Document F-bounds and inference using bounds for 3.7

[MaryaBelanger]

Fixes #6258

I added the section "Self-referential type parameter restriction" to the generics page, to support the "Inference using bounds" section on the Type system page (staged links).

This is a difficult topic to summarize simply enough for the language tour! I did my best, but please let me know if anything needs to change / if this misses the point of the new feature.

[dart-github-bot]

Visit the preview URL for this PR (updated for commit a180b33):

https://dart-dev--pr6403-inference-bounds-m64rfkos.web.app

[MaryaBelanger]

@parlough The error from test / analyze is, I think, just because inference using bounds is not stable yet...?

error - examples/type_system/lib/strong_analysis.dart:171:3 - Couldn't infer type parameter 'X'.
          
          Tried to infer 'C' for 'X' which doesn't work:
            Type parameter 'X' is declared to extend 'A<X>' producing 'A<C>'.
          The type 'C' was inferred from:
            Parameter 'x' declared as     'X'
                          but argument is 'C'.
          
          Consider passing explicit type argument(s) to the generic.
          
           - could_not_infer
   info - examples/misc/test/language_tour/generics_test.dart:70:[16](https://github.com/dart-lang/site-www/actions/runs/13251876512/job/36991328626?pr=6403#step:6:17) - Unnecessary use of an abstract class. Try making 'compareTo' a top-level function and removing the class. - one_member_abstracts

For the info message though, do you have an idea for a better way to write that? That's how I could get it to work in dartpad on main, but lmk if there's something neater I could do. Thanks!

[parlough]

@parlough The error from test / analyze is, I think, just because inference using bounds is not stable yet...?

Yep! Now I've updated main to 3.7 and merged it in, analysis seems to be working now 👍

For the info message though, do you have an idea for a better way to write that?

No, I think that's good. I pushed a change that adds an ignore comment. I also added the interface class modifier, which I think might be a good signal to not trigger the lint. I'll explore making that change in the SDK.

[chloestefantsova]

The edits LGTM. Thank you so much, Marya! I'd like to known @eernstg's opinion on this too.

[MaryaBelanger]

Thank you Chloe! And yes I will definitely wait for Erik's input, looking forward to it!

[eernstg]

Thanks for the kind words! I added a bunch of comments, hope they are helpful!

[eernstg]

Inference fails when this feature isn't available. When the feature is available, inference succeeds. I don't think that's obvious at this point. I also don't know how it could be explained in that comment without breaking the 80-or-whatever limit on the line length. ;-)

[MaryaBelanger]

Yeah I can see from all your suggestions that I wrote this section kind of on the line between treating this like a new feature/improvement, and trying to make it sound like this is "just the way it works". I'll clean up the whole section with that in mind, thank you for the thorough review!

[MaryaBelanger]

Maybe someone can help with code comment best practices, but I added a longer explanation here:

  f(C()); // OK. Without using bounds, inference relying on best-effort
  // approximations would fail after detecting that C is not a subtype of A<C>.

[eernstg]

Added a couple of comments.

[eernstg]

When we don't have f2, perhaps a name like f would suffice. (Unless there's a name clash with some other declaration named f nearby, but I couldn't see any).

[parlough]

Thanks @MaryaBelanger and everyone for working so hard on figuring out what to include here and how to word it.

Looks good to me! Just one suggestion to consider:

[MaryaBelanger]

Ahh sorry I merged it too fast! I'll reopen a pr for your fixes, thank you for checking