Make hitting the recursion limit in projection non-fatal

[matthewjasper]

This change was originally made in #80246 to avoid future (effectively) infinite loop bugs in projections,
but wundergraph relies on rustc recovering here.

cc #80953

r? @nikomatsakis

[jyn514]

I wonder if this would help with #79506.

[jyn514]

See also #79517.

[jyn514]

It does not fix #79506, because that gives an error in check_recursion_limit instead, which wasn't modified here:

rust/compiler/rustc_trait_selection/src/traits/select/mod.rs

Lines 953 to 960 in 5db5d8f

	match self.query_mode {
	TraitQueryMode::Standard => {
	self.infcx().report_overflow_error(error_obligation, true);
	}
	TraitQueryMode::Canonical => {
	return Err(OverflowError);
	}
	}

@matthewjasper should that match be changed too?

[matthewjasper]

This change already isn't really correct, it's just the safest fix for a backport.

For rustdoc it would make sense to move reporting recursion limit errors outside of the trait system, but that's not what this change does (this change incorrectly allows the trait system to recover from recursion limit errors).

[jyn514]

For rustdoc it would make sense to move reporting recursion limit errors outside of the trait system, but that's not what this change does

Got it, thanks. Do you have suggestions for moving recursion limit errors out of the trait system? I'd be interested in working on it.

[matthewjasper]

It's basically find all of the places that create a OverflowError or SelectionError::Overflow and make sure that they are always early returned until outside of rustc_trait_selection::traits::{select, project}, at which point they should be emitted. Overflow errors should never be converted to any other kind of error (such as EvaluatedToRecur or EvaluatedToError)

[apiraino]

beta backport approved, stable backport on hold, will be discussed next week (Zulip discussion)

[Mark-Simulacrum]

beta-nom label should not have been removed, see https://forge.rust-lang.org/release/beta-backporting.html -- can we add this to wg-prioritization docs perhaps? I think we're not highlighting those docs enough or they are confusing perhaps?

[nikomatsakis]

@bors r+

[bors]

📌 Commit 5db5d8f has been approved by nikomatsakis

[nikomatsakis]

@bors p=1

[apiraino]

discussed in T-compiler meeting, declined to backport to stable

@rustbot modify labels: -stable-nominated

[bors]

⌛ Testing commit 5db5d8f with merge c0b64d9...

[bors]

☀️ Test successful - checks-actions
Approved by: nikomatsakis
Pushing c0b64d9 to master...

[Aaron1011]

this change incorrectly allows the trait system to recover from recursion limit errors

@matthewjasper Is there an open issue for applying the correct fix?