Failpoints are getting executed while running tests using `go test -v`

使用 samplestruct.go、main.go 和 sample_test.go，通过比较 go test -v 与构建出的 ./modv1 可执行文件来复现该报告。检查在 TestFailpoint 期间如何设置和使用 GO_FAILPOINTS，并验证 failpoint 是否会在测试二进制文件中运行。完成意味着已解释该行为，并且通过可复现的测试或清晰的文档覆盖了测试时的预期行为。

Bug Report

1. What did you do? If possible, provide a recipe for reproducing the error.

I have the following setup (all in the same directory):
go.mod

module example.com/modv1

go 1.20

require (
	github.com/pingcap/failpoint v0.0.0-20220801062533-2eaa32854a6c
	github.com/stretchr/testify v1.8.4
)

require (
	github.com/davecgh/go-spew v1.1.1 // indirect
	github.com/pingcap/errors v0.11.4 // indirect
	github.com/pmezard/go-difflib v1.0.0 // indirect
	gopkg.in/yaml.v3 v3.0.1 // indirect
)

samplestruct.go

package main

import (
	"log"

	"github.com/pingcap/failpoint"
)

type SampleStruct struct {
	cnt int
}

func (s *SampleStruct) increment() {
	if _, _err_ := failpoint.Eval(_curpkg_("testPanic")); _err_ == nil {
		log.Printf("[SampleStruct] SUCCESS!!! Executed testPanic FP: The count for s is %v", s.cnt)
	}
	s.cnt++
}

func (s *SampleStruct) decrement() {
	s.cnt--
}

func (s *SampleStruct) incrementFP() string {
	log.Printf("[SampleStruct] Executing incrementFP with old cnt %v", s.cnt)
	s.cnt++
	log.Printf("[SampleStruct] Executing incrementFP with new cnt %v", s.cnt)
	if _, _err_ := failpoint.Eval(_curpkg_("doubleIncrement")); _err_ == nil {
		s.cnt++
		log.Printf("[SampleStruct] SUCCESS!!! Executed doubleIncrement FP. The count for s is %v", s.cnt)
	}
	return "single"
}

main.go

package main

import (
	"fmt"
	"log"
	"os"
)

func main() {
	log.Printf(fmt.Sprintf("[main] This is main function being executed in an executable at path %v", os.Args[0]))
	//os.Setenv("Go_FAILPOINTS", "main/doubleIncrement=return(true);main/testPanic=return(true)")
	log.Printf(fmt.Sprintf("[main] After FP code in main function"))

	sampleStruct := SampleStruct{cnt: 1}
	log.Printf("[main] sampleStruct cnt before increment is %v", sampleStruct.cnt)
	sampleStruct.increment()
	log.Printf("[main] sampleStruct cnt after increment is %v", sampleStruct.cnt)
	log.Printf("[main] sampleStruct cnt before double increment is %v", sampleStruct.cnt)
	sampleStruct.incrementFP()
	log.Printf("[main] sampleStruct cnt after double increment is %v", sampleStruct.cnt)
}

sample_test.go

package main

import (
	"log"
	"testing"
	"os/exec"
	"os"
	"fmt"
	"bytes"
	"github.com/stretchr/testify/assert"
)

func TestFailpoint(t *testing.T) {
	log.Printf("[TestFailpoint] Executing TestFailpoint")
	mySampleStruct := SampleStruct{cnt: 2}
	assert.Equal(t, mySampleStruct.cnt, 2)

	// fpsToEnable := map[string]struct{}{
	// 	"doubleIncrement": {},
	// }

	// cmd := exec.Command("export", "GO_FAILPOINTS='main/doubleIncrement=return(true);main/testPanic=return(true)'")
	// if err := cmd.Run(); err != nil {
	// 	log.Fatal(fmt.Sprintf("Encountered error while setting up the GO_FAILPOINTS environment variable. Error: %v", err))
	// } else {
	// 	log.Printf("Successfully set up the GO_FAILPOINTS environment variable")
	// }

	os.Setenv("GO_FAILPOINTS","'main/doubleIncrement=return(true);main/testPanic=return(true)'")
	log.Printf("[TestFailpoint] Successfully set up the GO_FAILPOINTS environment variable")

	cmd := exec.Command("./modv1")
	var out bytes.Buffer
	var stderr bytes.Buffer
	cmd.Stdout = &out
	cmd.Stderr = &stderr
	if err := cmd.Start(); err != nil {
		log.Printf(fmt.Sprintf("[TestFailpoint] Error while starting modv1 executable. Error: %v", err))
		log.Printf("[TestFailpoint] Command error is: " + stderr.String())
	} else {
		log.Printf("[TestFailpoint] Successfully started modv1 executable")
		log.Printf("[TestFailpoint] Result: " + out.String() + ",Error: " + stderr.String())
	}

	if err := cmd.Wait(); err != nil {
		log.Printf(fmt.Sprintf("[TestFailpoint] Error while waiting for modv1 executable. Error: %v", err))
		log.Printf("[TestFailpoint] Command error is: " + stderr.String())
	} else {
		log.Printf("[TestFailpoint] Successfully waited for modv1 executable")
		log.Printf("[TestFailpoint] Result: " + out.String() + ",Error: " + stderr.String())
	}

	//failpoint.Enable("doubleIncrement", "return(1)")
	log.Printf("[TestFailpoint] Calling increment")
	mySampleStruct.increment()
	log.Printf("[TestFailpoint] Called increment")
	log.Printf("[TestFailpoint] Calling incrementFP")
	result := mySampleStruct.incrementFP()
	log.Printf("[TestFailpoint] Called incrementFP")
	assert.Equal(t, mySampleStruct.cnt, 4)
	assert.Equal(t, result, "single")

	os.Unsetenv("GO_FAILPOINTS")
	log.Printf("[TestFailpoint] Successfully unset the GO_FAILPOINTS environment variable")
}

2. What did you expect to see?
   When I run go build and set the environment variable export GO_FAILPOINTS=main/doubleIncrement=return(true);main/testPanic=return(true) and then execute the ./modv1 executable I see that the failpoints in samplestruct.go are hit and the failpoint closure is executed. I expected same behavior when I run go test -v i.e. when the TestFailpoint test is executed.

3. What did you see instead?
   I did not see the failpoints in samplestruct.go getting executed while running the test TestFalipoint. Is this expected behavior? Is the failpoint code executed only in the compiled binary modv1 and not during test execution even though the environment variable GO_FAILPOINTS is kept around (and additionally also set/unset during) test execution?

4. Versions of the failpoint

$ ./failpoint-ctl -V
ReleaseVersion ed9079f
BuildTS 2023-09-01 02:53:58
GitHash ed9079f47f3761cbbeadaded49d48e3264c23427
GitBranch fp-testing
GoVersion go version go1.21.0 darwin/arm64