Initial #![no_std] mode

[est31]

Addresses most of #20 , except for the panic catching inside the linux module.

[est31]

libcore doesn't have functionality to catch panics. Also I'm not sure whether we'd be creating a circular dependency if we are calling this code inside the panic handler. Removing the catching is bad as well though because panicing through FFI boundaries is UB.

[philipc]

We could add an unsafe API where it is the caller's responsibility to ensure it doesn't panic.

[fitzgen]

It is really hard to verify code paths are panic-free (as I know you know ;) ).

Another option would be to internally do a full pass that copies the relevant info out, and then iterate over the copies with the user-supplied callback. This should make the code that needs to be panic-free easier to manage (certainly for users, since it is only the crate's responsibility now). It also doesn't sound very performant. Perhaps we could add some kind of query/filter API so we could only copy what is necessary and potentially break out of the loop early as well. Thoughts?

I need to mull this over a bit.

[philipc]

It's only a problem if it unwinds, right? So what I should have said is that it is the caller's responsibility to ensure it doesn't unwind. Maybe libstd's usage can be a bit special so that it can set an internal flag that forces any panics to abort instead of unwinding? So I'm hoping for a solution where we can still guarantee that it is correct (whereas it is hard to ensure any copying API we implement would never unwind).

[est31]

Maybe libstd's usage can be a bit special so that it can set an internal flag that forces any panics to abort instead of unwinding?

Seems that this is what is happening already. So I think we are fine with adding an unsafe variant of the API without any panic catching.

[est31]

The period of my contributions to Rust upstream has reached an end. Thus I'm unable to continue my work on this. I still think something like this is a great addition. I urge anyone interested in this change to adopt and continue it from here on.

Note: This PR isn't technically to a Rust upstream repository but I'm still closing it as it was created as a part of me pushing the libbacktrace → addr2line migration over the finish line.