rclone/fstest/test_all/run.go
2019-09-24 12:45:12 +01:00

394 lines
9.5 KiB
Go

// Run a test
// +build go1.11
package main
import (
"bytes"
"fmt"
"go/build"
"io"
"log"
"os"
"os/exec"
"path"
"regexp"
"runtime"
"sort"
"strconv"
"strings"
"sync"
"time"
"github.com/rclone/rclone/fs"
)
// Control concurrency per backend if required
var (
oneOnlyMu sync.Mutex
oneOnly = map[string]*sync.Mutex{}
)
// Run holds info about a running test
//
// A run just runs one command line, but it can be run multiple times
// if retries are needed.
type Run struct {
// Config
Remote string // name of the test remote
Backend string // name of the backend
Path string // path to the source directory
FastList bool // add -fast-list to tests
Short bool // add -short
NoRetries bool // don't retry if set
OneOnly bool // only run test for this backend at once
NoBinary bool // set to not build a binary
SizeLimit int64 // maximum test file size
Ignore map[string]struct{}
// Internals
cmdLine []string
cmdString string
try int
err error
output []byte
failedTests []string
runFlag string
logDir string // directory to place the logs
trialName string // name/log file name of current trial
trialNames []string // list of all the trials
}
// Runs records multiple Run objects
type Runs []*Run
// Sort interface
func (rs Runs) Len() int { return len(rs) }
func (rs Runs) Swap(i, j int) { rs[i], rs[j] = rs[j], rs[i] }
func (rs Runs) Less(i, j int) bool {
a, b := rs[i], rs[j]
if a.Backend < b.Backend {
return true
} else if a.Backend > b.Backend {
return false
}
if a.Remote < b.Remote {
return true
} else if a.Remote > b.Remote {
return false
}
if a.Path < b.Path {
return true
} else if a.Path > b.Path {
return false
}
if !a.FastList && b.FastList {
return true
} else if a.FastList && !b.FastList {
return false
}
return false
}
// dumpOutput prints the error output
func (r *Run) dumpOutput() {
log.Println("------------------------------------------------------------")
log.Printf("---- %q ----", r.cmdString)
log.Println(string(r.output))
log.Println("------------------------------------------------------------")
}
// This converts a slice of test names into a regexp which matches
// them.
func testsToRegexp(tests []string) string {
var split []map[string]struct{}
// Make a slice with maps of the used parts at each level
for _, test := range tests {
for i, name := range strings.Split(test, "/") {
if i >= len(split) {
split = append(split, make(map[string]struct{}))
}
split[i][name] = struct{}{}
}
}
var out []string
for _, level := range split {
var testsInLevel = []string{}
for name := range level {
testsInLevel = append(testsInLevel, name)
}
sort.Strings(testsInLevel)
if len(testsInLevel) > 1 {
out = append(out, "^("+strings.Join(testsInLevel, "|")+")$")
} else {
out = append(out, "^"+testsInLevel[0]+"$")
}
}
return strings.Join(out, "/")
}
var failRe = regexp.MustCompile(`(?m)^\s*--- FAIL: (Test.*?) \(`)
// findFailures looks for all the tests which failed
func (r *Run) findFailures() {
oldFailedTests := r.failedTests
r.failedTests = nil
excludeParents := map[string]struct{}{}
ignored := 0
for _, matches := range failRe.FindAllSubmatch(r.output, -1) {
failedTest := string(matches[1])
// Skip any ignored failures
if _, found := r.Ignore[failedTest]; found {
ignored++
} else {
r.failedTests = append(r.failedTests, failedTest)
}
// Find all the parents of this test
parts := strings.Split(failedTest, "/")
for i := len(parts) - 1; i >= 1; i-- {
excludeParents[strings.Join(parts[:i], "/")] = struct{}{}
}
}
// Exclude the parents
var newTests = r.failedTests[:0]
for _, failedTest := range r.failedTests {
if _, excluded := excludeParents[failedTest]; !excluded {
newTests = append(newTests, failedTest)
}
}
r.failedTests = newTests
if len(r.failedTests) == 0 && ignored > 0 {
log.Printf("%q - Found %d ignored errors only - marking as good", r.cmdString, ignored)
r.err = nil
r.dumpOutput()
return
}
if len(r.failedTests) != 0 {
r.runFlag = testsToRegexp(r.failedTests)
} else {
r.runFlag = ""
}
if r.passed() && len(r.failedTests) != 0 {
log.Printf("%q - Expecting no errors but got: %v", r.cmdString, r.failedTests)
r.dumpOutput()
} else if !r.passed() && len(r.failedTests) == 0 {
log.Printf("%q - Expecting errors but got none: %v", r.cmdString, r.failedTests)
r.dumpOutput()
r.failedTests = oldFailedTests
}
}
// nextCmdLine returns the next command line
func (r *Run) nextCmdLine() []string {
cmdLine := r.cmdLine
if r.runFlag != "" {
cmdLine = append(cmdLine, "-test.run", r.runFlag)
}
return cmdLine
}
// trial runs a single test
func (r *Run) trial() {
cmdLine := r.nextCmdLine()
cmdString := toShell(cmdLine)
msg := fmt.Sprintf("%q - Starting (try %d/%d)", cmdString, r.try, *maxTries)
log.Println(msg)
logName := path.Join(r.logDir, r.trialName)
out, err := os.Create(logName)
if err != nil {
log.Fatalf("Couldn't create log file: %v", err)
}
defer func() {
err := out.Close()
if err != nil {
log.Fatalf("Failed to close log file: %v", err)
}
}()
_, _ = fmt.Fprintln(out, msg)
// Early exit if --try-run
if *dryRun {
log.Printf("Not executing as --dry-run: %v", cmdLine)
_, _ = fmt.Fprintln(out, "--dry-run is set - not running")
return
}
// Internal buffer
var b bytes.Buffer
multiOut := io.MultiWriter(out, &b)
cmd := exec.Command(cmdLine[0], cmdLine[1:]...)
