Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


A super tiny package for error encapsulation and composition in idiomatic Go. A better and lighter alternative to pkg/errors.
(See below for comparison to pkg/errors)

Note: This package was previously called goerror. Now renamed to errors to be a drop-in replacement for std errors package.

Documentation: Read the source, Luke - it's tiny.


go get -u


  • httperror - Provides HttpError interface and HttpErr type
  • errutils - Provides a set of helpers to ease iteration, and message aggregation


Provides 3 interfaces

  • GoError
  • CodedError
  • ErrorGroup

Provides 3 concrete types

  • GoErr
  • CodedErr
  • ErrGroup


GoError provides a straightforward way to wrap the error interface as Cause. The utils subpackage provides a way to iterate over chained errors and to collect error messages easily.


CodedError adds just one item to Error - Code. It's useful for building any errors with an int code. The httperror subpackage, for example simply builds over this with an added code validation.


ErrGroup provides a clean way to aggregate errors as []error. But unlike other libraries out there - doesn't provides any way to add, combine, or remove errors. It prefers to sticks to Go's idiomatic way of keeping things minimalistic. You can use a Go's own make, append and other slice function to do the same. Once done, you simply use GroupFrom to add your slice to create a group. It's all just slices. The only thing ErrGroup does is provides you an error interface that by default prints a nice message, and a way to retrive all the original errors.


func TestErrors() {
    // New errors

    err := errors.New("some error")
    // Error with codes
    codedErr := errors.NewCoded(42, "compute overwhelmed. crashing")
    // HttpErrors are essentially, coded errors,
    // with http code validation
    httpErr := httperror.New(400, "naughty you! that's not valid", true)

    // Wrap existing existing errors
    wrapped1 := errors.From(errors.New("some other error"))
    wrapped2 := errors.From(wrapped1)
    // With additional message
    wrapper3 := errors.NewWithCause("wrapped as different msg", wrapped2)

    // Intuitively messages:
    // Output 
    // some other error

    // Same output
    // Output
    // some other error

    // Output
    // wrapped as different msg

    // You can get the real err by accessing `Cause`. Or more intuitively,
    // Collect from messages

    msgs := errutils.CollectMsg(wrapped3, nil)
    // msg has [ 'wrapped as different msg', 'some other error' ]

    // Error group
    errs := []error { err, codedErr, httpErr, wrapped1, wrapped2}
    errGroup := errors.GroupFrom(errs)

    // Let's make things interesting
    // Output
    // multiple errors:
    // ...
    // prints all of the given errors
    // ...

    // Or take control yourself
    allErrs = errGroup.Errors()
    msgs = errutils.CollectMsg(allErrs, nil)

A real example in middleware using prasannavl/mchain

func RequestIDMustInitHandler(next mchain.Handler) mchain.Handler {
	f := func(w http.ResponseWriter, r *http.Request) error {
		c := FromRequest(r)
		if _, ok := r.Header[RequestIDHeaderKey]; ok {
			msg := fmt.Sprintf("error: illegal header (%s)", RequestIDHeaderKey)
			return httperror.New(400, msg, true)
		var uid uuid.UUID
		c.RequestID = uid
		return next.ServeHTTP(w, r)
	return mchain.HandlerFunc(f)


Some of the features provided by this package are very similar to what the package pkg/errors offers. So, why do another error package?

  • pkg/errors has a very nice and simple way to chain errors with Cause. However, other than that I tend to disagree with the pattern the pkg/errors package tends to build on. In a language with exceptions, stack traces are critical - and so even though it may be expensive to take them - we try to get it where-ever we can. But Go takes a drastically different approach towards error handling. It's very simple. You just pass a simple structure around. And as such, is very lightweight. And it's very clean because, in order to identify where an error originated from, you can create errors that are unique with codes (which is the one of the purpose of error codes in the first place, beyond internal handling of error such as comparisons). But pkg/errors destroys this "lightness" by calling runtime.Callers everytime you create an error. I find taking stack traces along with every error mindlessly to be wasteful in a language without exceptions, and not idiomatic go.

    This package solves the problems that pkg/errors try to solve, and a lot more - by ErrorGroups, CodedError and utils, all while still retaining more simplicity than pkg/errors.

Other than this, goerror also provides a more complete error encapsulation solution:

  • It provides CodedError pattern by out of the box - which is what a lot of usage requires rather than stack traces.
  • Defines a neat way to group errors with GroupError.
  • It also provides httperror that's built over CodedError, out of the box but as separate package.

How is this more Go idiomatic?

As mentioned, stack-traces are absolutely essential when you work with language with execptions. Because they can throw from anywhere, and you can be catching it elsewhere - you're bound to loose track of where you are very quickly. However the approach Go takes is simple. You check your errors at every surface of an API. So, simple error type with kind can work very nicely. And that doesn't mean you can't or shouldn't take stack traces. You can always just wrap them in higher order functions. I've found it to be a better practice to take traces at component or contextual boundaries, when needed.

Simple composable errors with codes, should be the basic building block of error handling - while a package like pkg/errors that's intended to be a fundamental building block, unfortunately does caller stacks, is structured far more complex internally, and doesn't really help much with composition

I like stack-traces!

Then take them!

It's just not intended to go into the building block. They should be above it. So you have exactly that choice. There are times when stack traces are critical, and there are times when they are just a waste of system resources - but the truth is, error handling helpers like these can never know that regardless of any number of debates or discussions - and so this package does the absolute minimum to act as a building block, yet provides nice composition. Rest can be built over it.

How is this simpler?

Don't let the number of types fool you. Conceptually, it's all just in one file errors.go, which is less than 50 lines long. All it really does is just wrap things up neatly in a type, and allows you to compose them. If you're a reasonably proficient in Go, you should be able to read and understand the whole code of the library in just under 5 minutes - I encourage you to - you should know exactly what happens you instantiate an error type.

Drop-in replacement for std errors?

Yup. It is!


HttpError provides one additional method End that's useful to signify any middleware chain to stop processing.

Combining GoError and ErrorGroup should be sufficient to handle most complex error wrapping, merging scenarios in Go without the use of other packages that add too many whistles which are, in my opinion completely unnecessary - and frankly not idiomatic Go.


This project is licensed under either of the following, at your choice:

Code of Conduct

Contribution to the LiquidState project is organized under the terms of the Contributor Covenant, and as such the maintainer @prasannavl promises to intervene to uphold that code of conduct.


A super tiny package for error encapsulation in idiomatic Go




Apache-2.0, GPL-3.0 licenses found

Licenses found


Code of conduct





No releases published


No packages published