A Makefile for your Go project (2019)

Vincent Bernat

My most loathed feature of Go was the mandatory use of GOPATH: I do not want to put my own code next to its dependencies. I was not alone and people devised tools or crafted their own Makefile to avoid organizing their code around GOPATH.

Hopefully, since Go 1.11, it is possible to use Go’s modules to manage dependencies without relying on GOPATH. First, you need to convert your project to a module:1

$ go mod init hellogopher
go: creating new go.mod: module hellogopher
$ cat go.mod
module hellogopher

Then, you can invoke the usual commands, like go build or go test. The go command resolves imports by using versions listed in go.mod. When it runs into an import of a package not present in go.mod, it automatically looks up the module containing that package using the latest version and adds it.

$ go test ./...
go: finding github.com/spf13/cobra v0.0.5
go: downloading github.com/spf13/cobra v0.0.5
?       hellogopher     [no test files]
?       hellogopher/cmd [no test files]
ok      hellogopher/hello       0.001s
$ cat go.mod
module hellogopher

require github.com/spf13/cobra v0.0.5

If you want a specific version, you can either edit go.mod or invoke go get:

$ go get github.com/spf13/cobra@v0.0.4
go: finding github.com/spf13/cobra v0.0.4
go: downloading github.com/spf13/cobra v0.0.4
$ cat go.mod
module hellogopher

require github.com/spf13/cobra v0.0.4

Add go.mod to your version control system. Optionally, you can also add go.sum as a safety net against overriden tags. If you really want to vendor the dependencies, you can invoke go mod vendor and add the vendor/ directory to your version control system.

Thanks to the modules, in my opinion, Go’s dependency management is now on a par with other languages, like Ruby. While it is possible to run day-to-day operations—building and testing—with only the go command, a Makefile can still be useful to organize common tasks, a bit like Python’s setup.py or Ruby’s Rakefile. Let me describe mine.

Using third-party tools⚓︎

Most projects need some third-party tools for testing or building. We can either expect them to be already installed or compile them on the fly. For example, here is how code linting is done with Golint:

BIN = $(CURDIR)/bin
$(BIN):
    @mkdir -p $@
$(BIN)/%: | $(BIN)
    @tmp=$$(mktemp -d); \
       env GO111MODULE=off GOPATH=$$tmp GOBIN=$(BIN) go get $(PACKAGE) \
        || ret=$$?; \
       rm -rf $$tmp ; exit $$ret

$(BIN)/golint: PACKAGE=golang.org/x/lint/golint

GOLINT = $(BIN)/golint
lint: | $(GOLINT)
    $(GOLINT) -set_exit_status ./...

The first block defines how a third-party tool is built: go get is invoked with the package name matching the tool we want to install. We do not want to pollute our dependency management and therefore, we are working in an empty GOPATH. The generated binaries are put in bin/.

The second block extends the pattern rule defined in the first block by providing the package name for golint. Additional tools can be added by just adding another line like this.

The last block defines the recipe to lint the code. The default linting tool is the golint built using the first block but it can be overrided with make GOLINT=/usr/bin/golint.

Tests⚓︎

Here are some rules to help running tests:

TIMEOUT  = 20
PKGS     = $(or $(PKG),$(shell env GO111MODULE=on $(GO) list ./...))
TESTPKGS = $(shell env GO111MODULE=on $(GO) list -f \
            '{{ if or .TestGoFiles .XTestGoFiles }}{{ .ImportPath }}{{ end }}' \
            $(PKGS))

TEST_TARGETS := test-default test-bench test-short test-verbose test-race
test-bench:   ARGS=-run=__absolutelynothing__ -bench=.
test-short:   ARGS=-short
test-verbose: ARGS=-v
test-race:    ARGS=-race
$(TEST_TARGETS): test
check test tests: fmt lint
    go test -timeout $(TIMEOUT)s $(ARGS) $(TESTPKGS)

A user can invoke tests in different ways:

  • make test runs all tests;
  • make test TIMEOUT=10 runs all tests with a timeout of 10 seconds;
  • make test PKG=hellogopher/cmd only runs tests for the cmd package;
  • make test ARGS="-v -short" runs tests with the specified arguments;
  • make test-race runs tests with race detector enabled.

go test includes a test coverage tool. Unfortunately, it only handles one package at a time and you have to explicitely list the packages to be instrumented, otherwise the instrumentation is limited to the currently tested package. If you provide too many packages, the compilation time will skyrocket. Moreover, if you want an output compatible with Jenkins, you need some additional tools.

COVERAGE_MODE    = atomic
COVERAGE_PROFILE = $(COVERAGE_DIR)/profile.out
COVERAGE_XML     = $(COVERAGE_DIR)/coverage.xml
COVERAGE_HTML    = $(COVERAGE_DIR)/index.html
test-coverage-tools: | $(GOCOVMERGE) $(GOCOV) $(GOCOVXML) # ❶
test-coverage: COVERAGE_DIR := $(CURDIR)/test/coverage.$(shell date -u +"%Y-%m-%dT%H:%M:%SZ")
test-coverage: fmt lint test-coverage-tools
    @mkdir -p $(COVERAGE_DIR)/coverage
    @for pkg in $(TESTPKGS); do \ # ❷
        go test \
            -coverpkg=$$(go list -f '{{ join .Deps "\n" }}' $$pkg | \
                    grep '^$(MODULE)/' | \
                    tr '\n' ',')$$pkg \
            -covermode=$(COVERAGE_MODE) \
            -coverprofile="$(COVERAGE_DIR)/coverage/`echo $$pkg | tr "/" "-"`.cover" $$pkg ;\
     done
    @$(GOCOVMERGE) $(COVERAGE_DIR)/coverage/*.cover > $(COVERAGE_PROFILE)
    @go tool cover -html=$(COVERAGE_PROFILE) -o $(COVERAGE_HTML)
    @$(GOCOV) convert $(COVERAGE_PROFILE) | $(GOCOVXML) > $(COVERAGE_XML)

First, we define some variables to let the user override them. In ❶, we require the following tools—built like golint previously:

  • gocovmerge merges profiles from different runs into a single one;
  • gocov-xml converts a coverage profile to the Cobertura format, for Jenkins;
  • gocov is needed to convert a coverage profile to a format handled by gocov-xml.

In ❷, for each package to test, we run go test with the -coverprofile argument. We also explicitely provide the list of packages to instrument to -coverpkg by using go list to get a list of dependencies for the tested package and keeping only our owns.

Build⚓︎

Another useful recipe is to build the program. While this could be done with just go build, it is not uncommon to have to specify build tags, additional flags, or to execute supplementary build steps. In the following example, the version is extracted from Git tags. It will replace the value of the Version variable in the hellogopher/cmd package.

VERSION ?= $(shell git describe --tags --always --dirty --match=v* 2> /dev/null || \
            echo v0)
all: fmt lint | $(BIN)
    go build \
        -tags release \
        -ldflags '-X hellogopher/cmd.Version=$(VERSION)' \
        -o $(BIN)/hellogopher main.go

The recipe also runs code formatting and linting.


The excerpts provided in this post are a bit simplified. Have a look at the final result for more perks, including fancy output and integrated help!


  1. For an application not meant to be used as a library, I prefer to use a short name instead of a name derived from an URL, like github.com/vincentbernat/hellogopher. It makes it easier to read import sections:

    import (
            "fmt"
            "os"
    
            "hellogopher/cmd"
    
            "github.com/pkg/errors"
            "github.com/spf13/cobra"
    )
    
    ↩︎

Share this article