This patch makes error handling for region inference failures more
uniform by not reporting *any* region errors until the reigon inference
step. This requires threading through more information about what
caused a region constraint, so that we can still give informative
error messages.
I have only taken partial advantage of this information: when region
inference fails, we still report the same error we always did, despite
the fact that we now know precisely what caused the various constriants
and what the region variable represents, which we did not know before.
This change is required not only to improve error messages but
because the region hierarchy is not in fact fully known until regionck,
because it is not clear where closure bodies fit in (our current
treatment is unsound). Moreover, the relationships between free variables
cannot be fully determined until type inference is otherwise complete.
cc #3238.
What we now do is to create a region variable for each &
expression (and also each borrow). The lifetime of this
variable will be checked by borrowck to ensure it is not greater
than the lifetime of the underlying data. This both leads to
shorter lifetimes in some cases but also longer in others,
such as taking the address to the interior of unique boxes
tht are rooted in region pointers (e.g., returning a pointer
to the interior of a sendable map).
This may lead to issue #2977 if the rvalue is not POD, because
we may drop the data in trans sooner than borrowck expects us
to. Need to work out precisely where that fix ought to occur.
