More robust script handling and code cleanup

In this commit, I have made the script handling dramatically more
robust, and the code has been cleaned up a fair amount. We no longer
have to manually set is restarting or anything like that, and we dont
rely on a fragile exit code system.

main.go

@@ -8,6 +8,7 @@ import (
 	"flag"
 	"fmt"
 	"log"
+	"log/slog"
 	"net/http"
 	"os"
 	"os/exec"
@@ -15,6 +16,7 @@ import (
 	"path"
 	"path/filepath"
 	"runtime"
+	"strconv"
 	"strings"
 	"sync"
 	"syscall"
@@ -48,14 +50,12 @@ type Config struct {
 }
 
 type Script struct {
-	zqdgr        *ZQDGR
+	zqdgr      *ZQDGR
-	command      *exec.Cmd
+	command    *exec.Cmd
-	mutex        sync.Mutex
+	mutex      sync.Mutex
-	scriptName   string
+	scriptName string
-	isRestarting bool
+	// notified with the exit code of the script when it exits
-	wg           sync.WaitGroup
+	exitCode chan int
-	// the exit code of the script, only set after the script has exited
-	exitCode int
 }
 
 type ZQDGR struct {
@@ -132,10 +132,10 @@ func (zqdgr *ZQDGR) NewScript(scriptName string, args ...string) *Script {
 	}
 
 	return &Script{
-		zqdgr:        zqdgr,
+		zqdgr:      zqdgr,
-		command:      command,
+		command:    command,
-		scriptName:   scriptName,
+		scriptName: scriptName,
-		isRestarting: false,
+		exitCode:   make(chan int),
 	}
 }
 
@@ -143,69 +143,66 @@ func (s *Script) Start() error {
 	s.mutex.Lock()
 	defer s.mutex.Unlock()
 
-	s.wg.Add(1)
-
 	err := s.command.Start()
 	if err != nil {
-		s.wg.Done()
 		return err
 	}
 
 	go func() {
-		err := s.command.Wait()
+		processState, err := s.command.Process.Wait()
-		if err != nil {
+		slog.Debug("Script exited", "script", s.scriptName, "error", err)
-			if exitError, ok := err.(*exec.ExitError); ok {
-				s.exitCode = exitError.ExitCode()
-			} else {
-				// Other errors (e.g., process not found, permission denied)
-				if !s.isRestarting {
-					log.Printf("Error waiting for script %s: %v", s.scriptName, err)
-					s.exitCode = 1
-				}
-			}
-		}
 
-		if !s.isRestarting {
+		s.exitCode <- processState.ExitCode()
-			s.wg.Done()
-		}
 	}()
 
 	return err
 }
 
-func (s *Script) Stop(lock bool) error {
+// it is the caller's responsibility to lock the mutex before calling this function
-	if lock {
+func (s *Script) Stop() error {
-		s.mutex.Lock()
+	slog.Debug("Making sure process is still alive")
-		defer s.mutex.Unlock()
+	if runtime.GOOS == "windows" {
-	}
+		if _, err := os.FindProcess(s.command.Process.Pid); err != nil {
-
+			// process is already dead
-	s.isRestarting = true
+			return nil
-
-	if s.command.Process != nil {
-		var signal syscall.Signal
-		switch s.zqdgr.Config.ShutdownSignal {
-		case "SIGINT":
-			signal = syscall.SIGINT
-		case "SIGTERM":
-			signal = syscall.SIGTERM
-		case "SIGQUIT":
-			signal = syscall.SIGQUIT
-		default:
-			signal = syscall.SIGKILL
 		}
-
+	} else {
-		if err := syscall.Kill(-s.command.Process.Pid, signal); err != nil {
+		process, err := os.FindProcess(s.command.Process.Pid)
-			log.Printf("error killing previous process: %v", err)
+		if err != nil {
 			return err
 		}
+		// Sending signal 0 checks for existence and permissions
+		err = process.Signal(syscall.Signal(0))
+		if err != nil {
+			return nil
+		}
 	}
 
