rust/library/core/src/error.md

Interfaces for working with Errors.

Error Handling In Rust

The Rust language provides two complementary systems for constructing / representing, reporting, propagating, reacting to, and discarding errors. These responsibilities are collectively known as "error handling." The components of the first system, the panic runtime and interfaces, are most commonly used to represent bugs that have been detected in your program. The components of the second system, Result, the error traits, and user defined types, are used to represent anticipated runtime failure modes of your program.

The Panic Interfaces

The following are the primary interfaces of the panic system and the responsibilities they cover:

• [panic!] and panic_any (Constructing, Propagated automatically)
• PanicInfo (Reporting)
• set_hook, take_hook, and #[panic_handler] (Reporting)
• catch_unwind and resume_unwind (Discarding, Propagating)

The following are the primary interfaces of the error system and the responsibilities they cover:

• [Result] (Propagating, Reacting)
• The [Error] trait (Reporting)
• User defined types (Constructing / Representing)
• match and downcast (Reacting)
• The question mark operator (?) (Propagating)
• The partially stable Try traits (Propagating, Constructing)
• Termination (Reporting)

Converting Errors into Panics

The panic and error systems are not entirely distinct. Often times errors that are anticipated runtime failures in an API might instead represent bugs to a caller. For these situations the standard library provides APIs for constructing panics with an Error as it's source.

• [Result::unwrap]
• [Result::expect]

These functions are equivalent, they either return the inner value if the Result is Ok or panic if the Result is Err printing the inner error as the source. The only difference between them is that with expect you provide a panic error message to be printed alongside the source, whereas unwrap has a default message indicating only that you unwrapped an Err.

Of the two, expect is generally preferred since its msg field allows you to convey your intent and assumptions which makes tracking down the source of a panic easier. unwrap on the other hand can still be a good fit in situations where you can trivially show that a piece of code will never panic, such as "127.0.0.1".parse::<std::net::IpAddr>().unwrap() or early prototyping.

Common Message Styles

There are two common styles for how people word expect messages. Using the message to present information to users encountering a panic ("expect as error message") or using the message to present information to developers debugging the panic ("expect as precondition").

In the former case the expect message is used to describe the error that has occurred which is considered a bug. Consider the following example:

// Read environment variable, panic if it is not present
let path = std::env::var("IMPORTANT_PATH").unwrap();

In the "expect as error message" style we would use expect to describe that the environment variable was not set when it should have been:

let path = std::env::var("IMPORTANT_PATH")
    .expect("env variable `IMPORTANT_PATH` is not set");

In the "expect as precondition" style, we would instead describe the reason we expect the Result should be Ok. With this style we would prefer to write:

let path = std::env::var("IMPORTANT_PATH")
    .expect("env variable `IMPORTANT_PATH` should be set by `wrapper_script.sh`");

The "expect as error message" style does not work as well with the default output of the std panic hooks, and often ends up repeating information that is already communicated by the source error being unwrapped:

thread 'main' panicked at 'env variable `IMPORTANT_PATH` is not set: NotPresent', src/main.rs:4:6

In this example we end up mentioning that an env variable is not set, followed by our source message that says the env is not present, the only additional information we're communicating is the name of the environment variable being checked.

The "expect as precondition" style instead focuses on source code readability, making it easier to understand what must have gone wrong in situations where panics are being used to represent bugs exclusively. Also, by framing our expect in terms of what "SHOULD" have happened to prevent the source error, we end up introducing new information that is independent from our source error.

thread 'main' panicked at 'env variable `IMPORTANT_PATH` should be set by `wrapper_script.sh`: NotPresent', src/main.rs:4:6

In this example we are communicating not only the name of the environment variable that should have been set, but also an explanation for why it should have been set, and we let the source error display as a clear contradiction to our expectation.

Hint: If you're having trouble remembering how to phrase expect-as-precondition style error messages remember to focus on the word "should" as in "env variable should be set by blah" or "the given binary should be available and executable by the current user".