cmd.Stderr = multiOut
cmd.Stdout = multiOut
cmd.Dir = r.Path
start := time.Now()
r.err = cmd.Run()
r.output = b.Bytes()
duration := time.Since(start)
r.findFailures()
if r.passed() {
msg = fmt.Sprintf("%q - Finished OK in %v (try %d/%d)", cmdString, duration, r.try, *maxTries)
} else {
msg = fmt.Sprintf("%q - Finished ERROR in %v (try %d/%d): %v: Failed %v", cmdString, duration, r.try, *maxTries, r.err, r.failedTests)
}
log.Println(msg)
_, _ = fmt.Fprintln(out, msg)
}
// passed returns true if the test passed
func (r *Run) passed() bool {
return r.err == nil
}
// GOPATH returns the current GOPATH
func GOPATH() string {
gopath := os.Getenv("GOPATH")
if gopath == "" {
gopath = build.Default.GOPATH
}
return gopath
}
// BinaryName turns a package name into a binary name
func (r *Run) BinaryName() string {
binary := path.Base(r.Path) + ".test"
if runtime.GOOS == "windows" {
binary += ".exe"
}
return binary
}
// BinaryPath turns a package name into a binary path
func (r *Run) BinaryPath() string {
return path.Join(r.Path, r.BinaryName())
}
// PackagePath returns the path to the package
func (r *Run) PackagePath() string {
return path.Join(GOPATH(), "src", r.Path)
}
// MakeTestBinary makes the binary we will run
func (r *Run) MakeTestBinary() {
binary := r.BinaryPath()
binaryName := r.BinaryName()
log.Printf("%s: Making test binary %q", r.Path, binaryName)
cmdLine := []string{"go", "test", "-c"}
if *dryRun {
log.Printf("Not executing: %v", cmdLine)
return
}
cmd := exec.Command(cmdLine[0], cmdLine[1:]...)
cmd.Dir = r.Path
err := cmd.Run()
if err != nil {
log.Fatalf("Failed to make test binary: %v", err)
}
if _, err := os.Stat(binary); err != nil {
log.Fatalf("Couldn't find test binary %q", binary)
}
}
// RemoveTestBinary removes the binary made in makeTestBinary
func (r *Run) RemoveTestBinary() {
if *dryRun {
return
}
binary := r.BinaryPath()
err := os.Remove(binary) // Delete the binary when finished
if err != nil {
log.Printf("Error removing test binary %q: %v", binary, err)
}
}
// Name returns the run name as a file name friendly string
func (r *Run) Name() string {
ns := []string{
r.Backend,
strings.Replace(r.Path, "/", ".", -1),
r.Remote,
}
if r.FastList {
ns = append(ns, "fastlist")
}
ns = append(ns, fmt.Sprintf("%d", r.try))
s := strings.Join(ns, "-")
s = strings.Replace(s, ":", "", -1)
return s
}
// Init the Run
func (r *Run) Init() {
prefix := "-test."
if r.NoBinary {
prefix = "-"
r.cmdLine = []string{"go", "test"}
} else {
r.cmdLine = []string{"./" + r.BinaryName()}
}
r.cmdLine = append(r.cmdLine, prefix+"v", prefix+"timeout", timeout.String(), "-remote", r.Remote)
r.try = 1
if *verbose {
r.cmdLine = append(r.cmdLine, "-verbose")
fs.Config.LogLevel = fs.LogLevelDebug
}
if *runOnly != "" {
r.cmdLine = append(r.cmdLine, prefix+"run", *runOnly)
}
if r.FastList {
r.cmdLine = append(r.cmdLine, "-fast-list")
}
if r.Short {
r.cmdLine = append(r.cmdLine, "-short")
}
if r.SizeLimit > 0 {
r.cmdLine = append(r.cmdLine, "-size-limit", strconv.FormatInt(r.SizeLimit, 10))
}
r.cmdString = toShell(r.cmdLine)
}
// Logs returns all the log names
func (r *Run) Logs() []string {
return r.trialNames
}
// FailedTests returns the failed tests as a comma separated string, limiting the number
func (r *Run) FailedTests() string {
const maxTests = 5
ts := r.failedTests
if len(ts) > maxTests {
ts = ts[:maxTests:maxTests]
ts = append(ts, fmt.Sprintf("… (%d more)", len(r.failedTests)-maxTests))
}
return strings.Join(ts, ", ")
}
// Run runs all the trials for this test
func (r *Run) Run(logDir string, result chan<- *Run) {
if r.OneOnly {
oneOnlyMu.Lock()
mu := oneOnly[r.Backend]
if mu == nil {
mu = new(sync.Mutex)
oneOnly[r.Backend] = mu
}
oneOnlyMu.Unlock()
mu.Lock()
defer mu.Unlock()
}
r.Init()
r.logDir = logDir
for r.try = 1; r.try <= *maxTries; r.try++ {
r.trialName = r.Name() + ".txt"
r.trialNames = append(r.trialNames, r.trialName)
log.Printf("Starting run with log %q", r.trialName)
r.trial()
if r.passed() || r.NoRetries {
break
}
}
if !r.passed() {
r.dumpOutput()
}
result <- r
}