+	slog.Debug("Process is still alive, sending signal")
+
 	dead := make(chan bool)
 	go func() {
 		s.command.Wait()
 		dead <- true
 	}()
 
+	var signal syscall.Signal
+	switch s.zqdgr.Config.ShutdownSignal {
+	case "SIGINT":
+		signal = syscall.SIGINT
+	case "SIGTERM":
+		signal = syscall.SIGTERM
+	case "SIGQUIT":
+		signal = syscall.SIGQUIT
+	default:
+		signal = syscall.SIGKILL
+	}
+
+	if err := syscall.Kill(-s.command.Process.Pid, signal); err != nil {
+		log.Printf("error killing previous process: %v", err)
+		return err
+	}
+
 	shutdownTimeout := s.zqdgr.Config.ShutdownTimeout
 	if shutdownTimeout == 0 {
 		shutdownTimeout = 1
@@ -226,9 +223,11 @@ func (s *Script) Stop(lock bool) error {
 }
 
 func (s *Script) Restart() error {
+	slog.Debug("Restarting script", "script", s.scriptName)
+
 	s.mutex.Lock()
 
-	err := s.Stop(false)
+	err := s.Stop()
 	if err != nil {
 		s.mutex.Unlock()
 		return err
@@ -242,8 +241,6 @@ func (s *Script) Restart() error {
 		return nil
 	}
 
-	s.isRestarting = false
-
 	s.mutex.Unlock()
 
 	err = s.Start()
@@ -265,10 +262,6 @@ func (s *Script) Restart() error {
 	return err
 }
 
-func (s *Script) Wait() {
-	s.wg.Wait()
-}
-
 func (wsServer *WSServer) handleWs(w http.ResponseWriter, r *http.Request) {
 	conn, err := wsServer.upgrader.Upgrade(w, r, nil)
 	if err != nil {
@@ -319,18 +312,76 @@ func (zqdgr *ZQDGR) loadConfig() error {
 	return nil
 }
 
+func validatePattern(pattern string) ([]string, error) {
+	var paternArray []string
+	var currentPattern string
+	inMatch := false
+	// iterate over every letter in the pattern
+	for _, p := range pattern {
+		if string(p) == "{" {
+			if inMatch {
+				return nil, fmt.Errorf("unexpected { in pattern")
+			}
+
+			inMatch = true
+		}
+
+		if string(p) == "}" {
+			if !inMatch {
+				return nil, fmt.Errorf("enexpected } in pattern")
+			}
+
+			inMatch = false
+		}
+
+		if string(p) == "," && !inMatch {
+			paternArray = append(paternArray, currentPattern)
+			currentPattern = ""
+			continue
+		}
+
+		currentPattern += string(p)
+	}
+
+	if inMatch {
+		return nil, fmt.Errorf("unmatched } in pattern")
+	}
+
+	if currentPattern != "" {
+		paternArray = append(paternArray, currentPattern)
+	}
+
+	return paternArray, nil
+}
+
 func main() {
+	// var err error
+	var debugMode bool
+
 	noWs := flag.Bool("no-ws", false, "Disable WebSocket server")
 	configDir := flag.String("config", ".", "Path to the config directory")
-	disableReloadConfig := flag.Bool("no-reload-config", false, "Do not restart ZQDGR on config file change")
 	flag.StringVar(configDir, "C", *configDir, "Path to the config directory")
+	disableReloadConfig := flag.Bool("no-reload-config", false, "Do not restart ZQDGR on config file change")
 
 	flag.Parse()
 
+	// if ParseBool returns an error, ZQDGR_DEBUG is not a bool
+	debugMode, _ = strconv.ParseBool(os.Getenv("ZQDGR_DEBUG"))
+	if debugMode {
+		slog.SetLogLoggerLevel(slog.LevelDebug)
+	}
+
 	originalArgs := os.Args
 	os.Args = flag.Args()
 
