Error handling library that generates and injects context for you.
This library provides three main features:
#[oofs] attribute that generates and injects context to function calls with ? operators.Table of Content:
#[oofs] AttributeOof Error Struct._tag() and ._attach(_)#[async_trait]Below showcases the context injection.
```rust use oofs::{oofs, Oof};
fn outer_fn() -> Result<(), Oof> { let x = 123; let y = "hello world";
inner_fn(x, y)?;
Ok(())
}
fn inner_fn(x: usize, y: &str) -> Result<(), Oof> {
let _ = y.parse::
Ok(())
} ```
Printing error from outer_fn() outputs:
``
inner_fn($0, $1) failed atoofs/tests/basic.rs:6:5`
Parameters: $0: usize = 123 $1: &str = "hello world"
Caused by:
0: y.parse() failed at oofs/tests/basic.rs:17:13
1: invalid digit found in string
```
Now, let's look at a slightly longer example.
Below is an example from oofs/tests/basic.rs.
The example showcases context-generation, tagging, and attaching custom contexts.
```rust use oofs::{oofs, Oof, OofExt};
// Marker type used for tagging. struct RetryTag;
fn application() -> Result<(), Oof> {
if let Err(e) = middlelayer("hello world") {
// Check if any of internal errors is tagged as RetryTag; if so, try again.
if e.tagged_nested::
// If the call fails again, return it.
// Since `?` operator is used, context is generated and injected to the call.
middlelayer("hello world")?;
} else {
return Err(e);
}
}
Ok(())
}
fn middlelayer(text: &str) -> Result
// Passing an expression as arg is also fine.
// All args are evaluated before being debugged in the error.
// Context is generated and injected to both `?`s in this statement.
let ret = my_struct.failing_method(get_value()?)?;
Ok(ret)
}
fn get_value() -> Result
struct MyStruct { field: String, }
// #[oofs] can also be used to impl blocks.
// Context will be injected to all methods that return a Result.
impl MyStruct {
fn failingmethod(&self, x: usize) -> ResultRetryTag.
.attach(x) // attach anything that implements Debug as custom context.
.attach(&self.field) // attach the receiver as attachment to debug.
.attachlazy(|| "extra context")?; // lazily evaluate context; useful for something like || serde_json::to_string(&x).
Ok(ret)
}
} ```
If we print the error from application(), it would output:
``` Retrying middlelayer!
middlelayer($0) failed at oofs/tests/basic.rs:11:13
Parameters: $0: &str = "hello world"
Caused by:
0: mystruct.failingmethod($0) failed at oofs/tests/basic.rs:26:15
Parameters:
$0: usize = 123
1: self.field.parse() failed at `oofs/tests/basic.rs:46:14`
Attachments:
0: 123
1: "hello world"
2: "extra context"
2: invalid digit found in string
```
The error displays chain of methods that fail, their locations in code, the parameters' types and their debug values.
Same will be displayed when you format the error as Debug (i.e. {:?}).
Nice looking error is not all; we also get categorized error handling with tags.
At the source method failing_method, we tag the parse method with RetryTag type.
At the very top level function application, we call e.tagged_nested::<RetryTag> to check any interal calls were tagged with RetryTag.
When the tag is found, we handle the case by calling middlelayer again.
With tagging, we no longer have to go through every error variant at every level.
We just look for the tag we want to handle for, and we handle the tagged error accordingly.
In the above example, we retry calling middlelayer again if RetryTag tag is found.
As shown in the example above, you can tag an error with _tag and detect the tag with tagged and tagged_nested.
```rust struct MyTag;
fn applicationlevel() -> Result<(), Oof> {
if let Err(e) = source() {
if e.taggednested::
...
fn source() -> Result<(), Oof> {
somefn().tag::
Ok(())
} ```
This allows you to categorize errors into different tag groups, and handle for them accordingly. This gives a much better experience when handling errors compared to matching every enum variant in every nested function calls.
Note that you can also tag an error with multiple different tags.
I chose type as tag because types are small, readable and unique. String or usize can lead to having duplicate values by accident.
At some point, you may find the generated context is not enough. After all, it just shows the call that failed, and parameters that were passed to it. It will not capture all the other possibe context information.
You can attach your own context information to the error with _attach and _attach_lazy methods.
```rust
fn outer_fn() -> Result<(), Oof> { let x = 123usize; let y = std::time::Instant::now();
"hello world"
.parse::<usize>()
._attach(&x)
._attach(&y)?;
Ok(())
} ```
Above will print the following error:
``
$0.parse() failed atoofs/tests/basic.rs:10:10`
Parameters: $0: &str = "hello world"
Attachments: 0: 123 1: Instant { t: 11234993365176 }
Caused by: invalid digit found in string ```
_attach takes any type that implements std::fmt::Debug.
_attach_lazy, on the other hand, takes any closure that returns a type that implements ToString.
It can be something &str like ._attach_lazy(|| "some context"), String like ._attach_lazy(|| format!("some context {:?}", x)),
or some function that requires some work to display like ._attach_lazy(|| serde_json::to_string(&x)).
At some point, you also want to return your custom error.
For these cases, you have some options: oof!(...), wrap_err(_), ensure!(...) and ensure_eq!(...).
oof!(...): this is a lot like anyhow! or eyre!; you input to macro like you do for println!.
