Not compiling after adding the GAT Self: 'a bounds

[vandenheuvel]

After #89970 was merged, project relp no longer compiled as expected: I needed to add a few Self: 'a bounds.

After I did just that, I was left with the following error:

error[E0477]: the type `RemoveRows<'_, MP>` does not fulfill the required lifetime
  --> src/algorithm/two_phase/mod.rs:57:21
   |
57 |                     phase_two::primal::<_, _, SteepestDescentAlongObjective<_>>(&mut non_artificial)
   |                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

For more information about this error, try `rustc --explain E0477`.

After I made the following change:

-pub fn primal<IM, MP, PR>(
-    tableau: &mut Tableau<IM, NonArtificial<MP>>,
+pub fn primal<'provider, IM, MP, PR>(
+    tableau: &mut Tableau<IM, NonArtificial<'provider, MP>>,
 ) -> OptimizationResult<IM::F>
 where
-    IM: InverseMaintener<F: im_ops::FieldHR + im_ops::Column<<MP::Column as Column>::F>>,
-    for<'r> IM::F: im_ops::Cost<MP::Cost<'r>>,
+    IM: InverseMaintener<F:
+        im_ops::FieldHR +
+        im_ops::Column<<MP::Column as Column>::F> +
+        im_ops::Cost<MP::Cost<'provider>> +
+    >,

cargo build succeeded.

The core of this bug report is the question why the extra change above could have been necessary, as #89970 should not have influenced that.

Summarized from a private conversation with @jackh726.

[jackh726]

@rustbot ping cleanup

It would be really nice to get a minimal test for this. The intent for this lint was that in theory, existing code shouldn't break. So, either we need to update the outlives lint code to not error here or decide that this pattern isn't correct and we should figure out the correct fix and make sure it's documented.

[rustbot]

Hey Cleanup Crew ICE-breakers! This bug has been identified as a good
"Cleanup ICE-breaking candidate". In case it's useful, here are some
instructions for tackling these sorts of bugs. Maybe take a look?
Thanks! <3

cc @AminArria @ayazhafiz @camelid @chrissimpkins @contrun @elshize @h-michael @HallerPatrick @hdhoang @hellow554 @henryboisdequin @imtsuki @JamesPatrickGill @kanru @KarlK90 @matheus-consoli @mental32 @nmccarty @Noah-Kennedy @pard68 @PeytonT @pierreN @Redblueflame @RobbieClarken @RobertoSnap @robjtede @SarthakSingh31 @shekohex @sinato @smmalis37 @steffahn @Stupremee @tamuhey @turboladen @woshilapin @yerke

[jackh726]

For clarification, the most ideal MCVE would be of the code at this commit.

[vandenheuvel]

This issue might be related.

[vandenheuvel]

I am almost certain that this bug has to do with this line:

-    type Cost<'a> = MP::Cost<'a>;
+    type Cost<'a> where Self: 'a = MP::Cost<'a>;

The problem does not occur when the following change is applied before the where Self: 'a bound is added:

-    type Cost<'a> = MP::Cost<'a>;
+    type Cost<'a> = MP::Cost<'provider>;

On this branch, I applied the commits of this branch (where the original issue occured) in a different order. After starting with the commit above, I remove a few ugly HRTB's in an intermediate commit. After that, adding the Self: 'a clauses works as expected.

[jackh726]

@vandenheuvel were you able to fix your project and also include the required bounds?

[jackh726]

GATs issue triage: not blocking. It's really not clear what an actionable path forward here is. It would be nice to know what exactly led to these problems, but I can't see anything that changes the outlives lint rules. We also have the flexibility in the future to make the outlives lint more lenient.

[jackh726]

I'm going to go ahead and close this. As said in previous comment, not sure what is actionable here.