Support &str in the printer algebra

[lpil]

Currently the Document enum structure can only contain Strings, so any time we have a &str we need to allocate a String and copy the &str into it.

Add a new enum variant for Document which can hold a &str, allowing us to remove lots of cloning from the Erlang printer and Gleam formatter.

This will require adding an explicit lifetime to the Document enum.

[scrabsha]

Have you considered using Cow instead of String instead?

[lpil]

I had not! That could be a better solution :)

[scrabsha]

Nice! Can I start working on it?

[lpil]

That would be great! Thank you!

[scrabsha]

After some work, here is a report of my progress so far:

I have uploaded my code on my fork, on this precise commit.

The codebase still have 5 borrowing errors here and there. I am not sure about how to fix them.

I'll continue to work on that, but I may need some help.

[lpil]

Thanks for the patch. I checked out your commit but was unable to determine how to resolve the ownership problems while using Cow. I'm afraid I am far from an expert when it comes to Rust and I've not used Cow before. :(

[scrabsha]

Don't worry, we'll make it work :)

[scrabsha]

I spent a lot of time today fixing these errors. The solution was to add more and more Cow. This led to many other errors, which themselves required more Cow, and so on... The code is not entirely pushed yet, I'll push it when cargo check will report no error.

Also, I notice that the final PR will be very huge. I am currently at +644, -313, and expect it to grow, maybe to +700. As such, I think it would also require a lot of time to review.

Finally, once the PR is accepted, I'll try to create a derive macro on a separate crate which will make Cow handling a bit easier. Hopefully, this would allow us to remove some boilerplate introduced by this PR. Of course, it will have a pun-ish name :)

[scrabsha]

So this if compiling, and tests pass. There as still a lot of clones to remove, but it will be way easier now.

In order to reduce the amount of merge conflicts and the size of the PR, I'd like to split them. Are you okay with it?

[lpil]

Sounds great! :)

[scrabsha]

Hi,

Disclaimer: I use a very formal tone here. While I don't like this way to communicate, I need to ensure that I'll be understood. English is not my native language, sorry about that.

As I said previously, I am pretty sure I still don't understand what is exactly expected here. So I'm back with some questions.

1. Exact lifetime for Document:
   The most general way to fix this issue would be to add a generic lifetime to Document (say, 'a), and to add, as the issue description suggest, a BorrowedText variant, containing an &'a str. However, it would be easier to support static str in a first approach. This would allow us to make for instance "foo".to_doc() require no allocation and no copy. Unfortunately, this would not fix the issue since it has no impact on non-static str, but is already a good start.
   In order to make pull requests easy to review, I'd like to first do the "static str approach", then submit a second PR, which introduce a generic lifetime, and would close this issue. Are you ok with it?
   In my opinion, the current Text variant should be renamed OwnedText. I think it makes sense to show an explicitly-owned and another explicitly-borrowed variant. What do you think of it?

2. About the Document::Break variant:
   The Break variant of Document contains two String (the broken and the unbroken fields). Should I also create a BorrowedBreak variant, with the same properties as BorrowedText?

3. Exact meaning of "algebra":
   Until then, I was convinced that what is called "the printer algebra" refers to the printer algebraic data type, which is in fact the Document enum. Is this correct?

Have a nice day :)
Sasha