Go errors: Is() and As() claim to be recursive, is there any type that implements the error interface and supports this recursion - bug free?

Question

Everywhere I look, the "way" to "wrap" errors in Go is to use fmt.Errof with the %w verb

https://go.dev/blog/go1.13-errors

However, fmt.Errorf does not recursively wrap errors. There is no way to use it to wrap three previously defined errors (Err1, Err2, and Err3) and then check the result by using Is() and get true for each those three errors.

FINAL EDIT:

Thanks to @mkopriva's answer and comments below it, I now have a straightforward way to implement this (although, I am still curious if there is some standard type which does this). In the absence of an example, my attempts at creating one failed. The piece I was missing was adding an Is and As method to my type. Because the custom type needs to contain an error and a pointer to the next error, the custom Is and As methods allows us to compare the error contained in the custom type, rather than the custom type itself.

Here is a working example: https://go.dev/play/p/6BYGgIb728k

Highlights from the above link

type errorChain struct {
    err  error
    next *errorChain
}

//These two functions were the missing ingredient
//Defined this way allows for full functionality even if
//The wrapped errors are also chains or other custom types

func (c errorChain) Is(err error) bool { return errors.Is(c.err, err) }

func (c errorChain) As(target any) bool { return errors.As(c.err, target) }

//Omitting Error and Unwrap methods for brevity

func Wrap(errs ...error) error {
    out := errorChain{err: errs[0]}

    n := &out
    for _, err := range errs[1:] {
        n.next = &errorChain{err: err}
        n = n.next
    }
    return out
}

var Err0 = errors.New("error 0")
var Err1 = errors.New("error 1")
var Err2 = errors.New("error 2")
var Err3 = errors.New("error 3")

func main() {
    //Check basic Is functionality
    errs := Wrap(Err1, Err2, Err3)
    fmt.Println(errs)                            //error 1: error 2: error 3
    fmt.Println(errors.Is(errs, Err0))           //false
    fmt.Println(errors.Is(errs, Err2))           //true
}

While the Go source specifically mentions the ability to define an Is method, the example does not implement it in a way that can solve my issue and the discussion do not make it immediately clear that it would be needed to utilize the recursive nature of errors.Is.

AND NOW BACK TO THE ORIGINAL POST:

Is there something built into Go where this does work?

I played around with making one of my own (several attempts), but ran into undesirable issues. These issues stem from the fact that errors in Go appear to be compared by address. i.e. if Err1 and Err2 point to the same thing, they are the same.

This causes me issues. I can naively get errors.Is and errors.As to work recursively with a custom error type. It is straightforward.

1. Make a type that implements the error interface (has an Error() string method)
2. The type must have a member that represents the wrapped error which is a pointer to its own type.
3. Implement an Unwrap() error method that returns the wrapped error.
4. Implement some method which wraps one error with another

It seems good. But there is trouble.

Since errors are pointers, if I make something like myWrappedError = Wrap(Err1, Err2) (in this case assume Err1 is being wrapped by Err2). Not only will errors.Is(myWrappedError, Err1) and errors.Is(myWrappedError, Err2) return true, but so will errors.Is(Err2, Err1)

Should the need arise to make myOtherWrappedError = Wrap(Err3, Err2) and later call errors.Is(myWrappedError, Err1) it will now return false! Making myOtherWrappedError changes myWrappedError.

I tried several approaches, but always ran into related issues.

Is this possible? Is there a Go library which does this?

NOTE: I am more interested in the presumably already existing right way to do this rather than the specific thing that is wrong with my basic attempt

Edit 3: As suggested by one of the answers, the issue in my first code is obviously that I modify global errors. I am aware, but failed to adequately communicate. Below, I will include other broken code which uses no pointers and modifies no globals.

Edit 4: slight modification to make it work more, but it is still broken

See https://go.dev/play/p/bSytCysbujX

type errorGroup struct {
    err        error
    wrappedErr error
}

//...implemention Unwrap and Error excluded for brevity

func Wrap(inside error, outside error) error {
    return &errorGroup{outside, inside}
}

var Err1 = errorGroup{errors.New("error 1"), nil}
var Err2 = errorGroup{errors.New("error 2"), nil}
var Err3 = errorGroup{errors.New("error 3"), nil}

func main() {
    errs := Wrap(Err1, Err2)
    errs = Wrap(errs, Err3)
    fmt.Println(errs)//error 3: error 2: error 1
    fmt.Println(errors.Is(errs, Err1)) //true
    fmt.Println(errors.Is(errs, Err2)) //false <--- a bigger problem
    fmt.Println(errors.Is(errs, Err3)) //false <--- a bigger problem
}

Edit 2: playground version shortened

See https://go.dev/play/p/swFPajbMcXA for an example of this.

EDIT 1: A trimmed version of my code focusing on the important parts:

type errorGroup struct {
    err        error
    wrappedErr *errorGroup
}

//...implemention Unwrap and Error excluded for brevity

func Wrap(errs ...*errorGroup) (r *errorGroup) {
    r = &errorGroup{}
    for _, err := range errs {
        err.wrappedErr = r
        r = err

    }
    return
}

var Err0 = &errorGroup{errors.New("error 0"), nil}
var Err1 = &errorGroup{errors.New("error 1"), nil}
var Err2 = &errorGroup{errors.New("error 2"), nil}
var Err3 = &errorGroup{errors.New("error 3"), nil}

func main() {
    errs := Wrap(Err1, Err2, Err3)//error 3: error 2: error 1
    fmt.Println(errors.Is(errs, Err1)) //true

