Add default and latest stable edition to --edition in rustc

[sigaloid]

Fixes #106041

Boxes and leaks the formatted string as per #106041 (comment)

It will now output the default edition and latest stable edition in the help message for the --edition flag.

[rustbot]

r? @petrochenkov

(rustbot has picked a reviewer for you, use r? to override)

[kadiwa4]

I think you can avoid one allocation by writing Box::leak(format!(...).into_boxed_str())

[bjorn3]

Could you please avoid this Box::leak call? It is possible to invoke rustc multiple times in the same process. Each time would leak memory. You should be able to inline the opt_s function to pub fn opt_s(a: S, b: S, c: &str, d: S) -> R { to allow strings with arbitrary lifetimes and then pass &format(...).

[sigaloid]

Yep - I was actually pretty wary of this change but that's what I was recommended to do. I'll rewrite the type signature and see if it still works soon so we can avoid this leak.

[sigaloid]

Unfortunately I cannot refactor this out as the reference needs to outlive 'static

error[E0521]: borrowed data escapes outside of function
    --> compiler/rustc_session/src/config.rs:1290:9
     |
1289 |     pub fn opt_s(a: S, b: S, c: &str, d: S) -> R {
     |                  -           -  - let's call the lifetime of this reference `'1`
     |                  |           |
     |                  |           `c` is a reference that is only valid in the function body
     |                  `a` declared here, outside of the function body
1290 |         stable(longer(a, b), move |opts| opts.optopt(a, b, c, d))
     |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     |         |
     |         `c` escapes the function body here
     |         argument requires that `'1` must outlive `'static`

There's got to be something I'm missing - the enum variant for the formatted string are all static and embedded, and there's got to be a better option than either allocating, or just manually writing in the strings and commenting above the original variables, "this must be updated in xyz string". I don't know how but I think there might be a way to construct this statically somehow - it's just concatenating all-static strings.

[bjorn3]

I see. Another option would be to use std::sync::LazyLock to compute the description. If you put it in a static it should allow you to get a single &'static str without leaking anything.

[petrochenkov]

r? rust-lang/compiler
I'd like to take some break from reviewing.

[Noratrieb]

Thank you for your PR! But I see you accidentally added a merge commit (probably by updating your branch with the github UI). This repository doesn't allow merge commit and you should get rid of it. There's an explanation in the dev guide, feel free to ask further questions if that doesn't help.

As for the Box::leak, I would prefer the solution suggested by bjorn3 to use a LazyLock instead of leaking the string. While leaking a few bytes here isn't a problem memory wise, we should still leak as few bytes as possible to help leak detectors.

[sigaloid]

Thanks for the recommendation!

In regards to the LazyLock, are you recommending creating a static LazyLock at the top of that function and referencing it in that function? I attempted it and could not get it to become a &'static str despite my best efforts.

[bjorn3]

Did you try &*MY_STATIC?

[sigaloid]

That worked @bjorn3 - thanks!