This returns Oof struct, and you can call methods on the returned Oof like
rust
return oof!("my custom error").tag::<MyTag>().attach(&x).into_res();
into_res() wraps Oof into Result::Err(_).
wrap_err(_): function that wraps a custom error with Oof.
rust
return wrap_err(std::io::Error::new(std::io::ErrorKind::Other, "Some Error")).tag::<MyTag>().into_res();
into_res() wraps Oof into Result::Err(_).
ensure!(...): this is similar to a lot of other libraries with slight differences.
Optionally, you can input custom context message like for format!(...).
Also, you can optionally provide tags and attributes wrapped in braces.
rust
ensure!(false, "custom context with value {:?}", x, {
tags: [MyTag, OtherTag],
attach: [&y, "attachment", Instant::now()],
attach_lazy: [|| serde_json::to_string(&y), || format!("lazy attachment {}", &z)]
});
ensure_eq!(...): this is similar to a lot of other libraries with slight differences.
Optionally, you can input custom context message like for format!(...).
Also, you can optionally provide tags and attributes wrapped in braces.
rust
ensure_eq!(1u8, 2u8, "custom context with value {:?}", x, {
tags: [MyTag, OtherTag],
attach: [&y, "attachment", Instant::now()],
attach_lazy: [|| serde_json::to_string(&y), || format!("lazy attachment {}", &z)]
});
You can pass arguments to the attribute like #[oofs(tags(ThisTag, ThatTag))].
These arguments will be applied to the entire scope of the attribute.
For details about the attribute arguments, see documentation.
location (default: true): enables printing location of code that fails.debug_strategy_disabled (default: false): Disables debugging non-copy-able parameters.
Default behavior is to debug non-copyable parameters for debug mode, but not debugging them for release mode.
debug_strategy_full (default: false): Enables debugging non-copy-able parameters even for release mode.
Default behavior is to debug non-copyable parameters for debug mode, but not debugging them for release mode.
#[oofs] Attribute#[oofs] generates and injects contexts into all statements and expressions that have ? operator.
return Err(...) or last expression without semicolon do not get injected with context.
If the receiver of a method is a variable (i.e. x.some_method()), or a field of a variable (i.e. x.field.some_method()),
values of x or x.field are not displayed. This is because there is no way to determine in the macro whether this receiver
is a reference, mutable reference, or an owned variable.
x.some_method()._attach(&x) to display the value of x in the error.Oof Error StructOof does not implement From<E> where E: std::error::Error, and so must be built by attribute macro.
So, if you don't include #[oofs], it will throw a comiler error;
this is intentional because it will catch the user's eyes and force them to include the attribute.
Unlike anyhow::Error or eyre::Report, Oof does implement std::error::Error.
This is nice because it makes it compatible with these boxed error types.
For example, this works:
```rust
fn outerfn() -> Result<(), anyhow::Error> { innerfn()?; Ok(()) } ```
It works since ? operator will implicitly convert Oof into anyhow::Error.
._tag() and ._attach(_)In the basic examples above, you may have noticed that all the methods used for oof starts with an underscore; you could call them 'meta-methods' as they do not affect the logic, but only the result that is returned.
The reason for this is that there has to be a way for the macro to differentiate between functional methods and meta methods.
This is because macro would also try to include these meta methods as part of the displayed method chain, and things like
_attach(x) would be displayed twice in Parameters section and Attachments section.
This may seem disturbing and unnatural at first; it was for me, too. But after trying it out, I got used to it; and now I think I like it because I can easily differentiate between functional methods and meta methods.
I apologize for the inconvenience, and please let me know if there was a better way to do this.
One of the pain points while creating the library was to lazy-load values of copyable arguments and instantly load values of non-copyable arguments at compile time. I figured out how to do this with a cool rust hack.
Now, should the default behavior be to always instantly load values of non-copyable arguments? this could incur unwanted performance costs, as it would load them for non-error cases.
As a compromise, I made it so that, for debug mode, it will instantly load values of non-copyable arguments; and, for release mode, it will not load values of non-copyable arguments.
You can change this behavior with features debug_strategy_disabled and debug_strategy_full.
debug_strategy_disabled will disable loading values of non-copyable arguments even for debug mode.
debug_strategy_full will enable loading values of non-copyable arguments even for releaes mode.
#[async_trait]#[async_trait] parses and converts async fn in traits into fn -> Box<Future<Output = Result<...>>>.
Since #[oofs] by default only applies context injection to methods that returns Result<_, _>,
it will not apply injection once #[async_trait] is applied.
There are two ways to deal with this:
Place #[oofs] above #[async_trait], so that oofs is applied first, then #[async_trait].
```rust
impl Trait for Struct { ... } ```
In the impl block, place #[oofs(closures, async_blocks)] above fn ..., and oofs attr will tell the macro to apply injection regardless,
and closures and async_blocks will tell the macro to apply injection for closures and async blocks, which are disabled by default.
```rust
impl Trait for Struct {
async fn do_something() -> Result<(), _> { ... } } ```
This library is still very much WIP.
I plan to test the error handling for performance, optimize memory footprints of errors,
and implement attribute arguments like #[oofs(tag(MyTag))], #[oofs(skip)], etc.
Also, it does not inject context into closures and async blocks. I plan to add attribute args like #[oofs(closures)] and #[oofs(async_blocks)] to enable injecting context to closures and async blocks.