add control structures: if, for, while, until

- new parser (parser.go) for lists, pipelines, ! and if/elif/else/fi,
  for ... in ... do/done, while/until ... do/done; simple commands are
  kept as text and expanded when they run (in every iteration)
- break [n], continue [n] and the : builtin
- incomplete input (open control structure, trailing &&, |, quote or
  backslash) continues on the next line: prompt "> " in interactive
  shells, scripts and startup files read on
- Ctrl+C aborts the whole command line, including loops
- aliases are expanded at the start of every simple command, also inside
  control structures
- control structures in pipelines, with redirections or in the
  background are reported as not supported
- completion: commands after if, then, do etc.
- main_test.go reformatted with gofmt

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

README.md

@@ -19,6 +19,9 @@ Built-in commands:
   wait [%n]     - Wait for job n or all background jobs
   alias [name[=value] ...] - Define or show aliases (alias ll='ls -l')
   unalias name  - Remove an alias (unalias -a: all aliases)
+  :             - Do nothing (exit status 0)
+  break [n]     - Leave the innermost (or n-th) for/while/until loop
+  continue [n]  - Continue with the next iteration of a loop
   help          - Show this help
 
 Features:
@@ -34,14 +37,23 @@ Features:
   - Wildcards: * ? [abc] [a-z] [!abc] match file names (not hidden files
     unless the pattern starts with a dot); quoted wildcards and values of
     variables are not expanded, a pattern without match is kept as is
-  - Colored prompt (user@host green, red for root; directory blue),
-    configurable with $GOSH_COLORS, e.g. GOSH_COLORS='user=1;33:path=36'
-    (keys user, root, path; values are ANSI SGR codes, empty = no color);
-    disabled with $NO_COLOR, TERM=dumb or when output is not a terminal
+  - Prompt user@host:dir $, with [N] before the $ if the last command
+    failed with exit status N
+  - Colored prompt (user@host green, red for root; directory blue; exit
+    status red), configurable with $GOSH_COLORS, e.g.
+    GOSH_COLORS='user=1;33:path=36' (keys user, root, path, error; values
+    are ANSI SGR codes, empty = no color); disabled with $NO_COLOR,
+    TERM=dumb or when output is not a terminal
   - Environment variable expansion ($VAR, ${VAR}, ${VAR:-default});
     using an unset variable is an error, use ${VAR:-} for an empty default
+  - Command substitution: $(command) is replaced by the output of command
+    (run in a subshell); unquoted it is split into words, "$(command)"
+    stays one word
   - Command lists: a ; b (run both), a && b (b only if a succeeded),
     a || b (b only if a failed); $? is the exit status of the last command
+  - Control structures: if cmd; then ...; elif cmd; then ...; else ...; fi,
+    for x in words; do ...; done, while cmd; do ...; done, until ...,
+    ! cmd negates the status; they can span several lines (prompt "> ")
   - Pipes (a | b) and redirections: < file, > file, >> file, 2> file,
     2>> file, 2>&1, >&2, &> file (stdout and stderr), &>> file
   - Job control: command & runs in the background, Ctrl+Z stops the
@@ -119,6 +131,72 @@ export LANG=de_DE.UTF-8
 export PATH=$HOME/bin:$PATH
 ```
 
+### Kontrollstrukturen
+
+gosh kennt `if`, `for`, `while` und `until` wie `sh`, auf einer Zeile oder über mehrere Zeilen:
+
+```bash
+if [ -f ~/.goshrc ]; then
+  echo "rc vorhanden"
+elif [ -d ~/.config ]; then
+  echo "nur .config"
+else
+  echo "nichts"
+fi
+
+for f in *.log; do
+  gzip "$f"
+done
+
+for i in {1..5}; do echo "Durchlauf $i"; done
+
+set n=3
+while [ $n -gt 0 ]; do
+  echo $n
+  set n=$(expr $n - 1)
+done
+
+until ping -c1 -W1 server >/dev/null; do sleep 1; done
+```
+
+| Konstrukt                                   | Bedeutung                                                         |
+|---------------------------------------------|-------------------------------------------------------------------|
+| `if befehl; then …; elif …; then …; else …; fi` | führt den Zweig aus, dessen Bedingung den Exit-Status 0 hat   |
+| `for x in wörter; do …; done`               | setzt `x` nacheinander auf jedes Wort                             |
+| `while befehl; do …; done`                  | wiederholt, solange `befehl` erfolgreich ist                      |
+| `until befehl; do …; done`                  | wiederholt, bis `befehl` erfolgreich ist                          |
+| `! befehl`                                  | kehrt den Exit-Status um                                          |
+| `break [n]`, `continue [n]`                 | verlässt die (n-te) Schleife bzw. springt zum nächsten Durchlauf  |
+| `:`                                         | tut nichts, Exit-Status 0 (z. B. `while :; do …; done`)           |
+
+- Bedingungen sind normale Befehle (auch Listen mit `&&`/`||`), meist `test` bzw. `[ … ]` (z. B. `[ -f datei ]`, `[ "$a" = b ]`, `[ $n -lt 3 ]`).
+- Die Wörter nach `for … in` werden wie Argumente expandiert: `*.go`, `{1..10}`, `$VAR`, `$(befehl)`. Ohne `in` läuft die Schleife nicht (Positionsparameter gibt es nicht).
+- Variablen, `$(...)` und Wildcards im Rumpf werden bei jedem Durchlauf neu ersetzt.
+- Ist ein Befehl noch nicht vollständig (z. B. nach `if …; then` oder `for …` + Enter, nach `&&`, `|` oder einem `\` am Zeilenende), zeigt gosh den Fortsetzungs-Prompt `> `; Ctrl+C verwirft die angefangene Eingabe. Skripte und Startdateien lesen entsprechend weiter.
+- Ctrl+C bricht die ganze Befehlszeile ab, also auch eine Schleife (`while true; do sleep 1; done`).
+- Der Exit-Status ist der des zuletzt ausgeführten Befehls; ein `if` ohne passenden Zweig und eine Schleife ohne Durchlauf haben den Status 0.
+- Nicht unterstützt: Kontrollstrukturen in Pipes (`for …; done | sort`), mit Umleitung (`done > datei`) oder im Hintergrund (`done &`), außerdem `case`, Funktionen, `read` und Zuweisungen der Form `x=1` – Variablen setzt man mit `set x=1`, Arithmetik geht z. B. mit `$(expr $n + 1)`.
+
+### Befehlsersetzung
+
+`$(befehl)` wird durch die Ausgabe von `befehl` ersetzt:
+
+```
+$ echo "Heute ist $(date +%A)"
+Heute ist Montag
+$ kill $(pgrep -f mein-server)
+$ cd $(git rev-parse --show-toplevel)
+$ vim $(grep -l TODO *.go)
+$ echo "Dateien: $(ls | wc -l)"
+```
+
+- Der Befehl läuft in einer eigenen Subshell (einem neuen gosh-Prozess): `cd`, `set` oder `export` darin ändern die laufende Shell nicht. Er darf alles enthalten, was gosh kann (Pipes, `;`, `&&`, Umleitungen, verschachtelte `$(...)`), und liest von der gleichen Eingabe, Fehlermeldungen gehen auf stderr.
+- Abschließende Zeilenumbrüche der Ausgabe werden entfernt.
+- Ohne Anführungszeichen wird die Ausgabe an Leerzeichen, Tabs und Zeilenumbrüchen in mehrere Wörter geteilt, `"$(befehl)"` bleibt ein einziges Wort (mit den Zeilenumbrüchen darin).
+- Anders als in `sh` werden Wildcards in der Ausgabe nicht expandiert, und Zeichen wie `;` oder `|` darin sind normaler Text: Die Ausgabe ist immer nur Daten.
+- Besteht ein Befehl nur aus einer Ersetzung (`$(false)`), ist ihr Exit-Status `$?`, sonst zählt der Status des Befehls. Ctrl+C während der Ersetzung bricht die ganze Zeile ab.
+- In `'...'` oder mit `\$(` wird nichts ersetzt. Die alte Schreibweise mit Backticks (`` `befehl` ``) und Arithmetik `$((1+2))` gibt es nicht.
+
 ### Befehlslisten und Exit-Status
 
 Mehrere Befehle in einer Zeile werden wie in `sh` verknüpft:
@@ -289,6 +367,22 @@ $ bg %2                       # oder: angehaltenen Job im Hintergrund weiterlauf
 - Wird ein angehaltener Job von außen fortgesetzt (`kill -CONT`), zeigt `jobs` ihn weiter als `Stopped`, bis er endet oder erneut angehalten wird; `fg` und `bg` funktionieren trotzdem.
 - Einen Job beenden lässt sich mit `kill` über die Prozess-ID (z. B. `kill 12345`); `kill %1` versteht gosh noch nicht.
 
+### Prompt
+
+Der Prompt zeigt Benutzer, Rechner und Verzeichnis. Ist der letzte Befehl fehlgeschlagen, steht sein Exit-Status in eckigen Klammern davor, bis wieder ein Befehl erfolgreich war:
+
+```
+flow@host:~/src $ false
+flow@host:~/src [1] $ gibtsnicht
+gosh: gibtsnicht: command not found
+flow@host:~/src [127] $ sleep 100     # mit Ctrl+C abgebrochen
+^C
+flow@host:~/src [130] $ true
+flow@host:~/src $
+```
+
+Typische Werte: `1` allgemeiner Fehler, `2` Syntaxfehler, `126` nicht ausführbar, `127` Befehl nicht gefunden, `128+n` durch Signal `n` beendet (`130` = Ctrl+C, `148` = mit Ctrl+Z angehalten unter Linux). Eine leere Eingabe ändert den Status nicht.
