Auto merge of rust-lang#30533 - nikomatsakis:fulfillment-tree, r=aturon

This PR introduces an `ObligationForest` data structure that the fulfillment context can use to track what's going on, instead of the current flat vector. This enables a number of improvements:

1. transactional support, at least for pushing new obligations
2. remove the "errors will be reported" hack -- instead, we only add types to the global cache once their entire subtree has been proven safe. Before, we never knew when this point was reached because we didn't track the subtree.
   - this in turn allows us to limit coinductive reasoning to structural traits, which sidesteps rust-lang#29859
3. keeping the backtrace should allow for an improved error message, where we give the user full context
    - we can also remove chained obligation causes

This PR is not 100% complete. In particular:

- [x] Currently, types that embed themselves like `struct Foo { f: Foo }` give an overflow when evaluating whether `Foo: Sized`. This is not a very user-friendly error message, and this is a common beginner error. I plan to special-case this scenario, I think.
- [x] I should do some perf. measurements. (Update: 2% regression.)
- [x] More tests targeting rust-lang#29859
- [ ] The transactional support is not fully integrated, though that should be easy enough.
- [ ] The error messages are not taking advantage of the backtrace.

I'd certainly like to do 1 through 3 before landing, but 4 and 5 could come as separate PRs.

r? @aturon // good way to learn more about this part of the trait system
f? @arielb1 // already knows this part of the trait system :)

Author: bors

src/librustc/diagnostics.rs

@@ -679,6 +679,43 @@ There's no easy fix for this, generally code will need to be refactored so that
 you no longer need to derive from `Super<Self>`.
 "####,
 
+E0072: r##"
+When defining a recursive struct or enum, any use of the type being defined
+from inside the definition must occur behind a pointer (like `Box` or `&`).
+This is because structs and enums must have a well-defined size, and without
+the pointer the size of the type would need to be unbounded.
+
+Consider the following erroneous definition of a type for a list of bytes:
+
+```
+// error, invalid recursive struct type
+struct ListNode {
+    head: u8,
+    tail: Option<ListNode>,
+}
+```
+
+This type cannot have a well-defined size, because it needs to be arbitrarily
+large (since we would be able to nest `ListNode`s to any depth). Specifically,
+
+```plain
+size of `ListNode` = 1 byte for `head`
+                   + 1 byte for the discriminant of the `Option`
+                   + size of `ListNode`
+```
+
+One way to fix this is by wrapping `ListNode` in a `Box`, like so:
+
+```
+struct ListNode {
+    head: u8,
+    tail: Option<Box<ListNode>>,
+}
+```
+
+This works because `Box` is a pointer, so its size is well-known.
+"##,
+
 E0109: r##"
 You tried to give a type parameter to a type which doesn't need it. Erroneous
 code example:

src/librustc/middle/check_const.rs

@@ -125,7 +125,7 @@ impl<'a, 'tcx> CheckCrateVisitor<'a, 'tcx> {
             None => self.tcx.empty_parameter_environment()
         };
 
-        let infcx = infer::new_infer_ctxt(self.tcx, &self.tcx.tables, Some(param_env), false);
+        let infcx = infer::new_infer_ctxt(self.tcx, &self.tcx.tables, Some(param_env));
 
         f(&mut euv::ExprUseVisitor::new(self, &infcx))
     }
@@ -280,7 +280,7 @@ impl<'a, 'tcx> CheckCrateVisitor<'a, 'tcx> {
 
     fn check_static_type(&self, e: &hir::Expr) {
         let ty = self.tcx.node_id_to_type(e.id);
-        let infcx = infer::new_infer_ctxt(self.tcx, &self.tcx.tables, None, false);
+        let infcx = infer::new_infer_ctxt(self.tcx, &self.tcx.tables, None);
         let cause = traits::ObligationCause::new(e.span, e.id, traits::SharedStatic);
         let mut fulfill_cx = infcx.fulfillment_cx.borrow_mut();
         fulfill_cx.register_builtin_bound(&infcx, ty, ty::BoundSync, cause);