Field presence

[snproj]

Continuation of #243

Since I no longer have access to the ghpr-asia fork that I made that PR with.

• Some minor formatting changes rolled back to reduce the diff
• Custom default impls should now only be generated when there are custom default fields
• Added --generate-getters to generate getters for Proto2 fields with custom defaults

Minor notes:

• Forgot to mention in previous PR message, but custom default fields for strings and bytes must now begin and end with either single or double inverted commas (previously not enforced). I had to enforce this to predictably generate default const fields from user input.
• The minor change from #[derive(Debug, Default, PartialEq, Clone)] to #[derive(Default, Debug, PartialEq, Clone)] is there because Default is now being inserted programmatically into the list of derives, so it was best to simply append to the start

Future action

• Proto2 and Proto3 parsing will be made stricter; originally this was part of this PR (since the whole parsing system got restructured anyway), but it's probably better to separate it out. The changes are:
  • Proto3 enums must begin with index 0
  • Proto3 cannot have required fields
  • Proto3 cannot have custom defaults

[snproj]

Made the changes as per above, also removed some extra parentheses in the generated get_size()es that previous versions of this PR added in.

Edit: cleaning up diff

[snproj]

I've spent a few hours trying to get the diff as small as possible for the get_size() and write_message() parts, but the way that the code generation has changed in this PR means that eliminating all the differences is hard without a bunch of error-prone rewriting (that will eventually be reverted anyway when I make the PR to clean it up, since I already have the code for that). I don't know that it's impossible, but I didn't want to delay too long haha

If you're curious, the previous code had control flow in various parts that determine when to add a dereference into the code, which had built up over time into generating a bunch of &&*&**&. This is the main source of diff, since I decided to rewrite the control flow to avoid these chains of references and dereferences; which makes it hard to artificially add them back in and shrink the diff.

Hopefully it's ok 🙏

[thomaseizinger]

I've spent a few hours trying to get the diff as small as possible for the get_size() and write_message() parts, but the way that the code generation has changed in this PR means that eliminating all the differences is hard without a bunch of error-prone rewriting (that will eventually be reverted anyway when I make the PR to clean it up, since I already have the code for that). I don't know that it's impossible, but I didn't want to delay too long haha

If you're curious, the previous code had control flow in various parts that determine when to add a dereference into the code, which had built up over time into generating a bunch of &&*&**&. This is the main source of diff, since I decided to rewrite the control flow to avoid these chains of references and dereferences; which makes it hard to artificially add them back in and shrink the diff.

Hopefully it's ok pray

Yeah that is fine. There is a means to an end when it comes to reducing diff :)

[thomaseizinger]

https://github.com/libp2p/rust-libp2p/pull/3455 uses this branch and it works well, green light from our end!

[d-roak]

Hi @snproj!

Can you do a version bump in the near future to include this change?

Thanks 🙏

[mxinden]

Friendly ping. Could you cut a new release with this patch?