+
 ### Farben
 
 Der Prompt wird über die Umgebungsvariable `GOSH_COLORS` eingefärbt. Sie besteht aus durch `:` getrennten Einträgen `schlüssel=code`:
@@ -298,6 +392,7 @@ Der Prompt wird über die Umgebungsvariable `GOSH_COLORS` eingefärbt. Sie beste
 | `user`    | `user@host`                    | `1;32` (fett grün)  |
 | `root`    | `user@host`, wenn man root ist | `1;31` (fett rot)   |
 | `path`    | aktuelles Verzeichnis          | `1;34` (fett blau)  |
+| `error`   | Exit-Status `[N]` nach einem Fehler | `1;31` (fett rot) |
 
 Die Werte sind [ANSI-SGR-Codes](https://en.wikipedia.org/wiki/ANSI_escape_code#SGR_(Select_Graphic_Rendition)_parameters), mehrere werden mit `;` kombiniert:
 
@@ -335,6 +430,9 @@ GOSH_COLORS='user=1;38;2;136;192;208:path=38;2;216;222;233'
 # Ozean: türkis und kursiv blau
 GOSH_COLORS='user=1;96:path=3;94'
 
+# Fehler-Status gelb auf rotem Hintergrund
+GOSH_COLORS='error=1;33;41'
+
 # Root-Warnung: weiß auf rotem Hintergrund
 GOSH_COLORS='root=1;97;41'
 

alias.go

@@ -75,7 +75,7 @@ func expandAliasesIn(line string, active map[string]bool) string {
 	for i := 0; i < len(line); i++ {
 		ch := line[i]
 		if quote == 0 && !escaped && commandPos {
-			if ch == ' ' || ch == '\t' {
+			if ch == ' ' || ch == '\t' || ch == '\n' {
 				out.WriteByte(ch)
 				continue
 			}
@@ -96,6 +96,16 @@ func expandAliasesIn(line string, active map[string]bool) string {
 			}
 		}
 
+		if quote != '\'' && !escaped && isSubstStart(line, i) {
+			// $(...) is copied as it is; the subshell does not expand
+			// aliases
+			if end := substEnd(line, i); end >= 0 {
+				out.WriteString(line[i : end+1])
+				i = end
+				continue
+			}
+		}
+
 		out.WriteByte(ch)
 		switch {
 		case escaped:

brace.go

@@ -15,7 +15,8 @@ import (
 const maxBraceWords = 100000
 
 // expandBraceWords applies brace expansion to every word of a command (the
-// text of a pipeline). Quoted or escaped braces and ${...} are left alone.
+// text of a pipeline). Quoted or escaped braces, ${...} and $(...) are left
+// alone.
 // A redirection target must expand to a single word.
 func expandBraceWords(input string) (string, error) {
 	var (
@@ -69,6 +70,10 @@ func wordEnd(s string, i int) int {
 			}
 		case ch == '\\':
 			i++
+		case isSubstStart(s, i):
+			if j := substEnd(s, i); j >= 0 {
+				i = j
+			}
 		case ch == '$' && i+1 < len(s) && s[i+1] == '{':
 			if j := strings.IndexByte(s[i:], '}'); j >= 0 {
 				i += j
@@ -103,6 +108,12 @@ func braceScan(s string, fn func(i, depth int) bool) {
 		case ch == '\\':
 			i++
 			continue
+		case isSubstStart(s, i):
+			// The subshell does its own brace expansion
+			if j := substEnd(s, i); j >= 0 {
+				i = j
+			}
+			continue
 		case ch == '$' && i+1 < len(s) && s[i+1] == '{':
 			if j := strings.IndexByte(s[i:], '}'); j >= 0 {
 				i += j

main.go

@@ -131,6 +131,8 @@ type token struct {
 //   - '...' is taken literally, nothing is expanded inside
 //   - "..." expands $VAR and ${VAR}; \ only escapes $ " \ and `
 //   - outside quotes \ escapes any character
+//   - $(command) is replaced by the output of command (see commandSubst);
+//     unquoted, the output is split into words at blanks and newlines
 //   - $VAR, ${VAR} and ${VAR:-default} are expanded outside single quotes;
 //     an unset variable is an error, so that e.g. "rm -rf $DIR/x" never runs
 //     with an empty $DIR
@@ -190,6 +192,32 @@ func tokenize(input string) ([]token, error) {
 		case ch == '\\':
 			startRaw(i)
 			escaped = true
+		case isSubstStart(input, i):
+			startRaw(i)
+			end := substEnd(input, i)
+			if end < 0 {
+				return nil, errors.New("missing ) in command substitution")
+			}
+			out, err := commandSubst(input[i+2 : end])
+			if err != nil {
+				return nil, err
+			}
+			if quote == '"' {
+				literal(out)
+				inWord = true
+			} else {
+				// Unquoted, the output is split into words at blanks and
+				// newlines (but, unlike in sh, wildcards in it are not
+				// expanded)
+				for k, field := range strings.Fields(out) {
+					if k > 0 {
+						endWord()
+					}
+					literal(field)
+					inWord = true
+				}
+			}
+			i = end
 		case ch == '$':
 			startRaw(i)
 			value, n, err := expandVar(input[i:])
@@ -612,21 +640,23 @@ func getPrompt() string {
 		c := parseColors(os.Getenv("GOSH_COLORS"))
 		colors = &c
 	}
-	return buildPrompt(username, hostname, wd, os.Geteuid() == 0, colors)
+	return buildPrompt(username, hostname, wd, os.Geteuid() == 0, lastStatus, colors)
 }
 
 // promptColors holds the SGR parameters (e.g. "1;32") for the parts of the
 // prompt. An empty value means no color for that part.
 type promptColors struct {
-	user string // user@host
-	root string // user@host when running as root
-	path string // working directory
+	user  string // user@host
+	root  string // user@host when running as root
+	path  string // working directory
+	error string // exit status of the last command, if it failed
 }
 
-// defaultColors: bold green user, bold red root, bold blue directory
-var defaultColors = promptColors{user: "1;32", root: "1;31", path: "1;34"}
+// defaultColors: bold green user, bold red root, bold blue directory, bold
+// red exit status
+var defaultColors = promptColors{user: "1;32", root: "1;31", path: "1;34", error: "1;31"}
 
-// parseColors reads a GOSH_COLORS value like "user=1;33:root=1;31:path=36"
+// parseColors reads a GOSH_COLORS value like "user=1;33:path=36:error=31"
 // and overrides the corresponding defaults. Unknown keys and values that are
 // not plain SGR parameters (digits and ';') are ignored, so the variable
 // cannot be used to write arbitrary escape sequences to the terminal.
@@ -644,6 +674,8 @@ func parseColors(spec string) promptColors {
 			colors.root = value
 		case "path":
 			colors.path = value
+		case "error":
+			colors.error = value
 		}
 	}
 	return colors
@@ -669,20 +701,28 @@ func colorize(text, sgr string) string {
 	return "\x1b[" + sgr + "m" + text + "\x1b[0m"
 }
 
-// buildPrompt formats the prompt as user@host:path $. The parts are sanitized
+// buildPrompt formats the prompt as user@host:path $, or user@host:path [N] $
+// if the last command failed with exit status N. The parts are sanitized
 // before the color codes are added, so only our own escape sequences end up
 // in the prompt. colors == nil disables colors.
-func buildPrompt(username, hostname, path string, root bool, colors *promptColors) string {
+func buildPrompt(username, hostname, path string, root bool, status int, colors *promptColors) string {
 	userHost := sanitize(username + "@" + hostname)
 	path = sanitize(path)
+	var failed string
+	if status != 0 {
+		failed = " [" + strconv.Itoa(status) + "]"
+	}
 	if colors == nil {
-		return userHost + ":" + path + " $ "
+		return userHost + ":" + path + failed + " $ "
 	}
 	userColor := colors.user
 	if root {
 		userColor = colors.root
 	}
-	return colorize(userHost, userColor) + ":" + colorize(path, colors.path) + " $ "
+	if failed != "" {
+		failed = " " + colorize(failed[1:], colors.error)
+	}
+	return colorize(userHost, userColor) + ":" + colorize(path, colors.path) + failed + " $ "
 }
 
 // useColor reports whether the prompt should be colored: stdout must be a
@@ -702,7 +742,7 @@ type shellCompleter struct {
 
 func newShellCompleter() *shellCompleter {
 	return &shellCompleter{
-		commands: []string{"cd", "pwd", "echo", "exit", "quit", "set", "export", "source", "jobs", "fg", "bg", "wait", "alias", "unalias", "help"},
+		commands: []string{"cd", "pwd", "echo", "exit", "quit", "set", "export", "source", "jobs", "fg", "bg", "wait", "alias", "unalias", "break", "continue", "help"},
 	}
 }
 
@@ -716,6 +756,13 @@ func (c *shellCompleter) Do(line []rune, pos int) ([][]rune, int) {
 	before := strings.TrimSpace(string(line[:start]))
 	isFirstWord := before == "" || strings.HasSuffix(before, "|") ||
 		strings.HasSuffix(before, ";") || strings.HasSuffix(before, "&&")
+	// ... and after if, then, do etc.
+	if fields := strings.Fields(before); len(fields) > 0 {
+		switch fields[len(fields)-1] {
+		case "if", "then", "elif", "else", "do", "while", "until", "!":
+			isFirstWord = true
+		}
+	}
 
 	// Plain "~" -> "~/"
 	if word == "~" {
@@ -1030,28 +1077,56 @@ func sourceFile(cmd, path string) error {
 	return nil
 }
 
-// runScript executes the commands from r line by line. Like sh, it does not
-// stop at a failing command and returns the exit status of the last one.
-// Empty lines and lines starting with # (e.g. a #!/usr/bin/gosh line) are
-// skipped. If name is set, errors are reported with file name and line.
+// runScript executes the commands from r. Each command is executed as soon
+// as it is complete (a control structure can span several lines). Like sh,
+// it does not stop at a failing command and returns the exit status of the
+// last one. If name is set, errors are reported with file name and line.
 func runScript(r io.Reader, name string) int {
-	status := 0
-	for lineNo := 1; ; lineNo++ {
-		line, err := readLine(r)
-		if trimmed := strings.TrimSpace(line); trimmed != "" && !strings.HasPrefix(trimmed, "#") {
-			cmdErr := execInput(line)
-			status = lastStatus
-			if name != "" && cmdErr != nil {
-				cmdErr = fmt.Errorf("%s:%d: %w", sanitize(name), lineNo, cmdErr)
-			}
-			printError(cmdErr)
+	var (
+		buf       strings.Builder
+		startLine int
+	)
+	report := func(err error, line int) {
+		if name != "" && err != nil {
+			err = fmt.Errorf("%s:%d: %w", sanitize(name), line, err)
 		}
-		if err != nil {
-			if err != io.EOF {
-				fmt.Fprintln(os.Stderr, "gosh:", err)
+		printError(err)
+	}
+
+	for lineNo := 1; ; lineNo++ {
+		line, readErr := readLine(r)
+		if readErr == nil || line != "" {
+			if buf.Len() == 0 {
+				if strings.TrimSpace(line) == "" {
+					goto next
+				}
+				startLine = lineNo
+			}
+			buf.WriteString(line)
+			buf.WriteByte('\n')
+			tree, err := parse(buf.String())
+			if errors.Is(err, errIncomplete) {
+				goto next // read the rest of the command
+			}
+			buf.Reset()
+			if err != nil {
+				lastStatus = exitStatus(err)
+				report(err, startLine)
+			} else {
+				report(tree.exec(), startLine)
+			}
+		}
+	next:
+		if readErr != nil {
+			if readErr != io.EOF {
+				fmt.Fprintln(os.Stderr, "gosh:", readErr)
 				return 1
 			}
-			return status
+			if buf.Len() > 0 {
+				lastStatus = 2
+				report(errors.New("syntax error: unexpected end of file"), startLine)
+			}
+			return lastStatus
 		}
 	}
 }
@@ -1089,10 +1164,16 @@ func exitStatus(err error) int {
 	if errors.As(err, &statusErr) {
 		return int(statusErr)
 	}
-	var synErr syntaxError
-	if errors.As(err, &synErr) {
+	var (
+		synErr syntaxError
+		lc     loopControl
+	)
+	if errors.As(err, &synErr) || errors.Is(err, errIncomplete) {
 		return 2
 	}
+	if errors.As(err, &lc) {
+		return 0
+	}
 	var exitErr *exec.ExitError
 	if errors.As(err, &exitErr) {
 		if ws, ok := exitErr.Sys().(syscall.WaitStatus); ok && ws.Signaled() {
@@ -1144,23 +1225,35 @@ func interactive() {
 	}
 	defer l.Close()
 
-	// Main loop
+	// Main loop. A command that is not complete yet (e.g. if without fi)
+	// is continued on the next line with the prompt "> ".
