Option<snafu::Backtrace> handling should be configurable

[cyphar]

Right now, the only way to have runtime-controlled optional backtraces for some library or application is to use Option<snafu::Backtrace> and then the user must set RUST_BACKTRACE or RUST_LIB_BACKTRACE in order to enable it. While this does mean that the same UX for "getting a backtrace" is present for all snafu-based Rust programs, it has the major downside of making the UX feel strange for programs which are not necessarily related to Rust. For instance, a user may find it odd for a shared library (written in Rust) used by a C program to require you to set RUST_LIB_BACKTRACE=1 in the environment.

In addition, the fact it's set once and cannot be changed means that for long-running processes it's necessary to restart them in order to enable debugging -- when ideally you should be able to enable debugging after the program has been started and handled more than one error.

[shepmaster]

The functionality is configurable, just not via Option. You need to implement GenerateBacktrace for your type and then you can rely on whatever logic you want. An untested sketch:

use snafu::{Backtrace, GenerateBacktrace};
use std::sync::atomic::{AtomicBool, Ordering};

struct GlobalBacktrace(Option<Backtrace>);

static ENABLED: AtomicBool = AtomicBool::new(false);

impl GenerateBacktrace for GlobalBacktrace {
    fn generate() -> Self {
        Self(if ENABLED.load(Ordering::SeqCst) {
            Some(GenerateBacktrace::generate())
        } else {
            None
        })
    }

    fn as_backtrace(&self) -> Option<&Backtrace> {
        self.0.as_ref()
    }
}

[cyphar]

I will admit, I didn't think to implement GenerateBacktrace on a custom struct -- that should solve my problem. Thanks, and sorry for the noise.