    //Creating another wrapped error using the Err1, Err2, or Err3 breaks the previous wrap, errs.
    _ = Wrap(Err0, Err2, Err3)
    fmt.Println(errors.Is(errs, Err1)) //false <--- the problem
}

Answer 1

You can use something like this:

type errorChain struct {
    err  error
    next *errorChain
}

func Wrap(errs ...error) error {
    out := errorChain{err: errs[0]}

    n := &out
    for _, err := range errs[1:] {
        n.next = &errorChain{err: err}
        n = n.next
    }
    return out
}

func (c errorChain) Is(err error) bool {
    return c.err == err
}

func (c errorChain) Unwrap() error {
    if c.next != nil {
        return c.next
    }
    return nil
}

https://go.dev/play/p/6oUGefSxhvF

Answer 2

There arr several approaches but there is one thing that you should keep in mind: if you have multiple errors, you may need to handle it as a slice of errors

For instance, imagine you need to check if all errors are the same, or there is at least one error of certain type you can use the snippet below.

You can extend this concept or use some existing library to handle multierrors

type Errors []error

func (errs Errors) String() string {
  …
}

func (errs Errors) Any(target error) bool{
    for _, err := range errs {
        if errors.Is(err,target) {
            return true
        }
    }
    return false
}

func (errs Errors) All(target error) bool{
    if len(errs) == 0 { return false }
    
    for _, err := range errs {
        if !errors.Is(err,target) {
            return false
        }
    }
    return true
}

Answer 3

Your code modifies package-global error values, so it is inherently broken. This defect has nothing to do with Go's error handling mechanics.

Per the documentation you linked, there are two error-handling helpers: Is, and As. Is lets you recursively unwrap an error, looking for a specific error value, which is necessarily a package global for this to be useful. As, on the other hand, lets you recursively unwrap an error looking for any wrapped error value of a given type.

How does wrapping work? You wrap error A in a new error value B. A Wrap() helper would necessarily return a new value, as fmt.Errorf does in the examples in the linked documentation. A Wrap helper should never modify the value of the error being wrapped. That value should be considered immutable. In fact, in any normal implementation, the value would be of type error, so that you can wrap any error, rather than just wrapping concentric values of your custom error type in each other; and, in that case, you have no access to the fields of the wrapped error to modify them anyway. Essentially, Wrap should be roughly:

func Wrap(err error) error {
    return &errGroup{err}
}

And that's it. That's not very useful, because your implementation of errGroup doesn't really do anything - it provides no details about the error that occurred, it's just a container for other errors. For it to have value, it should have a string error message, or methods like some other error types' IsNotFound, or something that makes it more useful than just using error and fmt.Errorf.

Based on the usage in your example code, it also looks like you're presuming the use case is to say "I want to wrap A in B in C", which I've never seen in the wild and I cannot think of any scenario where that would be needed. The purpose of wrapping is to say "I've recieved error A, I'm going to wrap it in error B to add context, and return it". The caller might wrap that error in error C, and so on, which is what makes recursive wrapping valuable.

For example: https://go.dev/play/p/XeoONx19dgX

Answer 4

Instead of chaining/wrapping, you will "soon" (Go 1.20, as seen in Go 1.20-rc1 in Dec. 2022) be able to return a slice/tree of errors.
(In the meantime, mdobak/go-xerrors is a good alternative)

---

The release note explains:

Wrapping multiple errors

Go 1.20 expands support for error wrapping to permit an error to wrap multiple other errors.

An error e can wrap more than one error by providing an Unwrap method that returns a []error.

The errors.Is and errors.As functions have been updated to inspect multiply wrapped errors.

The fmt.Errorf function now supports multiple occurrences of the %w format verb, which will cause it to return an error that wraps all of those error operands.

The new function errors.Join returns an error wrapping a list of errors.

---

That comes from:

proposal: errors: add support for wrapping multiple errors

Background

Since Go 1.13, an error may wrap another by providing an Unwrap method returning the wrapped error.
The errors.Is and errors.As functions operate on chains of wrapped errors.

A common request is for a way to combine a list of errors into a single error.

Proposal

An error wraps multiple errors if its type has the method

Unwrap() []error

Reusing the name Unwrap avoids ambiguity with the existing singular Unwrap method.
Returning a 0-length list from Unwrap means the error doesn't wrap anything.

Callers must not modify the list returned by Unwrap.
The list returned by Unwrap must not contain any nil errors.

We replace the term "error chain" with "error tree".

The errors.Is and errors.As functions are updated to unwrap multiple errors.

• Is reports a match if any error in the tree matches.
• As finds the first matching error in a inorder preorder traversal of the tree.

The errors.Join function provides a simple implementation of a multierr.
It does not flatten errors.

// Join returns an error that wraps the given errors.
// Any nil error values are discarded.
// The error formats as the text of the given errors, separated by newlines.
// Join returns nil if errs contains no non-nil values.
func Join(errs ...error) error

The fmt.Errorf function permits multiple instances of the %w formatting verb.

The errors.Unwrap function is unaffected: It returns nil when called on an error with an Unwrap() []error method.

Why should this be in the standard library?

This proposal adds something which cannot be provided outside the standard library: Direct support for error trees in errors.Is and errors.As.

Existing combining errors operate by providing Is and As methods which inspect the contained errors, requiring each implementation to duplicate this logic, possibly in incompatible ways.
This is best handled in errors.Is and errors.As, for the same reason those functions handle singular unwrapping.

In addition, this proposal provides a common method for the ecosystem to use to represent combined errors, permitting interoperation between third-party implementations.

So far (Sept. 2022) this proposal seems a likely accept has been accepted!

CL 432575 starts the implementation.