+	var buf strings.Builder
 	for {
-		// Report finished and stopped background jobs
-		reapJobs(true)
-
-		// Update prompt dynamically (in case directory changed)
-		l.SetPrompt(getPrompt())
+		if buf.Len() == 0 {
+			// Report finished and stopped background jobs
+			reapJobs(true)
+			// Update prompt dynamically (in case directory changed)
+			l.SetPrompt(getPrompt())
+		} else {
+			l.SetPrompt("> ")
+		}
 
 		line, err := l.Readline()
 		if err != nil {
-			// Ctrl+C discards the current line
+			// Ctrl+C discards the current line (and an incomplete command)
 			if err == readline.ErrInterrupt {
+				buf.Reset()
 				continue
 			}
 			// Handle EOF (Ctrl+D)
 			if err == io.EOF {
 				fmt.Println()
+				if buf.Len() > 0 {
+					buf.Reset()
+					lastStatus = 2
+					fmt.Fprintln(os.Stderr, "gosh: syntax error: unexpected end of file")
+					continue
+				}
 				if hasStoppedJobs() && !exitWarned {
 					exitWarned = true
 					fmt.Fprintln(os.Stderr, "gosh: there are stopped jobs")
@@ -1173,7 +1266,7 @@ func interactive() {
 		}
 
 		// Skip empty input
-		if strings.TrimSpace(line) == "" {
+		if buf.Len() == 0 && strings.TrimSpace(line) == "" {
 			continue
 		}
 
@@ -1183,8 +1276,19 @@ func interactive() {
 			l.SaveHistory(line)
 		}
 
-		// Execute the input
-		printError(execInput(line))
+		buf.WriteString(line)
+		buf.WriteByte('\n')
+		tree, err := parse(buf.String())
+		if errors.Is(err, errIncomplete) {
+			continue
+		}
+		buf.Reset()
+		if err != nil {
+			lastStatus = exitStatus(err)
+			printError(err)
+			continue
+		}
+		printError(tree.exec())
 	}
 }
 
@@ -1216,145 +1320,31 @@ func printError(err error) {
 	var (
 		exitErr   *exec.ExitError
 		statusErr statusError
+		lc        loopControl
 	)
-	if err == nil || errors.As(err, &exitErr) || errors.As(err, &statusErr) {
+	if err == nil || errors.As(err, &exitErr) || errors.As(err, &statusErr) || errors.As(err, &lc) {
 		return
 	}
 	fmt.Fprintln(os.Stderr, "gosh:", err)
 }
 
-// listItem is a command of a command list together with the operator that
-// connects it to the previous command ("" for the first one).
-type listItem struct {
-	op         string // "", ";", "&&" or "||"
-	command    string
-	background bool // the command ends with &
-}
-
-// splitList splits input at unquoted ;, &, && and ||. Quotes, escapes, ${...}
-// and comments are handled like in splitWords, but nothing is expanded yet:
-// that happens per command when it runs, so "cd /tmp && echo $PWD" works.
-func splitList(input string) ([]listItem, error) {
-	var (
-		items     []listItem
-		op        string
-		start     int
-		quote     byte
-		escaped   bool
-		wordStart = true // the next character starts a word
-	)
-
-	end := len(input)
-	add := func(i int, next string) error {
-		command := input[start:i]
-		if strings.TrimSpace(command) == "" {
-			// "a;" is fine, but an operator needs a command on both sides
-			if next != "" || op == "&&" || op == "||" {
-				if next == "" {
-					next = op
-				}
-				return syntaxError(next)
-			}
-			return nil
-		}
-		items = append(items, listItem{op: op, command: command})
-		return nil
-	}
-
-scan:
-	for i := 0; i < len(input); i++ {
-		ch := input[i]
-		isWordStart := wordStart
-		wordStart = false
-		switch {
-		case escaped:
-			escaped = false
-		case quote == '\'':
-			if ch == '\'' {
-				quote = 0
-			}
-		case ch == '\\':
-			escaped = true
-		case ch == '$' && i+1 < len(input) && input[i+1] == '{':
-			// Skip ${...} as a whole, like expandVar does
-			if j := strings.IndexByte(input[i:], '}'); j >= 0 {
-				i += j
-			}
-		case quote == '"':
-			if ch == '"' {
-				quote = 0
-			}
-		case ch == '"' || ch == '\'':
-			quote = ch
-		case ch == ' ' || ch == '\t' || ch == '\n':
-			wordStart = true
-		case ch == '#' && isWordStart:
-			end = i
-			break scan
-		case ch == '&' && !(i+1 < len(input) && strings.IndexByte("&>", input[i+1]) >= 0) &&
-			!(i > 0 && strings.IndexByte("<>", input[i-1]) >= 0):
-			// A single & (not &&, &>, >& or <&) runs the command in the
-			// background
-			if err := add(i, "&"); err != nil {
-				return nil, err
-			}
-			items[len(items)-1].background = true
-			op = ";"
-			start = i + 1
-			wordStart = true
-		case ch == ';',
-			ch == '&' && i+1 < len(input) && input[i+1] == '&',
-			ch == '|' && i+1 < len(input) && input[i+1] == '|':
-			next := input[i : i+1]
-			if ch != ';' {
-				next = input[i : i+2]
-			}
-			if err := add(i, next); err != nil {
-				return nil, err
-			}
-			i += len(next) - 1
-			op = next
-			start = i + 1
-			wordStart = true
-		}
-	}
-
-	// An unterminated quote or a trailing \ is reported by splitWords
-	if err := add(end, ""); err != nil {
-		return nil, err
-	}
-	return items, nil
-}
-
-// execInput executes a line of input: a list of commands separated by ;, &&
-// and ||. It returns the error of the last command that was run; errors of
-// earlier commands are printed. $? is updated after each command.
+// execInput parses and executes input: a command line or a whole script.
+// It returns the error of the last command that was run; errors of earlier
+// commands are printed. $? is updated after each command.
 func execInput(input string) error {
-	if interactiveShell {
-		input = expandAliases(input)
-	}
-	items, err := splitList(input)
+	tree, err := parse(input)
 	if err != nil {
 		lastStatus = exitStatus(err)
 		return err
 	}
-
-	var lastErr error
-	for _, item := range items {
-		if (item.op == "&&" && lastStatus != 0) || (item.op == "||" && lastStatus == 0) {
-			continue
-		}
-		printError(lastErr)
-		lastErr = execCommand(item.command, item.background)
-		lastStatus = exitStatus(lastErr)
-	}
-	return lastErr
+	return tree.exec()
 }
 
 // execCommand executes a pipeline of simple commands with redirections, in
 // the background if background is set.
 func execCommand(input string, background bool) error {
 	// Brace expansion comes first, so that its results are expanded further
+	substStatus = 0
 	input, err := expandBraceWords(input)
 	if err != nil {
 		return err
@@ -1363,8 +1353,11 @@ func execCommand(input string, background bool) error {
 	// Split the input into words and operators, expanding variables and ~ on
 	// the way.
 	tokens, err := tokenize(input)
-	var synErr syntaxError
-	if errors.As(err, &synErr) {
+	var (
+		synErr    syntaxError
+		statusErr statusError
+	)
+	if errors.As(err, &synErr) || errors.As(err, &statusErr) {
 		return err
 	}
 	if err != nil {
@@ -1375,6 +1368,10 @@ func execCommand(input string, background bool) error {
 		return err
 	}
 	if len(stages) == 0 {
+		// A command that consists only of $(...) has its status
+		if substStatus != 0 {
+			return statusError(substStatus)
+		}
 		return nil
 	}
 	// "exit" with stopped jobs only warns, unless it is repeated right away
@@ -1648,21 +1645,24 @@ func applyRedirects(fds *[3]*os.File, redirects []redirect) ([]*os.File, error)
 // the shell cannot be used in a pipeline, since there they would run
 // concurrently (sh runs them in a subshell where they have no effect).
 var builtins = map[string]struct{ changesState bool }{
-	"cd":      {changesState: true},
-	"pwd":     {},
-	"exit":    {changesState: true},
-	"quit":    {changesState: true},
-	"set":     {changesState: true},
-	"export":  {changesState: true},
-	"help":    {},
-	"source":  {changesState: true},
-	".":       {changesState: true},
-	"jobs":    {},
-	"fg":      {changesState: true},
-	"bg":      {changesState: true},
-	"wait":    {changesState: true},
-	"alias":   {},
-	"unalias": {changesState: true},
+	"cd":       {changesState: true},
+	"pwd":      {},
+	"exit":     {changesState: true},
+	"quit":     {changesState: true},
+	"set":      {changesState: true},
+	"export":   {changesState: true},
+	"help":     {},
+	"source":   {changesState: true},
+	".":        {changesState: true},
+	"jobs":     {},
+	"fg":       {changesState: true},
+	"bg":       {changesState: true},
+	"wait":     {changesState: true},
+	"alias":    {},
+	"unalias":  {changesState: true},
+	":":        {},
+	"break":    {changesState: true},
+	"continue": {changesState: true},
 }
 
 // runBuiltin executes a built-in command, writing its output to stdout.
@@ -1742,6 +1742,11 @@ func runBuiltin(args []string, stdout io.Writer) error {
 		}
 		os.Setenv(parts[0], parts[1])
 		return nil
+	case ":":
+		// Does nothing, successfully (e.g. while :; do ...)
