`normalize_and_add_constraints` doesn't add entries in `universe_causes`
when creating new universes, causing an ICE. Remove it!

Add spans to track normalization constraints.

Fix couple places where `universe_causes` is not updated correctly to
track newly added universes.

There are times where computing a value may be cheap, or where
computing a reference may be expensive, so this fills out the
possibilities.

…ref_of}`

While the `provide_*` methods already short-circuit when a value has
been provided, there are times where an expensive computation is
needed to determine if the `provide_*` method can even be called.

Before, the MIR validator used RevealAll in its ParamEnv for type
checking. This could cause false negatives in some cases due to
RevealAll ParamEnvs not always use all predicates as expected here.

Since some MIR passes like inlining use RevealAll as well, keep using
it in the MIR validator too, but when it fails usign RevealAll, also
try the check without it, to stop false negatives.

Before, it called `normalize_erasing_regions` twice since
`equal_up_to_regions` called it as well for both types.

Previously we were just using the parent node as the scope for a
temporary value, but it turns out this is too narrow. For example, in
an expression like

    Foo {
        b: &42,
        a: async { 0 }.await,
    }

the scope for the &42 was set to the ExprField node for `b: &42`, when
we actually want to use the Foo struct expression.

We fix this by recursively searching through parent nodes until we find
a Node::Expr. It may be that we don't find one, and if so that's okay,
we will just fall back on the enclosing temporary scope which is always
sufficient.

revealed by tests, manually add a panic to test for dead code

…ossible...

- ... when creating diagnostics in rustc_metadata
-  use the error_code! macro
- pass macro output to diag.code()
- use fluent from within manual implementation of SessionDiagnostic
- emit the untested errors in case they occur in the wild
- stop panicking in the probably-not-dead code, add fixme to write test

This commit removes many cases of MIR opt tests emitting `.diff`s for more than one pass. These
tests cannot be `unit-test`s, and so they are easy to break, and they also provide little value due
to having excessively strong opinions over *how* a piece of code should be optimized.

Where reasonable, we instead add separate test files that only emit the `PreCodegen.after` MIR for
code where we want to track what the result of the net result of the optimization pipeline's output
is.

Add additional methods to the Demand type

This adds on to the original tracking issue rust-lang#96024

r? `@yaahc`

…stebank

don't ICE when normalizing closure input tys

We were ICEing while rendering diagnostics because `universe_causes` is expected to track every universe created in the typeck's infcx.

`normalize_and_add_constraints` doesn't update `universe_causes`
when creating new universes, causing an ICE. Remove it!

Add spans to better track normalization constraints.

Fix couple places where `universe_causes` is not updated correctly to
track newly added universes.

Fixes rust-lang#99665

… r=oli-obk

Try normalizing types without RevealAll in ParamEnv in MIR validation

Before, the MIR validator used RevealAll in its ParamEnv for type
checking. This could cause false negatives in some cases due to
RevealAll ParamEnvs not always use all predicates as expected here.

Since some MIR passes like inlining use RevealAll as well, keep using
it in the MIR validator too, but when it fails usign RevealAll, also
try the check without it, to stop false negatives.

Fixes rust-lang#99866

cc `@compiler-errors` who nicely helped me on zulip

…rochenkov

optimization of access level table construction

Refactoring which was mentioned in rust-lang#87487

…imulacrum

Change implementation of `-Z gcc-ld` and `lld-wrapper` again

This PR partially reverts rust-lang#97375 and uses the strategy described in rust-lang#97402 (comment) instead, thus fixes rust-lang#97755.

…iser

Simplify MIR opt tests

This commit removes many cases of MIR opt tests emitting `.diff`s for more than one pass. These tests cannot be `unit-test`s, and so they are easy to break, and they also provide little value due to having excessively strong opinions over *how* a piece of code should be optimized.

Where reasonable, we instead add separate test files that only emit the `PreCodegen.after` MIR for code where we want to track what the end to end effect of the optimization pipeline on the example code is.

r? `@wesleywiser`

…s, r=davidtwco

Migrate rustc_metadata to SessionDiagnostics

Migrate rustc_metadata to SessionDiagnostics.

Part of rust-lang#100717

[drop tracking] Use parent expression for scope, not parent node

Previously we were just using the parent node as the scope for a temporary value, but it turns out this is too narrow. For example, in an expression like

    Foo {
        b: &42,
        a: async { 0 }.await,
    }

the scope for the &42 was set to the ExprField node for `b: &42`, when we actually want to use the Foo struct expression.

We fix this by recursively searching through parent nodes until we find a Node::Expr. It may be that we don't find one, and if so that's okay, we will just fall back on the enclosing temporary scope which is always sufficient.

Helps with rust-lang#97331

r? `@jyn514`