-	zqdgr := NewZQDGR(*noWs, *configDir)
+	expandedConfigDir, err := filepath.Abs(*configDir)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	slog.Debug("noWS", "noWs", *noWs, "configDir", *configDir, "expandedConfigDir", expandedConfigDir, "disableReloadConfig", *disableReloadConfig)
+
+	zqdgr := NewZQDGR(*noWs, expandedConfigDir)
 	if zqdgr == nil {
 		return
 	}
@@ -338,10 +389,13 @@ func main() {
 	var command string
 	var commandArgs []string
 
+	// command name trimmed by flags.Args()
+	// os.Args ~= [script, --, arguments]
 	for i, arg := range os.Args {
 		if arg == "--" {
-			if i+2 < len(os.Args) {
+			slog.Debug("Found double-dash", "i", i, "len(os.Args)", len(os.Args), "os.Args", os.Args)
-				commandArgs = os.Args[i+2:]
+			if len(os.Args)-i > 1 {
+				commandArgs = os.Args[i+1:]
 			}
 			break
 		}
@@ -354,6 +408,8 @@ func main() {
 		commandArgs = append(commandArgs, arg)
 	}
 
+	slog.Debug("Collected", "command", command, "commandArgs", commandArgs)
+
 	watchMode := false
 	var scriptName string
 	switch command {
@@ -496,13 +552,7 @@ func main() {
 			log.Fatal("please specify a script to run")
 		}
 		watchMode = true
-		for i := range commandArgs {
+		scriptName = commandArgs[0]
-			if strings.HasPrefix(commandArgs[i], "-") {
-				continue
-			}
-
-			scriptName = commandArgs[i]
-		}
 	default:
 		scriptName = command
 	}
@@ -520,19 +570,9 @@ func main() {
 
 		log.Println("Received signal, exiting...")
 		if script.command != nil {
-			var signal syscall.Signal
+			script.mutex.Lock()
-			switch zqdgr.Config.ShutdownSignal {
+			script.Stop()
-			case "SIGINT":
+			script.mutex.Unlock()
-				signal = syscall.SIGINT
-			case "SIGTERM":
-				signal = syscall.SIGTERM
-			case "SIGQUIT":
-				signal = syscall.SIGQUIT
-			default:
-				signal = syscall.SIGKILL
-			}
-
-			syscall.Kill(-script.command.Process.Pid, signal)
 		}
 
 		os.Exit(0)
@@ -556,48 +596,21 @@ func main() {
 		}
 
 		// make sure the pattern is valid
-		var paternArray []string
+		patternArray, err := validatePattern(zqdgr.Config.Pattern)
-		var currentPattern string
+		if err != nil {
-		inMatch := false
+			log.Fatal(err)
-		// iterate over every letter in the pattern
-		for _, p := range zqdgr.Config.Pattern {
-			if string(p) == "{" {
-				if inMatch {
-					log.Fatal("unmatched { in pattern")
-				}
-
-				inMatch = true
-			}
-
-			if string(p) == "}" {
-				if !inMatch {
-					log.Fatal("unmatched } in pattern")
-				}
-
-				inMatch = false
-			}
-
-			if string(p) == "," && !inMatch {
-				paternArray = append(paternArray, currentPattern)
-				currentPattern = ""
-				inMatch = false
-				continue
-			}
-
-			currentPattern += string(p)
 		}
 
-		if inMatch {
+		for _, pattern := range zqdgr.Config.ExcludedGlobs {
-			log.Fatal("unmatched } in pattern")
+			_, err := validatePattern(pattern)
-		}
+			if err != nil {
-
+				log.Fatal(err)
-		if currentPattern != "" {
+			}
-			paternArray = append(paternArray, currentPattern)
 		}
 
 		watcherConfig := WatcherConfig{
 			excludedGlobs: globList(zqdgr.Config.ExcludedGlobs),
-			pattern:       paternArray,
+			pattern:       patternArray,
 		}
 
 		watcher, err := NewWatcher(&watcherConfig)
@@ -617,13 +630,12 @@ func main() {
 			log.Fatal(err)
 		}
 
