Home

Awesome

xgo

Go Reference Go Report Card Slack Widget Go Coverage CI Awesome Go

English | 简体中文

xgo provides all-in-one test utilities for golang, including:

As for the monkey patching part, xgo works as a preprocessor for go run,go build, and go test(see our blog).

See Quick Start and Documentation for more details.

Xgo Test Explorer

Xgo Test Explorer

By the way, I promise you this is an interesting project.

Installation

go install github.com/xhd2015/xgo/cmd/xgo@latest

Verify the installation:

xgo version
# output:
#   1.0.x

xgo help
# output: help messages

If xgo is not found, you may need to check if $GOPATH/bin is added to your PATH variable.

For CI jobs like github workflow, see doc/INSTALLATION.md.

Requirement

xgo requires at least go1.17 to compile.

There is no specific limitation on OS and Architecture.

All OS and Architectures are supported by xgo as long as they are supported by go.

x86x86_64 (amd64)arm64any other Arch...
LinuxYYYY
WindowsYYYY
macOSYYYY
any other OS...YYYY

Quick Start

Let's write a unit test with xgo:

  1. Ensure you have installed xgo by following the Installation section, and verify the installation with:
xgo version
# output
#   1.0.x
  1. Init a go project:
mkdir demo
cd demo
go mod init demo
  1. Add demo_test.go:
package demo_test

import (
	"testing"

	"github.com/xhd2015/xgo/runtime/mock"
)

func MyFunc() string {
	return "my func"
}

func TestFuncMock(t *testing.T) {
	mock.Patch(MyFunc, func() string {
		return "mock func"
	})
	text := MyFunc()
	if text != "mock func" {
		t.Fatalf("expect MyFunc() to be 'mock func', actual: %s", text)
	}
}
  1. Get the github.com/xhd2015/xgo/runtime dependency:
go get github.com/xhd2015/xgo/runtime
  1. Test the code:
xgo test -v ./
# NOTE: xgo will take some time to setup for the first time

Output:

=== RUN   TestFuncMock
--- PASS: TestFuncMock (0.00s)
PASS
ok      demo

NOTE: xgo is used to test your code, not just go.

Under the hood, xgo preprocess your code to add mock hooks, and then calls go to do remaining jobs.

The above code can be found at doc/demo.

API

Patch

Patch given function within current goroutine.

The API:

Cheatsheet:

// package level func
mock.Patch(SomeFunc, mockFunc)

// per-instance method
// only the bound instance `v` will be mocked
// `v` can be either a struct or an interface
mock.Patch(v.Method, mockMethod)

// per-TParam generic function
// only the specified `int` version will be mocked
mock.Patch(GenericFunc[int], mockFuncInt)

// per TParam and instance generic method
v := GenericStruct[int]
mock.Mock(v.Method, mockMethod)

// closure can also be mocked
// less used, but also supported
mock.Mock(closure, mockFunc)

Parameters:

Scope:

NOTE: fn and replacer should have the same signature.

Return:

Patch replaces the given fn with replacer in current goroutine. It will remove the replacer once current goroutine exits.

Example:

package patch_test

import (
	"testing"

	"github.com/xhd2015/xgo/runtime/mock"
)

func greet(s string) string {
	return "hello " + s
}

func TestPatchFunc(t *testing.T) {
	mock.Patch(greet, func(s string) string {
		return "mock " + s
	})

	res := greet("world")
	if res != "mock world" {
		t.Fatalf("expect patched result to be %q, actual: %q", "mock world", res)
	}
}

NOTE: Patch and Mock(below) supports top-level variables and consts, see runtime/mock/MOCK_VAR_CONST.md.

Notice for mocking stdlib: There are different modes for mocking stdlib functions,see runtime/mock/stdlib.md.

Mock

runtime/mock also provides another API called Mock, which is similar to Patch.

The the only difference between them lies in the the second parameter: Mock accepts an interceptor.

Mock can be used where Patch cannot be used, such as functions with unexported type.

