Slice regions upset in only one case

[graydon]

I don't quite see why but the regions being generated for c and cc in test/run-pass/estr-slice.rs are no longer popular with the compiler as of the fix for compiling const slices. Uncommenting the cases there makes them fail to compile, despite appearing almost identical to the a and b cases above.

[nikomatsakis]

ok, I tracked this down. It's kind of a general problem with the inference heuristics. They don't appear to work so great for regions. I am planning to reimplement the region inference, which is a far more limited problem than type inference. This'll be additional test case to consider.