enable error backtraces

[jbr]

This is the result of a yakshave to get backtraces on http-types errors, which were not previously functional as documented.
#[cfg(backtrace)] does nothing by itself. Anyhow provides this flag to itself using a build script, so this PR adds a nearly-identical build script to http-types.

It also correctly forwards Debug fmt to the inner error.

[jbr]

I'm not entirely sure how to conditionally test this, but on a nightly toolchain, given this file in http-types/examples/backtrace.rs:

fn main() {
    dbg!(http_types::format_err!("error").backtrace());
}

env RUST_LIB_BACKTRACE=1 cargo run --example backtrace →

[examples/backtrace.rs:2] http_types::format_err!("error").backtrace() = Backtrace [
    { fn: "backtrace::backtrace::libunwind::trace", file: "/Users/runner/.cargo/registry/src/github.com-1ecc6299db9ec823/backtrace-0.3.46/src/backtrace/libunwind.rs", line: 86 },
    { fn: "backtrace::backtrace::trace_unsynchronized", file: "/Users/runner/.cargo/registry/src/github.com-1ecc6299db9ec823/backtrace-0.3.46/src/backtrace/mod.rs", line: 66 },
    { fn: "std::backtrace::Backtrace::create", file: "src/libstd/backtrace.rs", line: 308 },
    { fn: "std::backtrace::Backtrace::capture", file: "src/libstd/backtrace.rs", line: 277 },
    { fn: "anyhow::error::<impl anyhow::Error>::msg", file: "/Users/jbr/.cargo/registry/src/github.com-1ecc6299db9ec823/anyhow-1.0.31/src/backtrace.rs", line: 10 },
    { fn: "http_types::error::Error::from_str", file: "src/error.rs", line: 40 },
    { fn: "http_types::error::Error::new_adhoc", file: "src/error.rs", line: 49 },
    { fn: "http_types::private::new_adhoc", file: "src/lib.rs", line: 199 },
    { fn: "backtrace::main", file: "examples/backtrace.rs", line: 2 },
    { fn: "std::rt::lang_start::{{closure}}", file: "/Users/jbr/.rustup/toolchains/nightly-x86_64-apple-darwin/lib/rustlib/src/rust/src/libstd/rt.rs", line: 67 },
    { fn: "std::rt::lang_start_internal::{{closure}}", file: "src/libstd/rt.rs", line: 52 },
    { fn: "std::panicking::try::do_call", file: "src/libstd/panicking.rs", line: 348 },
    { fn: "std::panicking::try", file: "src/libstd/panicking.rs", line: 325 },
    { fn: "std::panic::catch_unwind", file: "src/libstd/panic.rs", line: 394 },
    { fn: "std::rt::lang_start_internal", file: "src/libstd/rt.rs", line: 51 },
    { fn: "std::rt::lang_start", file: "/Users/jbr/.rustup/toolchains/nightly-x86_64-apple-darwin/lib/rustlib/src/rust/src/libstd/rt.rs", line: 67 },
    { fn: "main", line: 0 },
]

[jbr]

With the addition of 9446815, this is technically a breaking change. However, it's safe to say that zero apps were depending on this function, as was always conditionally compiled out regardless of toolchain and RUST_LIB_BACKTRACE. With that commit, here are three different outputs for the same examples/backtrace.rs example above:

$ RUST_LIB_BACKTRACE=0 cargo +nightly run --example backtrace
    Finished dev [unoptimized + debuginfo] target(s) in 0.13s
     Running `target/debug/examples/backtrace`
[examples/backtrace.rs:2] http_types::format_err!("error").backtrace() = None

$ RUST_LIB_BACKTRACE=1 cargo +nightly run --example backtrace
    Finished dev [unoptimized + debuginfo] target(s) in 0.11s
     Running `target/debug/examples/backtrace`
[examples/backtrace.rs:2] http_types::format_err!("error").backtrace() = Some(
    Backtrace [
        { fn: "backtrace::backtrace::libunwind::trace", file: "/Users/runner/.cargo/registry/src/github.com-1ecc6299db9ec823/backtrace-0.3.46/src/backtrace/libunwind.rs", line: 86 },
        { fn: "backtrace::backtrace::trace_unsynchronized", file: "/Users/runner/.cargo/registry/src/github.com-1ecc6299db9ec823/backtrace-0.3.46/src/backtrace/mod.rs", line: 66 },
        { fn: "std::backtrace::Backtrace::create", file: "src/libstd/backtrace.rs", line: 308 },
        { fn: "std::backtrace::Backtrace::capture", file: "src/libstd/backtrace.rs", line: 277 },
        { fn: "anyhow::error::<impl anyhow::Error>::msg", file: "/Users/jbr/.cargo/registry/src/github.com-1ecc6299db9ec823/anyhow-1.0.31/src/backtrace.rs", line: 10 },
        { fn: "http_types::error::Error::from_str", file: "src/error.rs", line: 40 },
        { fn: "http_types::error::Error::new_adhoc", file: "src/error.rs", line: 49 },
        { fn: "http_types::private::new_adhoc", file: "src/lib.rs", line: 199 },
        { fn: "backtrace::main", file: "examples/backtrace.rs", line: 2 },
        { fn: "std::rt::lang_start::{{closure}}", file: "/Users/jbr/.rustup/toolchains/nightly-x86_64-apple-darwin/lib/rustlib/src/rust/src/libstd/rt.rs", line: 67 },
        { fn: "std::rt::lang_start_internal::{{closure}}", file: "src/libstd/rt.rs", line: 52 },
        { fn: "std::panicking::try::do_call", file: "src/libstd/panicking.rs", line: 348 },
        { fn: "std::panicking::try", file: "src/libstd/panicking.rs", line: 325 },
        { fn: "std::panic::catch_unwind", file: "src/libstd/panic.rs", line: 394 },
        { fn: "std::rt::lang_start_internal", file: "src/libstd/rt.rs", line: 51 },
        { fn: "std::rt::lang_start", file: "/Users/jbr/.rustup/toolchains/nightly-x86_64-apple-darwin/lib/rustlib/src/rust/src/libstd/rt.rs", line: 67 },
        { fn: "main", line: 0 },
    ],
)

$ cargo +stable run --example backtrace # stable ignores rust_lib_backtrace
    Finished dev [unoptimized + debuginfo] target(s) in 0.14s
     Running `target/debug/examples/backtrace`
[examples/backtrace.rs:2] http_types::format_err!("error").backtrace() = None

[yoshuawuyts]

Woah; this is a massive hack. Very clever! Can we perhaps add a big prelude to the build file, linking to anyhow's file and this PR? Once Rust gains feature detection we can probably remove the file, but until then having a clear explainer and reference for what it is we're doing in the build.rs file would probably be good 😅

[jbr]

Added a nice big comment explaining how this works. Didn't link to the PR because I think the comment probably explains it better than this PR does thus far, but can tack a link on if you think it helps

[Fishrock123]

The comment helps a lot, this seems pretty sketch but I guess if it does what we need then that's just how it is. I can't say I understand a lot about rust backtraces, so...

[jbr]

I suggest we go with this pr instead of #215, since this is only semver-minor, and treat eyre as a later change

[jbr]

Going to merge this since it was approved and just enables the feature as documented, basically making this a bugfix