don't add implied bounds after normalization

[aliemjay]

Forked from #99217 because it is a preexisting problem. cc @lcnr

We should not add implied bounds for the normalized type because we don't know for sure if the normalization holds.

Consider this example:

trait Tr {
    type Ty;
}

impl<T: 'static> Tr for T {
    type Ty = ();
}

fn test(_: <&u8 as Tr>::Ty) {}

This now fails in stable as expected, but with this change it passes!

-    type Ty = ();
+    type Ty = &'static M;

r? @ghost

[rust-log-analyzer]

The job x86_64-gnu-llvm-12 failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

---- [ui] src/test/ui/nll/issue-52057.rs stdout ----

error: test compilation failed although it shouldn't!
status: exit status: 1
command: "/checkout/obj/build/x86_64-unknown-linux-gnu/stage2/bin/rustc" "/checkout/src/test/ui/nll/issue-52057.rs" "-Zthreads=1" "--target=x86_64-unknown-linux-gnu" "-O" "--error-format" "json" "--json" "future-incompat" "-Ccodegen-units=1" "-Zui-testing" "-Zdeduplicate-diagnostics=no" "-Cstrip=debuginfo" "-C" "prefer-dynamic" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/test/ui/nll/issue-52057/a" "-Crpath" "-Cdebuginfo=0" "-Lnative=/checkout/obj/build/x86_64-unknown-linux-gnu/native/rust-test-helpers" "-L" "/checkout/obj/build/x86_64-unknown-linux-gnu/test/ui/nll/issue-52057/auxiliary"
stdout: none
--- stderr -------------------------------
error[E0311]: the parameter type `I` may not live long enough
   |
note: the parameter type `I` must be valid for the anonymous lifetime defined here...
   |
LL |     fn parse_first(_: &mut Self::Input) {}
   |                       ^^^^^^^^^^^^^^^^
Some tests failed in compiletest suite=ui mode=ui host=x86_64-unknown-linux-gnu target=x86_64-unknown-linux-gnu
Some tests failed in compiletest suite=ui mode=ui host=x86_64-unknown-linux-gnu target=x86_64-unknown-linux-gnu
   = note: ...so that the type `I` will meet its required lifetime bounds
help: consider adding an explicit lifetime bound...
   |
LL | impl<'a, I: 'b, P: ?Sized> Parser for &'a mut P

error: aborting due to previous error
------------------------------------------

[jackh726]

Is this only a problem in #99217, not on master? If so, this needs to be merged into that PR.

I don't think we should not treat normalized types as implied, but rather I think we need to fully prove (including region obligations) that the projection ty normalizes.

[lcnr]

Is this only a problem in #99217, not on master? If so, this needs to be merged into that PR.

It's a problem on master, or well, I am not sure if it is a problem, because while surprising, i wasn't able to cause ub with it

[aliemjay]

I agree this is not a soundness problem. Moreover I think with #99217 we can even ignore the region obligations generated by normalizing <&u8 as Tr>::Ty when borrow-checking test because we can be sure that they're checked by the caller.

It was an experiment to resolve this surprising behavior, but I missed the case where the projection is normalized to a type parameter in src/test/ui/nll/issue-52057.rs. Closing.