+		return nil
+	case "break", "continue":
+		return builtinLoopControl(args)
 	case "alias":
 		return builtinAlias(args, stdout)
 	case "unalias":
@@ -1778,6 +1783,9 @@ Built-in commands:
   wait [%n]     - Wait for job n or all background jobs
   alias [name[=value] ...] - Define or show aliases (alias ll='ls -l')
   unalias name  - Remove an alias (unalias -a: all aliases)
+  :             - Do nothing (exit status 0)
+  break [n]     - Leave the innermost (or n-th) for/while/until loop
+  continue [n]  - Continue with the next iteration of a loop
   help          - Show this help
 
 Features:
@@ -1793,14 +1801,23 @@ Features:
   - Wildcards: * ? [abc] [a-z] [!abc] match file names (not hidden files
     unless the pattern starts with a dot); quoted wildcards and values of
     variables are not expanded, a pattern without match is kept as is
-  - Colored prompt (user@host green, red for root; directory blue),
-    configurable with $GOSH_COLORS, e.g. GOSH_COLORS='user=1;33:path=36'
-    (keys user, root, path; values are ANSI SGR codes, empty = no color);
-    disabled with $NO_COLOR, TERM=dumb or when output is not a terminal
+  - Prompt user@host:dir $, with [N] before the $ if the last command
+    failed with exit status N
+  - Colored prompt (user@host green, red for root; directory blue; exit
+    status red), configurable with $GOSH_COLORS, e.g.
+    GOSH_COLORS='user=1;33:path=36' (keys user, root, path, error; values
+    are ANSI SGR codes, empty = no color); disabled with $NO_COLOR,
+    TERM=dumb or when output is not a terminal
   - Environment variable expansion ($VAR, ${VAR}, ${VAR:-default});
     using an unset variable is an error, use ${VAR:-} for an empty default
+  - Command substitution: $(command) is replaced by the output of command
+    (run in a subshell); unquoted it is split into words, "$(command)"
+    stays one word
   - Command lists: a ; b (run both), a && b (b only if a succeeded),
     a || b (b only if a failed); $? is the exit status of the last command
+  - Control structures: if cmd; then ...; elif cmd; then ...; else ...; fi,
+    for x in words; do ...; done, while cmd; do ...; done, until ...,
+    ! cmd negates the status; they can span several lines (prompt "> ")
   - Pipes (a | b) and redirections: < file, > file, >> file, 2> file,
     2>> file, 2>&1, >&2, &> file (stdout and stderr), &>> file
   - Job control: command & runs in the background, Ctrl+Z stops the

main_test.go

@@ -3,6 +3,7 @@ package main
 import (
 	"bytes"
 	"errors"
+	"fmt"
 	"io"
 	"os"
 	"os/exec"
@@ -841,7 +842,7 @@ func TestShellCompleter_Do(t *testing.T) {
 		want   []string
 		length int
 	}{
-		{"c", []string{"d "}, 1},
+		{"c", []string{"d ", "ontinue "}, 1},
 		{"e", []string{"cho ", "xit ", "xport "}, 1},
 		{"gi", []string{"tty "}, 2},
 		{"  gi", []string{"tty "}, 2},
@@ -892,6 +893,11 @@ func TestMain(m *testing.M) {
 	// Never read the system's startup files; tests that read the user's
 	// files set $HOME themselves
 	sysConfDir = filepath.Join(dir, "etc")
+	// Command substitution runs gosh, not the test binary
+	subshellPath = filepath.Join(dir, "gosh")
+	if out, err := exec.Command("go", "build", "-o", subshellPath, ".").CombinedOutput(); err != nil {
+		panic(fmt.Sprintf("building gosh: %v\n%s", err, out))
+	}
 	code := m.Run()
 	os.RemoveAll(dir)
 	os.Exit(code)
@@ -1054,24 +1060,24 @@ func TestGetPrompt_EscapesControlCharacters(t *testing.T) {
 }
 
 func TestBuildPrompt_Color(t *testing.T) {
-	got := buildPrompt("flow", "host", "~/src", false, &defaultColors)
+	got := buildPrompt("flow", "host", "~/src", false, 0, &defaultColors)
 	want := "\x1b[1;32mflow@host\x1b[0m:\x1b[1;34m~/src\x1b[0m $ "
 	if got != want {
 		t.Errorf("buildPrompt() = %q, want %q", got, want)
 	}
 
-	got = buildPrompt("root", "host", "/", true, &defaultColors)
+	got = buildPrompt("root", "host", "/", true, 0, &defaultColors)
 	if !strings.HasPrefix(got, "\x1b[1;31mroot@host") {
 		t.Errorf("expected red user for root, got %q", got)
 	}
 
-	got = buildPrompt("flow", "host", "~/src", false, nil)
+	got = buildPrompt("flow", "host", "~/src", false, 0, nil)
 	if got != "flow@host:~/src $ " {
 		t.Errorf("buildPrompt() without color = %q", got)
 	}
 
 	colors := promptColors{user: "38;5;208", path: ""}
-	got = buildPrompt("flow", "host", "~/src", false, &colors)
+	got = buildPrompt("flow", "host", "~/src", false, 0, &colors)
 	if got != "\x1b[38;5;208mflow@host\x1b[0m:~/src $ " {
 		t.Errorf("buildPrompt() with custom colors = %q", got)
 	}
@@ -1083,13 +1089,14 @@ func TestParseColors(t *testing.T) {
 		want promptColors
 	}{
 		{"", defaultColors},
-		{"user=1;33", promptColors{user: "1;33", root: "1;31", path: "1;34"}},
-		{"user=33:root=41;97:path=38;5;208", promptColors{user: "33", root: "41;97", path: "38;5;208"}},
-		{"path=", promptColors{user: "1;32", root: "1;31", path: ""}},
+		{"user=1;33", promptColors{user: "1;33", root: "1;31", path: "1;34", error: "1;31"}},
+		{"user=33:root=41;97:path=38;5;208", promptColors{user: "33", root: "41;97", path: "38;5;208", error: "1;31"}},
+		{"error=33", promptColors{user: "1;32", root: "1;31", path: "1;34", error: "33"}},
+		{"path=", promptColors{user: "1;32", root: "1;31", path: "", error: "1;31"}},
 		{"path=0m\x1b[2J", defaultColors}, // escape sequences are rejected
 		{"path=red", defaultColors},       // only numeric codes
 		{"foo=1:user", defaultColors},     // unknown key, missing value
-		{"user=35::path=36", promptColors{user: "35", root: "1;31", path: "36"}},
+		{"user=35::path=36", promptColors{user: "35", root: "1;31", path: "36", error: "1;31"}},
 	}
 	for _, tc := range tests {
 		if got := parseColors(tc.spec); got != tc.want {
@@ -1098,8 +1105,27 @@ func TestParseColors(t *testing.T) {
 	}
 }
 
+func TestBuildPrompt_Status(t *testing.T) {
+	if got := buildPrompt("flow", "host", "~", false, 0, nil); got != "flow@host:~ $ " {
+		t.Errorf("status 0: %q", got)
+	}
+	if got := buildPrompt("flow", "host", "~", false, 127, nil); got != "flow@host:~ [127] $ " {
+		t.Errorf("status 127 without color: %q", got)
+	}
+	got := buildPrompt("flow", "host", "~", false, 1, &defaultColors)
+	want := "\x1b[1;32mflow@host\x1b[0m:\x1b[1;34m~\x1b[0m \x1b[1;31m[1]\x1b[0m $ "
+	if got != want {
+		t.Errorf("status 1 with color = %q, want %q", got, want)
+	}
+	colors := defaultColors
+	colors.error = ""
+	if got := buildPrompt("flow", "host", "~", false, 2, &colors); !strings.HasSuffix(got, " [2] $ ") {
+		t.Errorf("status without error color = %q", got)
+	}
+}
+
 func TestBuildPrompt_ColorEscapesControlCharacters(t *testing.T) {
-	got := buildPrompt("flow", "host", "evil\x1b[2J\x07", false, &defaultColors)
+	got := buildPrompt("flow", "host", "evil\x1b[2J\x07", false, 0, &defaultColors)
 	if strings.Contains(got, "\x1b[2J") || strings.Contains(got, "\x07") {
 		t.Errorf("prompt contains control characters from path: %q", got)
 	}
@@ -1361,48 +1387,170 @@ func TestReadLine_DoesNotReadAhead(t *testing.T) {
 	}
 }
 
-func TestSplitList(t *testing.T) {
+// treeString formats a syntax tree for the tests: simple commands in
+// brackets, background commands with a trailing &.