-		// We use this timer to deduplicate events.
+		// tailing edge debounce of file system events
 		var (
-			// Wait 100ms for new events; each new event resets the timer.
 			waitFor = 100 * time.Millisecond
 
-			// Keep track of the timers, as path → timer.
+			mu sync.Mutex
-			mu     sync.Mutex
+			// watched filepath -> timer
 			timers = make(map[string]*time.Timer)
 		)
 		go func() {
@@ -640,25 +652,37 @@ func main() {
 
 					if !ok {
 						timer = time.AfterFunc(waitFor, func() {
-							if event.Op&fsnotify.Remove == fsnotify.Remove || event.Op&fsnotify.Write == fsnotify.Write || event.Op&fsnotify.Create == fsnotify.Create {
+							slog.Debug("FSnotify event received", "event", event)
-								if os.Getenv("ZQDGR_DEBUG") != "" {
-									fmt.Println("File changed:", event.Name)
-								}
 
-								if strings.HasSuffix(event.Name, "zqdgr.config.json") {
+							if event.Op&fsnotify.Remove == fsnotify.Remove || event.Op&fsnotify.Write == fsnotify.Write || event.Op&fsnotify.Create == fsnotify.Create {
+								slog.Debug("File changed", "file", event.Name)
+
+								fullEventPath, err := filepath.Abs(event.Name)
+								if err != nil {
+									log.Fatal(err)
+								}
+								slog.Debug("expanded event path", "path", fullEventPath)
+
+								// check against the fullpath to make sure that the config file that was changed is
+								// actually the one we are using
+								if fullEventPath == filepath.Join(zqdgr.WorkingDirectory, "zqdgr.config.json") {
 									// re-exec the exact same command
 									if !*disableReloadConfig {
-										log.Println("zqdgr.config.json has changed, restarting...")
+										fmt.Println("zqdgr.config.json has changed, restarting...")
 										executable, err := os.Executable()
 										if err != nil {
 											log.Fatal(err)
 										}
 
-										err = script.Stop(true)
+										script.mutex.Lock()
+
+										err = script.Stop()
 										if err != nil {
 											log.Fatal(err)
 										}
 
+										script.mutex.Unlock()
+
 										err = syscall.Exec(executable, originalArgs, os.Environ())
 										if err != nil {
 											log.Fatal(err)
@@ -668,8 +692,16 @@ func main() {
 									}
 								}
 
-								if directoryShouldBeTracked(&watcherConfig, event.Name) {
+								if pathShouldBeTracked(&watcherConfig, event.Name) {
-									watcher.(NotifyWatcher).watcher.Add(event.Name)
+									if event.Op&fsnotify.Create == fsnotify.Create {
+										slog.Debug("Adding new file to watcher", "file", event.Name)
+										watcher.(NotifyWatcher).watcher.Add(event.Name)
+									}
+
+									if event.Op&fsnotify.Remove == fsnotify.Remove {
+										slog.Debug("Removing file from watcher", "file", event.Name)
+										watcher.(NotifyWatcher).watcher.Remove(event.Name)
+									}
 								}
 
 								if pathMatches(&watcherConfig, event.Name) {
@@ -702,8 +734,17 @@ func main() {
 				}
 			}
 		}()
+
+		// block until the script exits with a zero (aka, it normally exited on its own)
+		for {
+			exitCode := <-script.exitCode
+			slog.Debug("Script exited", "script", scriptName, "exitCode", exitCode)
+			if exitCode == 0 {
+				os.Exit(0)
+			}
+		}
 	}
 
