tried this on Protobuf.Encoder and it broke conformance tests, despite the fact that in the docs it seems like this is the expected behaviour there as well (just as we skip 0 for non-optional integers, I'd expect it to skip the default enum value...)

[v0idpwn]

I'll investigate this further

[v0idpwn]

I think it's okay to be reviewed, @whatyouhide :)

[v0idpwn]

Actually, I found that there are conformance tests for text encoding. I'll work on them.