+func treeString(l *listNode) string {
+	var parts []string
+	for _, e := range l.items {
+		s := pipelineString(e.andOr.first)
+		for _, r := range e.andOr.rest {
+			s += " " + r.op + " " + pipelineString(r.pipeline)
+		}
+		if e.background {
+			s += " &"
+		}
+		parts = append(parts, s)
+	}
+	return strings.Join(parts, "; ")
+}
+
+func pipelineString(pl *pipelineNode) string {
+	s := ""
+	if pl.negate {
+		s = "! "
+	}
+	switch n := pl.cmd.(type) {
+	case *simpleNode:
+		s += "[" + n.text + "]"
+	case *ifNode:
+		s += "if"
+		for i := range n.conds {
+			if i > 0 {
+				s += " elif"
+			}
+			s += " " + treeString(n.conds[i]) + " then " + treeString(n.bodies[i])
+		}
+		if n.elseBody != nil {
+			s += " else " + treeString(n.elseBody)
+		}
+		s += " fi"
+	case *forNode:
+		s += "for " + n.name
+		if n.hasIn {
+			s += " in (" + n.words + ")"
+		}
+		s += " do " + treeString(n.body) + " done"
+	case *whileNode:
+		if n.until {
+			s += "until "
+		} else {
+			s += "while "
+		}
+		s += treeString(n.cond) + " do " + treeString(n.body) + " done"
+	}
+	return s
+}
+
+func TestParse_Lists(t *testing.T) {
 	tests := []struct {
 		input string
-		want  []listItem
+		want  string
 	}{
-		{"", nil},
-		{"   ", nil},
-		{"# only a comment", nil},
-		{"ls", []listItem{{"", "ls", false}}},
-		{"a; b", []listItem{{"", "a", false}, {";", " b", false}}},
-		{"a;", []listItem{{"", "a", false}}},
-		{"a && b || c", []listItem{{"", "a ", false}, {"&&", " b ", false}, {"||", " c", false}}},
-		{`echo "a;b" 'c&&d' e\;f`, []listItem{{"", `echo "a;b" 'c&&d' e\;f`, false}}},
-		{`echo "x\"; y"`, []listItem{{"", `echo "x\"; y"`, false}}},
-		{"echo ${X:-a;b}; c", []listItem{{"", "echo ${X:-a;b}", false}, {";", " c", false}}},
-		{"echo a # b; c", []listItem{{"", "echo a ", false}}},
-		{"echo a#b; c", []listItem{{"", "echo a#b", false}, {";", " c", false}}},
-		{"echo a|b", []listItem{{"", "echo a|b", false}}},
-		{"a & b", []listItem{{"", "a ", true}, {";", " b", false}}},
-		{"a &", []listItem{{"", "a ", true}}},
-		{"a&", []listItem{{"", "a", true}}},
-		{"a 2>&1 &", []listItem{{"", "a 2>&1 ", true}}},
-		{"a &> f &", []listItem{{"", "a &> f ", true}}},
-		{"a >&2 && b", []listItem{{"", "a >&2 ", false}, {"&&", " b", false}}},
-		{`echo "a & b" \&`, []listItem{{"", `echo "a & b" \&`, false}}},
+		{"", ""},
+		{"   ", ""},
+		{"# only a comment", ""},
+		{"ls", "[ls]"},
+		{"a; b", "[a]; [b]"},
+		{"a;", "[a]"},
+		{"a\nb\n", "[a]; [b]"},
+		{"a && b || c", "[a] && [b] || [c]"},
+		{"a &&\n  b", "[a] && [b]"},
+		{"a | b|c", "[a | b|c]"},
+		{"a |\n b", "[a |\n b]"},
+		{`echo "a;b" 'c&&d' e\;f`, `[echo "a;b" 'c&&d' e\;f]`},
+		{`echo "x\"; y"`, `[echo "x\"; y"]`},
+		{"echo ${X:-a;b}; c", "[echo ${X:-a;b}]; [c]"},
+		{"echo a # b; c", "[echo a]"},
+		{"echo a#b; c", "[echo a#b]; [c]"},
+		{"a & b", "[a] &; [b]"},
+		{"a&", "[a] &"},
+		{"a 2>&1 &", "[a 2>&1] &"},
+		{"a &> f &", "[a &> f] &"},
+		{"a >&2 && b", "[a >&2] && [b]"},
+		{`echo "a & b" \&`, `[echo "a & b" \&]`},
+		{"echo $(a; b && c | d) \"$(e || f)\" & g", `[echo $(a; b && c | d) "$(e || f)"] &; [g]`},
+		{"! false", "! [false]"},
+		{"echo a \\\n b", "[echo a  b]"},
+		// Reserved words are only special as the first word of a command
+		{"echo if then fi do done", "[echo if then fi do done]"},
 	}
 	for _, tc := range tests {
-		got, err := splitList(tc.input)
+		tree, err := parse(tc.input)
 		if err != nil {
-			t.Errorf("splitList(%q) returned error: %v", tc.input, err)
+			t.Errorf("parse(%q) returned error: %v", tc.input, err)
 			continue
 		}
-		if !reflect.DeepEqual(got, tc.want) {
-			t.Errorf("splitList(%q) = %+v, want %+v", tc.input, got, tc.want)
+		if got := treeString(tree); got != tc.want {
+			t.Errorf("parse(%q) = %s, want %s", tc.input, got, tc.want)
 		}
 	}
 }
 
-func TestSplitList_SyntaxErrors(t *testing.T) {
-	for _, input := range []string{"; a", "a ;; b", "&& a", "a &&", "a ||", "a && ; b", "a || && b", "& a", "a & && b", "a & & b"} {
-		if _, err := splitList(input); err == nil || !strings.Contains(err.Error(), "syntax error") {
-			t.Errorf("splitList(%q) = %v, want syntax error", input, err)
+func TestParse_ControlStructures(t *testing.T) {
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"if a; then b; fi", "if [a] then [b] fi"},
+		{"if a\nthen\n  b\n  c\nfi", "if [a] then [b]; [c] fi"},
+		{"if a; then b; elif c; then d; else e; fi", "if [a] then [b] elif [c] then [d] else [e] fi"},
+		{"if a && ! b; then c; fi; echo x", "if [a] && ! [b] then [c] fi; [echo x]"},
+		{"for x in a b c; do echo $x; done", "for x in (a b c) do [echo $x] done"},
+		{"for x in *.go {1..3}\ndo\n  echo $x\ndone", "for x in (*.go {1..3}) do [echo $x] done"},
+		{"for x; do echo; done", "for x do [echo] done"},
+		{"for x in; do echo; done", "for x in () do [echo] done"},
+		{"while a; do b; done", "while [a] do [b] done"},
+		{"until a; do b; done", "until [a] do [b] done"},
+		{"while a; do if b; then break; fi; done", "while [a] do if [b] then [break] fi done"},
+		{"for i in 1 2; do for j in 3 4; do echo $i$j; done; done", "for i in (1 2) do for j in (3 4) do [echo $i$j] done done"},
+		{"if a; then if b; then c; fi fi", "if [a] then if [b] then [c] fi fi"},
+	}
+	for _, tc := range tests {
+		tree, err := parse(tc.input)
+		if err != nil {
+			t.Errorf("parse(%q) returned error: %v", tc.input, err)
+			continue
+		}
+		if got := treeString(tree); got != tc.want {
+			t.Errorf("parse(%q) = %s, want %s", tc.input, got, tc.want)
+		}
+	}
+}
+
+func TestParse_Incomplete(t *testing.T) {
+	for _, input := range []string{
+		"if a", "if a; then", "if a; then b", "if a; then b; else", "if a; then b; elif c",
+		"for x", "for x in a b", "for x in a; do", "for x in a; do b",
+		"while a", "while a; do b", "until a; do",
+		"a &&", "a ||", "a |", "echo 'open", `echo "open`, "echo $(a", "echo a \\",
+	} {
+		if _, err := parse(input); !errors.Is(err, errIncomplete) {
+			t.Errorf("parse(%q) = %v, want incomplete", input, err)
+		}
+	}
+}
+
+func TestParse_SyntaxErrors(t *testing.T) {
+	for _, input := range []string{
+		"; a", "a ;; b", "&& a", "a && ; b", "a || && b", "& a", "a & && b", "a & & b",
+		"fi", "then", "done", "if a; then fi", "if ; then a; fi", "while a; do done",
+		"if a; then b; done", "for 1x in a; do b; done", "for x in a do b; done; x",
+		"a | | b", "| a",
+	} {
+		_, err := parse(input)
+		if err == nil || errors.Is(err, errIncomplete) {
+			t.Errorf("parse(%q) = %v, want syntax error", input, err)
+		}
+	}
+	for input, want := range map[string]string{
+		"for x in a; do b; done | cat": "control structures in pipelines are not supported",
+		"echo | while a; do b; done":   "control structures in pipelines are not supported",
+		"if a; then b; fi > out":       "redirections of control structures are not supported (near `>')",
+		"for x in a; do b; done &":     "only simple commands and pipelines can run in the background",
+		"a && b &":                     "only simple commands and pipelines can run in the background",
+	} {
+		if _, err := parse(input); err == nil || err.Error() != want {
+			t.Errorf("parse(%q) = %v, want %q", input, err, want)
 		}
 	}
 }

parser.go

@@ -0,0 +1,841 @@
+package main
+
+// Parser and executor for command lines and scripts: lists (; & && || and
+// newlines), pipelines, ! and the control structures if, for, while and
+// until. Simple commands (with their pipes and redirections) are kept as
+// text; they are expanded and run by execCommand only when they are
+// executed, so that e.g. the variable of a for loop is expanded anew in
+// every iteration.
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"strconv"
+	"strings"
+	"syscall"
+)
+
+// incompleteError is returned by parse if the input ends inside a
+// construct (if without fi, an open quote, a trailing && ...): an
+// interactive shell then reads another line.
+type incompleteError struct{}
+
+func (incompleteError) Error() string { return "syntax error: unexpected end of input" }
+
+var errIncomplete error = incompleteError{}
+
+// loopControl is returned by break and continue and ends the given number of
+// enclosing loops (break) or iterations (continue).
+type loopControl struct {
+	isBreak bool
+	n       int
+}
+
+func (c loopControl) Error() string {
+	if c.isBreak {
+		return "break"
+	}
+	return "continue"
+}
+
+// loopDepth is the number of loops that are currently running.
+var loopDepth int
+
+// aliasExpansion is false while the result of an alias expansion runs, so
+// that it is not expanded again.
+var aliasExpansion = true
+
+// reserved words that end a list; they are only recognized as the first word
+// of a command.
+var terminators = map[string]bool{"then": true, "elif": true, "else": true, "fi": true, "do": true, "done": true}
+
+// ---------------------------------------------------------------------------
+// Lexer
+
+type tokenKind int
+
+const (
+	tWord tokenKind = iota
+	tOp             // ; & && || | or newline
+)
+
+type rawToken struct {
+	kind       tokenKind
+	text       string
+	start, end int
+}
+
+// joinContinuationLines removes backslash-newline (outside single quotes),
+// which continues a line. continued reports whether src ends with such a
+// continuation (so that the command goes on in the next line).
+func joinContinuationLines(src string) (joined string, continued bool) {
+	if !strings.Contains(src, "\\\n") {
+		return src, false
+	}
+	var b strings.Builder
+	var quote byte
+	for i := 0; i < len(src); i++ {
+		ch := src[i]
+		switch {
+		case quote == '\'':
+			if ch == '\'' {
+				quote = 0
+			}
+		case ch == '\\' && i+1 < len(src):
+			if src[i+1] == '\n' {
+				i++
+				continued = i == len(src)-1
+				continue
+			}
+			continued = false
+			b.WriteByte(ch)
+			i++
+			ch = src[i]
+		case ch == '"':
+			if quote == '"' {
+				quote = 0
+			} else {
+				quote = '"'
+			}
+		case ch == '\'' && quote == 0:
+			quote = '\''
+		}
+		if ch != '\n' && ch != ' ' && ch != '\t' {
+			continued = false
+		}
+		b.WriteByte(ch)
+	}
+	return b.String(), continued
+}
+
+// lex splits src into words and list operators. Words keep their quotes,
+// escapes, $(...) and redirections; they are only split at blanks and list
+// operators.
+func lex(src string) ([]rawToken, error) {
+	var toks []rawToken
+	op := func(text string, i int) {
+		toks = append(toks, rawToken{tOp, text, i, i + len(text)})
+	}
+	for i := 0; i < len(src); {
+		ch := src[i]
+		next := byte(0)
+		if i+1 < len(src) {
+			next = src[i+1]
+		}
+		switch {
+		case ch == ' ' || ch == '\t':
+			i++
+		case ch == '\n' || ch == ';':
+			op(string(ch), i)
+			i++
+		case ch == '#':
+			// Comment until the end of the line
+			for i < len(src) && src[i] != '\n' {
+				i++
+			}
+		case ch == '|' && next == '|', ch == '&' && next == '&':
+			op(src[i:i+2], i)
+			i += 2
+		case ch == '|', ch == '&' && next != '>':
+			op(string(ch), i)
+			i++
+		default:
+			end, err := lexWord(src, i)
+			if err != nil {
+				return nil, err
+			}
+			toks = append(toks, rawToken{tWord, src[i:end], i, end})
+			i = end
+		}
+	}
+	return toks, nil
+}
+
+// lexWord returns the end of the word that starts at src[i].