-	script.Wait()
+	// block until the script exits
-	os.Exit(script.exitCode)
+	os.Exit(<-script.exitCode)
 }

watcher.go

@@ -4,7 +4,7 @@ import (
 	"errors"
 	"fmt"
 	"log"
-	"os"
+	"log/slog"
 	"path/filepath"
 	"strings"
 
@@ -34,7 +34,9 @@ func (g *globList) Matches(value string) bool {
 
 func matchesPattern(pattern []string, path string) bool {
 	for _, p := range pattern {
+		slog.Debug("checking path against pattern", "pattern", p, "path", path)
 		if matched, _ := doublestar.Match(p, path); matched {
+			slog.Debug("path matches pattern", "pattern", p, "path", path)
 			return true
 		}
 	}
@@ -42,17 +44,13 @@ func matchesPattern(pattern []string, path string) bool {
 	return false
 }
 
-func directoryShouldBeTracked(cfg *WatcherConfig, path string) bool {
+func pathShouldBeTracked(cfg *WatcherConfig, path string) bool {
 	base := filepath.Dir(path)
 
-	if os.Getenv("ZQDGR_DEBUG") != "" {
+	slog.Debug("checking file against path", "path", path, "base", base)
-		log.Printf("checking %s against %s %v\n", path, base, *cfg)
-	}
 
 	if cfg.excludedGlobs.Matches(base) {
-		if os.Getenv("ZQDGR_DEBUG") != "" {
+		slog.Debug("file is excluded", "path", path)
-			log.Printf("%s is excluded\n", base)
-		}
 		return false
 	}
 
@@ -86,9 +84,7 @@ func (n NotifyWatcher) Close() error {
 }
 
 func (n NotifyWatcher) AddFile(path string) error {
-	if os.Getenv("ZQDGR_DEBUG") != "" {
+	slog.Debug("manually adding file", "file", path)
-		log.Printf("manually adding file\n")
-	}
 
 	return n.add(path)
 }
@@ -125,24 +121,36 @@ func NewWatcher(cfg *WatcherConfig) (FileWatcher, error) {
 func addFiles(fw FileWatcher) error {
 	cfg := fw.getConfig()
 	for _, pattern := range cfg.pattern {
-		if os.Getenv("ZQDGR_DEBUG") != "" {
+		slog.Debug("adding pattern", "pattern", pattern)
-			fmt.Printf("processing glob %s\n", pattern)
-		}
 
 		matches, err := doublestar.Glob(pattern)
 		if err != nil {
 			log.Fatalf("Bad pattern \"%s\": %s", pattern, err.Error())
 		}
 
+		trackedDirs := make(map[string]bool)
 		for _, match := range matches {
-			if os.Getenv("ZQDGR_DEBUG") != "" {
+			base := filepath.Dir(match)
-				log.Printf("checking %s\n", match)
+			// this allows us to track file creations and deletions
+			if !trackedDirs[base] {
+				if cfg.excludedGlobs.Matches(base) {
+					slog.Debug("directory is excluded", "file", match)
+					continue
+				}
+
+				trackedDirs[base] = true
+
+				slog.Debug("adding directory", "dir", base)
+
+				if err := fw.add(base); err != nil {
+					return fmt.Errorf("FileWatcher.Add(): %v", err)
+				}
 			}
 
-			if directoryShouldBeTracked(cfg, match) {
+			slog.Debug("adding file", "file", match)
-				if os.Getenv("ZQDGR_DEBUG") != "" {
+
-					log.Printf("%s is not excluded\n", match)
+			if pathShouldBeTracked(cfg, match) {
-				}
+				slog.Debug("path should be tracked", "file", match)
 
 				if err := fw.add(match); err != nil {
 					return fmt.Errorf("FileWatcher.Add(): %v", err)

zqdgr.config.json

@@ -1,6 +1,6 @@
 {
     "name": "zqdgr",
-    "version": "0.0.4",
+    "version": "0.0.6",
     "description": "zqdgr is a quick and dirty Golang runner",
     "author": "juls0730",
     "license": "BSL-1.0",