Stacked Errors

A crate for high level error propogation with software controlled backtraces that are entirely independent of the RUST_BACKTRACE system.

In Rust development, major crates will often have their own error enums that work well in their own specialized domain, but when orchestrating many domains together we run into issues. map_err is very annoying to work with. In async call stacks or cases where we can't easily control RUST_BACKTRACE, we run into an especially annoying problem where the same kind of error can be returned from multiple places, and we are sometimes forced into println debugging to find out where it is actually from. This crate introduces the StackableErr trait and a "stackable" error type that allows for both software-defined error backtraces and easily converting errors into the stackable error type.

This crate is similar to eyre, but has a more efficient internal layout with a ThinVec array of SmallBoxes, works with no_std, implements core::error::Error, and more.

Some partial examples of what using the crate looks like:

f.map_err(|e| Error::from_err(e))?;
// replace the above with
f.stack()?; // uses `#[track_caller]` when an error is being propagated

let dir = self
    .path
    .parent()
    .stack_err("FileOptions::preacquire() -> empty path")?
    .to_str()
    .stack_err("bad OsStr conversion")?;

// arbitrary things implementing `Display + Send + Sync + 'static` can be stacked
f.stack_err(arbitrary)?;
// readily swappable with `anyhow` and `eyre` due to extra method, trait, and
// struct aliases
f.wrap_err(arbitrary)?;
f.context(arbitrary)?;

// The trait is implemented for options so that we don't need `OptionExt` like
// `eyre` does
option.take()
    .stack_err("`Struct` has already been taken")?
    .wait_with_output()
    .await
    .stack_err_with(|| {
        format!("{self:?}.xyz() -> failed when waiting")
    })?;

return Err(Error::from_err(format!(
    "failure of {x:?} to complete"
)))
// replace the above with
bail!("failure of {x:?} to complete")

// when the error type is already `stacked_errors::Error` you can do this if it is
// preferable over `map`
return match ... {
    Ok(ok) => {
        ...
    }
    Err(e) => Err(e.add_err(format!("myfunction(.., host: {host})"))),
}

use stacked_errors::{bail, Error, Result, StackableErr};

// Note that `Error` uses `ThinVec` internally, which means that it often
// takes up only the stack space of a `usize` or the size of the `T` plus
// a byte.
fn innermost(s: &str) -> Result<u8> {
    if s == "return error" {
        bail!("bottom level `StrErr`")
    }
    if s == "parse invalid" {
        // This is the common case where we have some external
        // crate function that returns a `Result<..., E: Error>`. We
        // usually call `StackableErr::stack_err` if we want to attach
        // some message to it right away (it is called with a closure
        // so that it doesn't have impact on the `Ok` cases). Otherwise, we
        // just call `StackableErr::stack` so that just the location is
        // pushed on the stack. We can then use `?` directly.

        let _: () = ron::from_str("invalid")
            .stack_err_with(|| format!("parsing error with \"{s}\""))?;
    }
    Ok(42)
}

fn inner(s: &str) -> Result<u16> {
    // Chainable with other combinators. Use `stack_err` with a message for
    // propogating up the stack when the error is something that should
    // have some mid layer information attached for it for quick diagnosis
    // by the user. Otherwise use just `stack` which will also do error
    // conversion if necessary, avoiding needing to wrangle with `map_err`.

    let x = innermost(s)
        .map(|x| u16::from(x))
        .stack_err_with(|| format!("error from innermost(\"{s}\")"))?;
    Ok(x)
}

fn outer(s: &str) -> Result<u64> {
    // ...

    let x = inner(s).stack()?;

    // ...
    Ok(u64::from(x))
}

let res = format!("{:?}", outer("valid"));
assert_eq!(res, "Ok(42)");

// The line numbers are slightly off because this is a doc test.
// In order from outer to the innermost call, it lists the location of the
// `stack` call from `outer`, the location of `stack_err` from `inner`,
// the associated error message, the location of either the `Error::from`
// or `stack_err` from `innermost`, and finally the root error message.

// note that debug mode (used when returning errors from the main function)
// includes terminal styling
println!("{:?}", outer("return error"));
let res = format!("{}", outer("return error").unwrap_err());
assert_eq!(
    res,
    r#"
  at src/lib.rs 45:22
    error from innermost("return error") at src/lib.rs 38:10
    bottom level `StrErr` at src/lib.rs 12:9"#
);

println!("{:?}", outer("parse invalid"));
let res = format!("{}", outer("parse invalid").unwrap_err());
assert_eq!(
    res,
    r#"
  at src/lib.rs 45:22
    error from innermost("parse invalid") at src/lib.rs 38:10
    parsing error with "parse invalid" at src/lib.rs 24:14
    1:1: Expected unit"#
);

// in commonly used functions you may want `_locationless` to avoid adding
// on unnecessary information if the location is already being added on
return Err(e.add_err_locationless(ErrorKind::TimeoutError)).stack_err_with(|| {
    format!(
        "wait_for_ok(num_retries: {num_retries}, delay: {delay:?}) timeout, \
         last error stack was:"
    )
})