+func lexWord(src string, i int) (int, error) {
+	var quote byte
+	for j := i; j < len(src); j++ {
+		ch := src[j]
+		switch {
+		case quote == '\'':
+			if ch == '\'' {
+				quote = 0
+			}
+		case ch == '\\':
+			if j+1 == len(src) {
+				return 0, errIncomplete
+			}
+			j++
+		case isSubstStart(src, j):
+			end := substEnd(src, j)
+			if end < 0 {
+				return 0, errIncomplete
+			}
+			j = end
+		case ch == '$' && j+1 < len(src) && src[j+1] == '{':
+			if k := strings.IndexByte(src[j:], '}'); k >= 0 {
+				j += k
+			}
+		case quote == '"':
+			if ch == '"' {
+				quote = 0
+			}
+		case ch == '"' || ch == '\'':
+			quote = ch
+		case ch == ' ' || ch == '\t' || ch == '\n' || ch == ';' || ch == '|':
+			return j, nil
+		case ch == '&':
+			// &>, >& and <& are redirections, not the & operator
+			if !(j+1 < len(src) && src[j+1] == '>') && !(j > i && (src[j-1] == '>' || src[j-1] == '<')) {
+				return j, nil
+			}
+		}
+	}
+	if quote != 0 {
+		return 0, errIncomplete
+	}
+	return len(src), nil
+}
+
+// ---------------------------------------------------------------------------
+// Syntax tree
+
+// node is a command that can be executed: it returns the error of the last
+// command it ran (so that exitStatus gives its status).
+type node interface {
+	exec() error
+}
+
+// listNode is a list of and-or lists separated by ;, & or newlines.
+type listNode struct {
+	items []listEntry
+}
+
+type listEntry struct {
+	andOr      *andOrNode
+	background bool
+}
+
+// andOrNode is a pipeline followed by && and || pipelines.
+type andOrNode struct {
+	first *pipelineNode
+	rest  []andOrPart
+}
+
+type andOrPart struct {
+	op       string // && or ||
+	pipeline *pipelineNode
+}
+
+// pipelineNode is a simple command (whose text may contain a pipeline) or a
+// control structure, optionally negated with !.
+type pipelineNode struct {
+	negate bool
+	cmd    node
+}
+
+// simpleNode is the text of a pipeline of simple commands.
+type simpleNode struct {
+	text string
+}
+
+type ifNode struct {
+	conds    []*listNode
+	bodies   []*listNode
+	elseBody *listNode
+}
+
+type forNode struct {
+	name  string
+	words string // the words after "in", expanded when the loop starts
+	hasIn bool
+	body  *listNode
+}
+
+type whileNode struct {
+	until bool
+	cond  *listNode
+	body  *listNode
+}
+
+// singleSimple returns the simple command if the list is just that.
+func (l *listNode) singleSimple() *simpleNode {
+	if len(l.items) != 1 || l.items[0].background {
+		return nil
+	}
+	ao := l.items[0].andOr
+	if len(ao.rest) > 0 || ao.first.negate {
+		return nil
+	}
+	s, _ := ao.first.cmd.(*simpleNode)
+	return s
+}
+
+// ---------------------------------------------------------------------------
+// Parser
+
+type parser struct {
+	src  string
+	toks []rawToken
+	pos  int
+}
+
+// parse parses a command line or script. It returns errIncomplete if the
+// input ends inside a construct.
+func parse(src string) (*listNode, error) {
+	src, continued := joinContinuationLines(src)
+	if continued {
+		return nil, errIncomplete
+	}
+	toks, err := lex(src)
+	if err != nil {
+		return nil, err
+	}
+	p := &parser{src: src, toks: toks}
+	list, err := p.parseList(nil)
+	if err != nil {
+		return nil, err
+	}
+	if t := p.peek(); t != nil {
+		return nil, p.unexpected(t)
+	}
+	return list, nil
+}
+
+func (p *parser) peek() *rawToken {
+	if p.pos < len(p.toks) {
+		return &p.toks[p.pos]
+	}
+	return nil
+}
+
+func (p *parser) peekWord(words ...string) bool {
+	t := p.peek()
+	if t == nil || t.kind != tWord {
+		return false
+	}
+	for _, w := range words {
+		if t.text == w {
+			return true
+		}
+	}
+	return false
+}
+
+func (p *parser) skipNewlines() {
+	for t := p.peek(); t != nil && t.kind == tOp && t.text == "\n"; t = p.peek() {
+		p.pos++
+	}
+}
+
+func (p *parser) unexpected(t *rawToken) error {
+	if t.text == "\n" {
+		return syntaxError("newline")
+	}
+	return syntaxError(t.text)
+}
+
+// expect consumes the reserved word w.
+func (p *parser) expect(w string) error {
+	t := p.peek()
+	if t == nil {
+		return errIncomplete
+	}
+	if t.kind != tWord || t.text != w {
+		return p.unexpected(t)
+	}
+	p.pos++
+	return nil
+}
+
+// parseList parses and-or lists until the end of the input or one of the
+// reserved words in terms at the start of a command.
+func (p *parser) parseList(terms map[string]bool) (*listNode, error) {
+	list := &listNode{}
+	for {
+		p.skipNewlines()
+		t := p.peek()
+		if t == nil || t.kind == tWord && terms[t.text] {
+			return list, nil
+		}
+		ao, err := p.parseAndOr()
+		if err != nil {
+			return nil, err
+		}
+		entry := listEntry{andOr: ao}
+
+		t = p.peek()
+		switch {
+		case t == nil:
+		case t.kind == tOp && (t.text == ";" || t.text == "&" || t.text == "\n"):
+			p.pos++
+			entry.background = t.text == "&"
+		case t.kind == tWord && terms[t.text]:
+			// e.g. "fi" right after the end of a nested control structure
+		default:
+			return nil, p.unexpected(t)
+		}
+		if entry.background {
+			if _, ok := ao.first.cmd.(*simpleNode); !ok || len(ao.rest) > 0 || ao.first.negate {
+				return nil, errors.New("only simple commands and pipelines can run in the background")
+			}
+		}
+		list.items = append(list.items, entry)
+	}
+}
+
+// nonEmpty returns an error if a list that must contain commands is empty.
+func (p *parser) nonEmpty(list *listNode) error {
+	if len(list.items) > 0 {
+		return nil
+	}
+	if t := p.peek(); t != nil {
+		return p.unexpected(t)
+	}
+	return errIncomplete
+}
+
+func (p *parser) parseAndOr() (*andOrNode, error) {
+	first, err := p.parsePipeline()
+	if err != nil {
+		return nil, err
+	}
+	ao := &andOrNode{first: first}
+	for {
+		t := p.peek()
+		if t == nil || t.kind != tOp || (t.text != "&&" && t.text != "||") {
+			return ao, nil
+		}
+		p.pos++
+		p.skipNewlines()
+		if p.peek() == nil {
+			return nil, errIncomplete
+		}
+		next, err := p.parsePipeline()
+		if err != nil {
+			return nil, err
+		}
+		ao.rest = append(ao.rest, andOrPart{op: t.text, pipeline: next})
+	}
+}
+
+func (p *parser) parsePipeline() (*pipelineNode, error) {
+	pl := &pipelineNode{}
+	if p.peekWord("!") {
+		pl.negate = true
+		p.pos++
+	}
+	t := p.peek()
+	if t == nil {
+		return nil, errIncomplete
+	}
+	if t.kind == tOp || terminators[t.text] {
+		return nil, p.unexpected(t)
+	}
+
+	var err error
+	compound := true
+	switch t.text {
+	case "if":
+		pl.cmd, err = p.parseIf()
+	case "for":
+		pl.cmd, err = p.parseFor()
+	case "while", "until":
+		pl.cmd, err = p.parseWhile()
+	default:
+		compound = false
+		pl.cmd, err = p.parseSimple()
+	}
+	if err != nil {
+		return nil, err
+	}
+
+	if compound {
+		if t := p.peek(); t != nil {
+			switch {
+			case t.kind == tOp && t.text == "|":
+				return nil, errors.New("control structures in pipelines are not supported")
+			case t.kind == tWord && !terminators[t.text]:
+				return nil, fmt.Errorf("redirections of control structures are not supported (near `%s')", t.text)
+			}
+		}
+	}
+	return pl, nil
+}
+
+// parseSimple parses a pipeline of simple commands and returns its text.
+func (p *parser) parseSimple() (*simpleNode, error) {
+	start := p.toks[p.pos].start
+	end := start
+	for t := p.peek(); t != nil; t = p.peek() {
+		if t.kind == tWord {
+			end = t.end
+			p.pos++
+			continue
+		}
+		if t.text != "|" {
+			break
+		}
+		p.pos++
+		end = t.end
+		p.skipNewlines()
+		next := p.peek()
+		switch {
+		case next == nil:
+			return nil, errIncomplete
+		case next.kind == tOp || terminators[next.text]:
+			return nil, p.unexpected(next)
+		case next.text == "if" || next.text == "for" || next.text == "while" || next.text == "until":
+			return nil, errors.New("control structures in pipelines are not supported")
+		}
+	}
+	return &simpleNode{text: p.src[start:end]}, nil
+}
+
+func (p *parser) parseIf() (*ifNode, error) {
+	p.pos++ // if
+	n := &ifNode{}
+	for {
+		cond, err := p.parseList(map[string]bool{"then": true})
+		if err == nil {
+			err = p.nonEmpty(cond)
+		}
+		if err == nil {
+			err = p.expect("then")
+		}
+		if err != nil {
+			return nil, err
+		}
+		body, err := p.parseList(map[string]bool{"elif": true, "else": true, "fi": true})
+		if err == nil {
+			err = p.nonEmpty(body)
+		}
+		if err != nil {
+			return nil, err
+		}
+		n.conds = append(n.conds, cond)
+		n.bodies = append(n.bodies, body)
+
+		t := p.peek()
+		if t == nil {
+			return nil, errIncomplete
+		}
+		p.pos++
+		switch t.text {
+		case "elif":
+			continue
+		case "else":
+			body, err := p.parseList(map[string]bool{"fi": true})
+			if err == nil {
+				err = p.nonEmpty(body)
+			}
+			if err == nil {
+				err = p.expect("fi")
+			}
+			if err != nil {
+				return nil, err
+			}
+			n.elseBody = body
+		}
+		return n, nil // fi
+	}
+}
+
+func (p *parser) parseFor() (*forNode, error) {
+	p.pos++ // for
+	t := p.peek()
+	if t == nil {
+		return nil, errIncomplete
+	}
+	if t.kind != tWord || !isVarName(t.text) {
+		return nil, fmt.Errorf("for: `%s': not a valid identifier", sanitize(t.text))
+	}
+	n := &forNode{name: t.text}
+	p.pos++
+
+	p.skipNewlines()
+	if p.peekWord("in") {
+		p.pos++
+		n.hasIn = true
+		var words []string
+		for t := p.peek(); t != nil && t.kind == tWord; t = p.peek() {
+			words = append(words, t.text)
+			p.pos++
+		}
+		n.words = strings.Join(words, " ")
+	}
+	t = p.peek()
+	switch {
+	case t == nil:
+		return nil, errIncomplete
+	case t.kind == tOp && (t.text == ";" || t.text == "\n"):
+		p.pos++
+	case !n.hasIn && t.kind == tWord && t.text == "do":
+	default:
+		return nil, p.unexpected(t)
+	}
+
+	p.skipNewlines()
+	body, err := p.parseLoopBody()
+	if err != nil {
+		return nil, err
+	}
+	n.body = body
+	return n, nil
+}
+
+func (p *parser) parseWhile() (*whileNode, error) {
+	n := &whileNode{until: p.peek().text == "until"}
+	p.pos++
+	cond, err := p.parseList(map[string]bool{"do": true})
+	if err == nil {
+		err = p.nonEmpty(cond)
+	}
+	if err != nil {
+		return nil, err
+	}
+	n.cond = cond
+	body, err := p.parseLoopBody()
+	if err != nil {
+		return nil, err
+	}
+	n.body = body
+	return n, nil
+}
+
+// parseLoopBody parses "do list done".
