Upgrade to rustc 1.0.0-nightly (4db0b3246 2015-02-25) (built 2015-02-26)

[SimonSapin]

This makes it build again, but it have no idea what I’m doing. Please review for memory safety.

libcore now has impl<T> !Send for *mut T { }, which might be what causes:

src/atom/mod.rs:59:1: 59:42 error: conflicting implementations for trait `core::marker::Send` [E0119]
src/atom/mod.rs:59 unsafe impl Send for *mut StringCacheEntry { }
                   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
src/atom/mod.rs:59:1: 59:42 note: conflicting implementation in crate `core`
src/atom/mod.rs:59 unsafe impl Send for *mut StringCacheEntry { }
                   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[SimonSapin]

It looks like this is failing on Travis because examples/summarize-events has a git dependency on string-cache from github, instead of a path dependency from its grand-parent directory. Should that be changed?

[kmcallister]

examples/summarize-events has a git dependency on string-cache from github, instead of a path dependency from its grand-parent directory. Should that be changed?

Oops, it totally should, now that it's part of the Travis build.

[SimonSapin]

What about examples/event-log?

[kmcallister]

It seems like both buckets and next_in_bucket should use ptr::Unique<StringCacheEntry>. This expresses the ownership relationship and therefore enables StringCache to be Send without an unsafe impl.

I started making this change, and it's rather tricky! For one, Unique is non-nullable. That's solvable with Option but then I ran into the problem that Unique isn't Copy (nor should it be), which complicates the insert / remove code. I think it's pushing me towards a design with less aliasing and more clear ownership, but I didn't quite get there.

The unsafe impl is a reasonable fix for now. It avoids sweeping changes to unsafe code that by all indications is working correctly.