Implement (most of) RFC 1214 #27641
Implement (most of) RFC 1214 #27641
Commits on Aug 12, 2015
-
Don't report a hard error if there are inference failures until
after we check casts, because sometimes casts can influence inference, unfortunately. We do re-run `select_all_trait_obligations` during regionck anyhow.
-
Define the
wfandoutlivesrelation separately, unlike the existing`implicator`. These definitions are also in accordance with RFC 1214 (or more so), and hence somewhat different from the implicator. This commit also modifies the implicator to remove the older rules for projections, which can easily trigger infinite loops.
-
-
Extend ParameterEnvironment to remember the free_id, and to be usable
on more kind of items
-
Implement a new wfcheck to replace the old wf; this new code only issues
warnings. It also checks more conditions than the old code. Keep the old wf code around unchanged so that we can continue to issue errors for the cases where we used to report errors. As part of this, remove the where-clauses-must-reference-parameter rule, which is easily circumvented.
-
New WF condition requires checking that argument types are WF
on every call. This ensures that implies bounds are reasonable (the older code only checked that the values provided had WF types, but we also must know that the formal types of the arguments are WF.)
-
Introduce a "origin/cause" for new requirements (or bugfixes...) intr…
…oduced by RFC 1214, and issue a warning (and explanatory note) when we encounter such a thing.
-
Generalize the outlives rule for projections to handle the new cases;
also, generalize VerifyBounds to include OR conditions.
-
-
-
Fallout in tests -- we now report an error if you even reference a type
`&Foo` where `Foo` is a trait that is not object-safe
-
Fallout in tests -- break the object safety part into a separate file…
… because error will be in future reported by wfcheck, which runs in a later stage than coherence
-
Fallout in tests -- break this test into three tests, since we later saw
staging differences in terms of when errors were reported
-
-
Fallout in tests -- explain an interesting test failure having to
do with dropck and the new outlives rules
-
-
-
-
-
-
regionck.rs: correct misuse of ty.regions() rather than regions()
and add a test that was (incorrectly) failing to compile with existing code
-
traits: consider whether origin is RFC1214 when caching, ensuring
that the test rfc1214-warn-and-error.rs reports an error
-
outlives.rs: remove use of ty.walk and replace with recursive of
ty.walk_shallow, add add'l comments.
-
check/wf.rs: change to use correct span and older WF algorithm;
at the time I reinstituted the old code, I hadn't given up yet and brought back the implicator.
-
Update test error messages based on changes to wfcheck; also, break
apart the tests that tested many things at once.
-
-
-
-
regionck.rs: experimentally adopt a more conservative strategy for
projection outlives relations that prefers not to add extract edges to region graph
Commits on Aug 13, 2015
-
expr_use_visitor: Remove FIXME that is no longer needed (and in fact
causes errors, post rebase). Issue rust-lang#27592 is still alive and well, whatever it is.
-