+func (p *parser) parseLoopBody() (*listNode, error) {
+	if err := p.expect("do"); err != nil {
+		return nil, err
+	}
+	body, err := p.parseList(map[string]bool{"done": true})
+	if err == nil {
+		err = p.nonEmpty(body)
+	}
+	if err == nil {
+		err = p.expect("done")
+	}
+	return body, err
+}
+
+// ---------------------------------------------------------------------------
+// Execution
+
+// isAbort reports whether err stops the execution of the enclosing lists:
+// break, continue, or a command that was interrupted with Ctrl+C (like in
+// bash, Ctrl+C aborts the whole command line, e.g. a loop).
+func isAbort(err error) bool {
+	var lc loopControl
+	return errors.As(err, &lc) || exitStatus(err) == 128+int(syscall.SIGINT)
+}
+
+// silent returns an error with only the exit status of err, for errors that
+// were already reported.
+func silent(err error) error {
+	if err == nil {
+		return nil
+	}
+	return statusError(exitStatus(err))
+}
+
+func (l *listNode) exec() error {
+	var err error
+	for _, entry := range l.items {
+		printError(err)
+		if entry.background {
+			err = entry.andOr.first.cmd.(*simpleNode).run(true)
+		} else {
+			err = entry.andOr.exec()
+		}
+		lastStatus = exitStatus(err)
+		if isAbort(err) {
+			return err
+		}
+	}
+	return err
+}
+
+func (a *andOrNode) exec() error {
+	err := a.first.exec()
+	for _, part := range a.rest {
+		lastStatus = exitStatus(err)
+		if isAbort(err) {
+			return err
+		}
+		if (part.op == "&&") != (lastStatus == 0) {
+			continue
+		}
+		printError(err)
+		err = part.pipeline.exec()
+	}
+	return err
+}
+
+func (pl *pipelineNode) exec() error {
+	err := pl.cmd.exec()
+	if !pl.negate || isAbort(err) {
+		return err
+	}
+	printError(err)
+	if exitStatus(err) == 0 {
+		return statusError(1)
+	}
+	return nil
+}
+
+func (s *simpleNode) exec() error {
+	return s.run(false)
+}
+
+// run executes the simple command, in the background if background is set.
+// Aliases are expanded here (in interactive shells), so that they work at
+// the start of every command, also inside control structures.
+func (s *simpleNode) run(background bool) error {
+	text := s.text
+	if interactiveShell && aliasExpansion {
+		if expanded := expandAliases(text); expanded != text {
+			tree, err := parse(expanded)
+			if err != nil {
+				return err
+			}
+			if simple := tree.singleSimple(); simple != nil {
+				text = simple.text
+			} else {
+				if background {
+					return errors.New("an alias with a command list cannot run in the background")
+				}
+				aliasExpansion = false
+				defer func() { aliasExpansion = true }()
+				return tree.exec()
+			}
+		}
+	}
+	return execCommand(text, background)
+}
+
+func (n *ifNode) exec() error {
+	for i, cond := range n.conds {
+		err := cond.exec()
+		if isAbort(err) {
+			return err
+		}
+		printError(err)
+		if exitStatus(err) == 0 {
+			return n.bodies[i].exec()
+		}
+	}
+	if n.elseBody != nil {
+		return n.elseBody.exec()
+	}
+	lastStatus = 0
+	return nil
+}
+
+// runLoopBody runs one iteration of a loop. It returns whether the loop
+// must stop, and the error to return from the loop.
+func runLoopBody(body *listNode, result *error) (stop bool, err error) {
+	err = body.exec()
+	var lc loopControl
+	if errors.As(err, &lc) {
+		*result = nil
+		if lc.n > 1 {
+			lc.n--
+			return true, lc // ends an outer loop too
+		}
+		return lc.isBreak, nil
+	}
+	if isAbort(err) {
+		return true, err
+	}
+	printError(err)
+	*result = silent(err)
+	return false, nil
+}
+
+func (n *whileNode) exec() error {
+	loopDepth++
+	defer func() { loopDepth-- }()
+
+	var result error
+	for {
+		err := n.cond.exec()
+		if isAbort(err) {
+			return err
+		}
+		printError(err)
+		if (exitStatus(err) == 0) == n.until {
+			break
+		}
+		if stop, err := runLoopBody(n.body, &result); stop {
+			if err != nil {
+				return err
+			}
+			break
+		}
+	}
+	lastStatus = exitStatus(result)
+	return result
+}
+
+func (n *forNode) exec() error {
+	var words []string
+	if n.hasIn {
+		var err error
+		if words, err = expandWordList(n.words); err != nil {
+			return err
+		}
+	}
+
+	loopDepth++
+	defer func() { loopDepth-- }()
+
+	var result error
+	for _, word := range words {
+		os.Setenv(n.name, word)
+		if stop, err := runLoopBody(n.body, &result); stop {
+			if err != nil {
+				return err
+			}
+			break
+		}
+	}
+	lastStatus = exitStatus(result)
+	return result
+}
+
+// expandWordList expands the words of a for loop like the arguments of a
+// command (braces, ~, variables, $(...), wildcards).
+func expandWordList(text string) ([]string, error) {
+	text, err := expandBraceWords(text)
+	if err != nil {
+		return nil, err
+	}
+	tokens, err := tokenize(text)
+	if err != nil {
+		return nil, fmt.Errorf("invalid input: %w", err)
+	}
+	var words []string
+	for _, t := range tokens {
+		if t.op != "" {
+			return nil, syntaxError(t.op)
+		}
+		words = append(words, expandWord(t)...)
+	}
+	return words, nil
+}
+
+// builtinLoopControl implements break [n] and continue [n].
+func builtinLoopControl(args []string) error {
+	n := 1
+	if len(args) > 1 {
+		var err error
+		if n, err = strconv.Atoi(args[1]); err != nil || n < 1 {
+			return fmt.Errorf("%s: %s: loop count out of range", args[0], sanitize(args[1]))
+		}
+	}
+	if loopDepth == 0 {
+		return fmt.Errorf("%s: only meaningful in a loop", args[0])
+	}
+	return loopControl{isBreak: args[0] == "break", n: min(n, loopDepth)}
+}

parser_test.go

@@ -0,0 +1,164 @@
+package main
+
+import (
+	"os"
+	"strings"
+	"testing"
+)
+
+// runIn runs input in a temporary directory and returns the content of the
+// file "out" there.