API details: runtime/mock/README.md

The Mock API:

Parameters:

Interceptor Signature: func(ctx context.Context, fn *core.FuncInfo, args core.Object, results core.Object) error

There are other 2 APIs can be used to setup mock based on name, check runtime/mock/README.md for more details.

Method mock example:

type MyStruct struct {
    name string
}
func (c *MyStruct) Name() string {
    return c.name
}

func TestMethodMock(t *testing.T){
    myStruct := &MyStruct{
        name: "my struct",
    }
    otherStruct := &MyStruct{
        name: "other struct",
    }
    mock.Mock(myStruct.Name, func(ctx context.Context, fn *core.FuncInfo, args core.Object, results core.Object) error {
        results.GetFieldIndex(0).Set("mock struct")
        return nil
    })

    // myStruct is affected
    name := myStruct.Name()
    if name!="mock struct"{
        t.Fatalf("expect myStruct.Name() to be 'mock struct', actual: %s", name)
    }

    // otherStruct is not affected
    otherName := otherStruct.Name()
    if otherName!="other struct"{
        t.Fatalf("expect otherStruct.Name() to be 'other struct', actual: %s", otherName)
    }
}

Trace

Trace might be the most powerful tool provided by xgo, this blog has a more thorough example: https://blog.xhd2015.xyz/posts/xgo-trace_a-powerful-visualization-tool-in-go

It is painful when debugging with a deep call stack.

Trace addresses this issue by collecting the hierarchical stack trace and stores it into file for later use.

Needless to say, with Trace, debug becomes less usual:

package trace_test

import (
    "fmt"
    "testing"
)

func TestTrace(t *testing.T) {
    A()
    B()
    C()
}

func A() { fmt.Printf("A\n") }
func B() { fmt.Printf("B\n");C(); }
func C() { fmt.Printf("C\n") }

Run with xgo:

# run the test
# this will write the trace into TestTrace.json
# --strace represents stack trace
xgo test --strace ./

# view the trace
xgo tool trace TestTrace.json

Output: trace html

Another more complicated example from runtime/test/stack_trace/update_test.go: trace html

Real world examples:

Trace helps you get started to a new project quickly.

By default, Trace will write traces to a temp directory under current working directory. This behavior can be overridden by setting XGO_TRACE_OUTPUT to different values:

Besides the --strace flag, xgo allows you to define which span should be collected, using trace.Begin():

import "github.com/xhd2015/xgo/runtime/trace"

func TestTrace(t *testing.T) {
    A()
    finish := trace.Begin()
    defer finish()
    B()
    C()
}

The trace will only include B() and C().

Trap

Xgo preprocess the source code and IR(Intermediate Representation) before invoking go, providing a chance for user to intercept any function when called.

Trap allows developer to intercept function execution on the fly.

Trap is the core of xgo as it is the basis of other abilities like Mock and Trace.

The following example logs function execution trace by adding a Trap interceptor:

(check test/testdata/trap/trap.go for more details.)

package main

import (
    "context"
    "fmt"

    "github.com/xhd2015/xgo/runtime/core"
    "github.com/xhd2015/xgo/runtime/trap"
)

func init() {
    trap.AddInterceptor(&trap.Interceptor{
        Pre: func(ctx context.Context, f *core.FuncInfo, args core.Object, results core.Object) (interface{}, error) {
            if f.Name == "A" {
                fmt.Printf("trap A\n")
                return nil, nil
            }
            if f.Name == "B" {
                fmt.Printf("abort B\n")
                return nil, trap.ErrAbort
            }
            return nil, nil
        },
    })
}

func main() {
    A()
    B()
}

func A() {
    fmt.Printf("A\n")
}

func B() {
    fmt.Printf("B\n")
}

Run with go:

go run ./
# output:
#   A
#   B

Run with xgo:

xgo run ./
# output:
#   trap A
#   A
#   abort B

AddInterceptor() add given interceptor to either global or local, depending on whether it is called from init or after init:

When AddInterceptor() is called after init, it will return a dispose function to clear the interceptor earlier before current goroutine exits.

Example:

func main(){
    clear := trap.AddInterceptor(...)
    defer clear()
    ...
}

Trap also have a helper function called Direct(fn), which can be used to bypass any trap and mock interceptors, calling directly into the original function.

