Adding a method to clear the Gimli symbol cache

[Tomcc]

As discussed in #222 , in my use case I can't really afford each process to lock hundreds of mb's per thread in the backtrace symbol cache.

I've added a clear_symbol_cache() function to the root of the crate that clears the Gimli mapping cache and uh, does nothing at all if you aren't using Gimli.
I tried to fix libbacktrace as well, but that one doesn't expose a way to free its state even in C.

I'm not that good at Rust still so this PR is more of a prototype, let me know if anything is terrible :)
I know that a few things could be improved:

• Is it fine to take a dependency on lazy_static?
• The Mutex on the static is redundant, but I couldn't placate the compiler.
• Defining fn noop_clear_symbol_cache() {} again for each backend is ugly.

Anyway, there it is.

[alexcrichton]

Thanks for the PR! Could this perhaps avoid picking up a dependency on lazy_static for now? It's probably best to tackle #225 in a dedicated PR.

Additionally if this didn't actually fix the issue for you I wonder if there's some more investigation to do? There may be more caches that need to be cleared in this crate somewhere I'm not realizing?

(one may be libunwind which may have its own set of internal caches)

[Tomcc]

Sure, I can avoid lazy_static.

Additionally if this didn't actually fix the issue for you I wonder if there's some more investigation to do? There may be more caches that need to be cleared in this crate somewhere I'm not realizing?

Actually this PR fixed my problem! The extra memory leftover that I mentioned on the other issue was... the println! that I used to debug it.
Interestingly, printing out anything from a background thread seems to allocate a bunch of caches as well... I've already deployed this code though, and it does fix the problem. I caused a few backtraces and it's still sitting at 38mb :)
I could even downsize the VM down to 1gb and the swapping doesn't occur anymore, so I trust what docker stats is telling me.

[alexcrichton]

Thanks again for this!

[Tomcc]

Thank you for helping me with this!
Btw, I saw that all my commits with horrible messages went in as-is... I thought I'd have a chance to clean them up before merging them :P
I guess it's too late now!

[alexcrichton]

Ah no worries I don't tend to be too strict with commit history, it's all good!