Don't allow flattened format_args in const.

[m-ou-se]

Fixes #139136

Fixes #139621

We allow format_args!("a") in const, but don't allow any format_args with arguments in const, such as format_args!("{}", arg).

However, we accidentally allow format_args!("hello {}", "world") in const, as it gets flattened to format_args!("hello world").

This also applies to panic in const.

This wasn't supposed to happen. I added protection against this in the format args flattening code, but I accidentally marked a function as const that shouldn't have been const but this was removed in #135139.

This is a breaking change. The crater found no breakage, however.

This breaks things like:

const _: () = if false { panic!("a {}", "a") };

and

const F: std::fmt::Arguments<'static> = format_args!("a {}", "a");

[m-ou-se]

@bors try

[bors]

⌛ Trying commit cc791eb with merge 6619027...

[bors]

☀️ Try build successful - checks-actions
Build commit: 6619027 (6619027b13a87a3597115f2c8231f107bf3a7a5f)

[m-ou-se]

@craterbot check

[craterbot]

👌 Experiment pr-139624 created and queued.
🤖 Automatically detected try build 6619027
🔍 You can check out the queue and this experiment's details.

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🚧 Experiment pr-139624 is now running

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🎉 Experiment pr-139624 is completed!
📊 3 regressed and 5 fixed (612545 total)
📰 Open the full report.

⚠️ If you notice any spurious failure please add them to the denylist!
ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[RalfJung]

These all seem spurious.

Landing this requires FCP, but I'm not sure which team -- lang and/or libs-api?

[m-ou-se]

@rfcbot merge

[rfcbot]

Team member @m-ou-se has proposed to merge this. The next step is review by the rest of the tagged team members:

• @Amanieu
• @BurntSushi
• @dtolnay
• @joshtriplett
• @m-ou-se
• @nikomatsakis
• @scottmcm
• @tmandry
• @traviscross

No concerns currently listed.

Once a majority of reviewers approve (and at most 2 approvals are outstanding), this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

cc @rust-lang/lang-advisors: FCP proposed for lang, please feel free to register concerns.
See this document for info about what commands tagged team members can give me.

[traviscross]

@rfcbot reviewed

If we didn't mean to do it and there's no breakage, we should indeed claw it back.

I'm curious, though, to know whether this is something we might later want to allow and what would need to happen for us to be happy to do that.

[RalfJung]

I'm curious, though, to know whether this is something we might later want to allow and what would need to happen for us to be happy to do that.

As a first step, we should have a very clear idea of which cases we want to allow, and not have this be just "whatever that compiler pass can deal with". :)

[m-ou-se]

I'm curious, though, to know whether this is something we might later want to allow and what would need to happen for us to be happy to do that.

Making some guesses: My expectation is that in the future we will be able to allow all format_args!() invocations in const (since it doesn't do any formatting yet, just sets up the structure). For panic!(), I think the best plan is to accept all panic!() invitations in const, and do the formatting of the message (for the diagnostic) on a best-effort basis (e.g. only format builtin types).

[joshtriplett]

I would love to see all format_args! invocations supported in const, but until then, 👍 for reverting this and not accidentally supporting whatever the flattening pass happens to handle.

[nikomatsakis]

@rfcbot reviewed

[rfcbot]

🔔 This is now entering its final comment period, as per the review above. 🔔

[tmandry]

@rfcbot reviewed

[m-ou-se]

but I accidentally marked a function as const that shouldn't have been const

This wasn't actually the case. I did implement this correctly. The problem was caused by #135139, which added const to a bunch of functions without properly considering the effect on the format_args!() macro.

This means that this problem has only been around since 1.86.0, which explains why we didn't find any actual breakage.

It might make sense to backport this change to land it in 1.87.0.

[RalfJung]

Can you add a test to tests/ui/consts/const-eval/format.rs as well? That seems to be where we have similar tests like this -- using a doctest is somewhat unconventional.