Tools

Test Explorer

The xgo e sub command will open a test explorer UI in the browser, provide an easy way to test and debug go code.

With the test explorer, xgo test is used instead of go test, to enable mocking functionalities.

$ xgo e
Server listen at http://localhost:7070
<img width="1792" alt="test-explorer" src="https://github.com/xhd2015/xgo/assets/14964938/f8689a2d-b422-4101-acef-0aa41ad7e246">

P.S.: xgo e is an alias for xgo tool test-explorer.

For help, run xgo e help.

See doc/test-explorer/README.md for configuration details.

Incremental Coverage

The xgo tool coverage sub command extends go's builtin go tool cover for better visualization.

First, run go test or xgo test to get a coverage profile:

go test -cover -coverpkg ./... -coverprofile cover.out ./...

Then, use xgo to display the coverage:

xgo tool coverage serve cover.out

Output:

coverage

The displayed coverage is a combination of coverage and git diff. By default, only modified lines were shown:

This helps to quickly locate changes that were not covered, and add tests for them incrementally.

Concurrent safety

I know you guys from other monkey patching library suffer from the unsafety implied by these frameworks.

But I guarantee you mocking in xgo is builtin concurrent safe. That means, you can run multiple tests concurrently as long as you like.

Why? when you run a test, you setup some mock, these mocks will only affect the goroutine test running the test. And these mocks get cleared when the goroutine ends, no matter the test passed or failed.

Want to know why? Stay tuned, we are working on internal documentation.

Implementation Details

Working in progress...

See Issue#7 for more details.

This blog has a basic explanation: https://blog.xhd2015.xyz/posts/xgo-monkey-patching-in-go-using-toolexec

Why xgo?

The reason is simple: NO interface.

Yes, no interface, just for mocking. If the only reason to abstract an interface is to mock, then it only makes me feel boring, not working.

Extracting interface just for mocking is never an option to me. To the domain of the problem, it's merely a workaround. It enforces the code to be written in one style, that's why we don't like it.

Monkey patching simply does the right thing for the problem. But existing library are bad at compatibility.

So I created xgo, and hope it will finally take over other solutions to the mocking problem.

Comparing xgo with monkey

The project bouk/monkey, was initially created by bouk, as described in his blog https://bou.ke/blog/monkey-patching-in-go.

In short, it uses a low level assembly hack to replace function at runtime. Which exposes lots of confusing problems to its users as it gets used more and more widely(especially on macOS).

Then it was archived and no longer maintained by the author himself. However, two projects later take over the ASM idea and add support for newer go versions and architectures like Apple M1.

Still, the two does not solve the underlying compatibility issues introduced by ASM, including cross-platform support, the need to write to a read-only section of the execution code and lacking of general mock.

So developers still get annoying failures every now and then.

Xgo managed to solve these problems by avoiding low level hacking of the language itself. Instead, it relies on the IR representation employed by the go compiler.

It does a so-called IR Rewriting on the fly when the compiler compiles the source code. The IR(Intermediate Representation) is closer to the source code rather than the machine code. Thus it is much more stable than the monkey solution.

In conclusion, xgo and monkey are compared as the following:

xgomonkey
TechniqueIRASM
Function MockYY
Unexported Function MockYN
Per-Instance Method MockYN
Per-Goroutine MockYN
Per-Generic Type MockYY
Var MockYN
Const MockYN
Closure MockYY
Stack TraceYN
General TrapYN
CompatiblilityNO LIMITlimited to amd64 and arm64
APIsimplecomplex
Integration Efforteasyhard

Contribution

Want to help contribute to xgo? Great! Check CONTRIBUTING for help.

Evolution of xgo

xgo is the successor of the original go-mock, which works by rewriting go code before compile.

The strategy employed by go-mock works well but causes much longer build time for larger projects due to source code explosion.

However, go-mock is remarkable for it's discovery of Trap, Trace besides Mock, and additional abilities like trapping variable and disabling map randomness.

It is the shoulder which xgo stands on.