+func runIn(t *testing.T, input string) string {
+	t.Helper()
+	origDir, _ := os.Getwd()
+	defer os.Chdir(origDir)
+	os.Chdir(t.TempDir())
+	execInput(input)
+	data, _ := os.ReadFile("out")
+	return string(data)
+}
+
+func TestExec_If(t *testing.T) {
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"if true; then echo yes >> out; fi", "yes\n"},
+		{"if false; then echo yes >> out; fi", ""},
+		{"if false; then echo a >> out; else echo b >> out; fi", "b\n"},
+		{"if false; then echo a >> out; elif true; then echo b >> out; else echo c >> out; fi", "b\n"},
+		{"if false; then echo a >> out; elif false; then echo b >> out; else echo c >> out; fi", "c\n"},
+		{"if ! false; then echo negated >> out; fi", "negated\n"},
+		{"if true && false; then echo a >> out; else echo b >> out; fi", "b\n"},
+		{"if false || true; then echo a >> out; fi", "a\n"},
+		{"if echo cond >> out; then echo body >> out; fi", "cond\nbody\n"},
+		{"if true\nthen\n  echo multi >> out\n  echo line >> out\nfi", "multi\nline\n"},
+		{"if true; then if false; then echo a >> out; else echo nested >> out; fi; fi", "nested\n"},
+		{"if [ -d / ]; then echo dir >> out; fi", "dir\n"},
+	}
+	for _, tc := range tests {
+		if got := runIn(t, tc.input); got != tc.want {
+			t.Errorf("%q: out = %q, want %q", tc.input, got, tc.want)
+		}
+	}
+}
+
+func TestExec_Loops(t *testing.T) {
+	defer os.Unsetenv("GOSH_I")
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"for x in a b c; do echo $x >> out; done", "a\nb\nc\n"},
+		{"for x in {1..3}; do echo $x >> out; done", "1\n2\n3\n"},
+		{"for x in $(echo a b); do echo $x >> out; done", "a\nb\n"},
+		{"for x in \"a b\" c; do echo $x >> out; done", "a b\nc\n"},
+		{"touch f1 f2; for f in f*; do echo $f >> out; done", "f1\nf2\n"},
+		{"for x in; do echo never >> out; done", ""},
+		{"for x; do echo never >> out; done", ""},
+		{"set GOSH_I=0; while [ $GOSH_I -lt 3 ]; do echo $GOSH_I >> out; set GOSH_I=$(expr $GOSH_I + 1); done", "0\n1\n2\n"},
+		{"set GOSH_I=3; until [ $GOSH_I -eq 0 ]; do echo $GOSH_I >> out; set GOSH_I=$(expr $GOSH_I - 1); done", "3\n2\n1\n"},
+		{"while false; do echo never >> out; done", ""},
+		// The loop variable keeps the last value
+		{"for GOSH_I in a b; do :; done; echo $GOSH_I > out", "b\n"},
+		// break and continue
+		{"for x in 1 2 3 4; do if [ $x = 3 ]; then break; fi; echo $x >> out; done", "1\n2\n"},
+		{"for x in 1 2 3; do if [ $x = 2 ]; then continue; fi; echo $x >> out; done", "1\n3\n"},
+		{"for a in 1 2; do for b in x y; do if [ $b = y ]; then continue 2; fi; echo $a$b >> out; done; done", "1x\n2x\n"},
+		{"for a in 1 2; do for b in x y; do echo $a$b >> out; break 2; done; done", "1x\n"},
+		{"for a in 1 2; do for b in x y; do echo $a$b >> out; break 5; done; done", "1x\n"},
+		{"while true; do echo once >> out; break; done", "once\n"},
+		{"for x in 1 2; do echo $x >> out; break; echo never >> out; done; echo after >> out", "1\nafter\n"},
+	}
+	for _, tc := range tests {
+		if got := runIn(t, tc.input); got != tc.want {
+			t.Errorf("%q: out = %q, want %q", tc.input, got, tc.want)
+		}
+	}
+	if loopDepth != 0 {
+		t.Errorf("loopDepth = %d after the loops", loopDepth)
+	}
+}
+
+func TestExec_Status(t *testing.T) {
+	tests := []struct {
+		input  string
+		status int
+	}{
+		{"if false; then :; fi", 0},
+		{"if true; then false; fi", 1},
+		{"if false; then :; else sh -c 'exit 3'; fi", 3},
+		{"for x in a; do false; done", 1},
+		{"for x in a; do false; done; true", 0},
+		{"for x in; do false; done", 0},
+		{"while false; do :; done", 0},
+		{"for x in a b; do sh -c 'exit 4'; break; done", 0},
+		{"! true", 1},
+		{"! false", 0},
+		{"! gosh-no-such-command", 0},
+		{":", 0},
+		{"break", 1},
+		{"continue", 1},
+		{"for x in a; do break x; done", 1},
+		{"if a; then", 2},
+		{"fi", 2},
+	}
+	for _, tc := range tests {
+		runIn(t, tc.input)
+		if lastStatus != tc.status {
+			t.Errorf("%q: $? = %d, want %d", tc.input, lastStatus, tc.status)
+		}
+	}
+	if err := execInput("break"); err == nil || err.Error() != "break: only meaningful in a loop" {
+		t.Errorf("break outside a loop: %v", err)
+	}
+}
+
+func TestRunScript_MultiLine(t *testing.T) {
+	origDir, _ := os.Getwd()
+	defer os.Chdir(origDir)
+	os.Chdir(t.TempDir())
+
+	script := `# a script with control structures
+for x in a b
+do
+  if [ $x = a ]; then
+    echo first >> out
+  else
+    echo second >> out
+  fi
+done
+echo continued \
+  line >> out
+`
+	if status := runScript(strings.NewReader(script), ""); status != 0 {
+		t.Errorf("status = %d", status)
+	}
+	if got, _ := os.ReadFile("out"); string(got) != "first\nsecond\ncontinued line\n" {
+		t.Errorf("out = %q", got)
+	}
+
+	// A script that ends inside a control structure is a syntax error
+	if status := runScript(strings.NewReader("echo ok >> out2\nif true; then\n  echo never >> out2\n"), ""); status != 2 {
+		t.Errorf("status = %d for an unterminated if, want 2", status)
+	}
+	if got, _ := os.ReadFile("out2"); string(got) != "ok\n" {
+		t.Errorf("out2 = %q", got)
+	}
+}
+
+func TestExec_AliasesInControlStructures(t *testing.T) {
+	withAliases(t, map[string]string{"say": "echo said", "both": "echo one >> out; echo two"})
+	defer func(saved bool) { interactiveShell = saved }(interactiveShell)
+	interactiveShell = true
+
+	if got := runIn(t, "for x in a b; do say $x >> out; done"); got != "said a\nsaid b\n" {
+		t.Errorf("alias in for: out = %q", got)
+	}
+	if got := runIn(t, "if true; then both >> out; fi"); got != "one\ntwo\n" {
+		t.Errorf("alias with a list in if: out = %q", got)
+	}
+	if got := runIn(t, "if say x > out; then :; fi"); got != "said x\n" {
+		t.Errorf("alias in condition: out = %q", got)
+	}
+}

subst.go

@@ -0,0 +1,98 @@
+package main
+
+// Command substitution: $(command) is replaced by the output of command,
+// which runs in a subshell (a new gosh process), so that cd, set etc. in it
+// do not change the shell.
+
+import (
+	"bytes"
+	"errors"
+	"os"
+	"os/exec"
+	"strings"
+	"syscall"
+)
+
+// subshellPath is the gosh binary used for command substitution; empty means
+// the running executable. Tests set it, since there the running executable is
+// the test binary.
+var subshellPath string
+
+// substStatus is the exit status of the last command substitution of the
+// current command (a command that consists only of a substitution has this
+// status, like in sh).
+var substStatus int
+
+// isSubstStart reports whether a command substitution $( starts at s[i].
+func isSubstStart(s string, i int) bool {
+	return s[i] == '$' && i+1 < len(s) && s[i+1] == '('
+}
+
+// substEnd returns the index of the ) that ends the command substitution
+// starting with $( at s[i], or -1 if it is not terminated. Quotes, escapes
+// and nested substitutions inside are taken into account.
+func substEnd(s string, i int) int {
+	depth := 0
+	var quote byte
+	for j := i + 1; j < len(s); j++ {
+		ch := s[j]
+		switch {
+		case quote == '\'':
+			if ch == '\'' {
+				quote = 0
+			}
+		case ch == '\\':
+			j++
+		case quote == '"':
+			if ch == '"' {
+				quote = 0
+			} else if isSubstStart(s, j) {
+				end := substEnd(s, j)
+				if end < 0 {
+					return -1
+				}
+				j = end
+			}
+		case ch == '"' || ch == '\'':
+			quote = ch
+		case ch == '(':
+			depth++
+		case ch == ')':
+			depth--
+			if depth == 0 {
+				return j
+			}
+		}
+	}
+	return -1
+}
+
+// commandSubst runs command in a subshell and returns its output without
+// trailing newlines. stdin and stderr are those of the shell.
+func commandSubst(command string) (string, error) {
+	if strings.HasPrefix(command, "(") {
+		return "", errors.New("arithmetic expansion $((...)) is not supported")
+	}
+	path := subshellPath
+	if path == "" {
+		var err error
+		if path, err = os.Executable(); err != nil {
+			return "", err
+		}
+	}
+
+	var out bytes.Buffer
+	cmd := exec.Command(path, "-c", command)
+	cmd.Stdin, cmd.Stdout, cmd.Stderr = os.Stdin, &out, os.Stderr
+	err := cmd.Run()
+	substStatus = exitStatus(err)
+	if substStatus == 128+int(syscall.SIGINT) {
+		// Ctrl+C aborts the whole command, like in bash
+		return "", statusError(substStatus)
+	}
+	var exitErr *exec.ExitError
+	if err != nil && !errors.As(err, &exitErr) {
+		return "", err
+	}
+	return strings.TrimRight(out.String(), "\n"), nil
+}

subst_test.go

@@ -0,0 +1,104 @@
+package main
+
+import (
+	"os"
+	"testing"
+)
+
+func TestSubstEnd(t *testing.T) {
+	tests := []struct {
+		s    string
+		want int
+	}{
+		{"$(echo a)", 8},
+		{"$(echo a) b", 8},
+		{"$(echo $(date))x", 14},
+		{`$(echo ")")`, 10},
+		{`$(echo ')')`, 10},
+		{`$(echo \))`, 9},
+		{`$(echo "$(echo ")")")`, 20},
+		{"$(echo (a))", 10},
+		{"$(echo a", -1},
+		{`$(echo ")`, -1},
+	}
+	for _, tc := range tests {
+		if got := substEnd(tc.s, 0); got != tc.want {
+			t.Errorf("substEnd(%q) = %d, want %d", tc.s, got, tc.want)
+		}
+	}
+}
+
+func TestExpandAliasesAndBraces_SkipSubstitution(t *testing.T) {
+	withAliases(t, map[string]string{"ll": "ls -l"})
+	if got := expandAliases("echo $(ll; ll) && ll"); got != "echo $(ll; ll) && ls -l" {
+		t.Errorf("expandAliases = %q", got)
+	}
+	got, err := expandBraceWords("echo $(echo {a,b}) {c,d}")
+	if err != nil || got != "echo $(echo {a,b}) c d" {
+		t.Errorf("expandBraceWords = %q, %v", got, err)
+	}
+}
+
+func TestExecInput_CommandSubstitution(t *testing.T) {
+	origDir, _ := os.Getwd()
+	defer os.Chdir(origDir)
+	dir := t.TempDir()
+	os.Chdir(dir)
+	os.WriteFile("a.go", nil, 0644)
+	os.WriteFile("b.go", nil, 0644)
+
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"echo $(echo hello) > out", "hello\n"},
+		// Unquoted, the output is split into words
+		{"printf '[%s]' $(printf 'a b\\nc\\n\\n') > out", "[a][b][c]"},
+		{"printf '[%s]' x$(echo a b)y > out", "[xa][by]"},
+		// Quoted, it is a single word, only trailing newlines are removed
+		{`printf '[%s]' "$(printf 'a  b\nc\n\n')" > out`, "[a  b\nc]"},
+		{`printf '[%s]' "$(true)" > out`, "[]"},
+		{"printf '[%s]' $(true) x > out", "[x]"},
+		// Nested, with quotes and operators inside
+		{"echo $(echo $(echo nested)) > out", "nested\n"},
+		{`echo "$(echo "a;b|c")" > out`, "a;b|c\n"},
+		{"echo $(echo a | tr a-z A-Z) > out", "A\n"},
+		// Wildcards in the output are not expanded, in the command they are
+		{`echo $(echo "*") > out`, "*\n"},
+		{"echo $(echo *.go) > out", "a.go b.go\n"},
+		// Quoted or escaped, $( is literal
+		{`echo '$(echo a)' \$(echo b) > out`, "$(echo a) $(echo b)\n"},
+		// The subshell does not change the shell
+		{"echo $(cd /; pwd) > out; pwd >> out", "/\n" + dir + "\n"},
+		{"set GOSH_SUBST=$(echo value); echo $GOSH_SUBST > out", "value\n"},
+	}
+	for _, tc := range tests {
+		if err := execInput(tc.input); err != nil {
+			t.Errorf("%q: %v", tc.input, err)
+		}
+		if got, _ := os.ReadFile("out"); string(got) != tc.want {
+			t.Errorf("%q: out = %q, want %q", tc.input, got, tc.want)
+		}
+	}
+	os.Unsetenv("GOSH_SUBST")
+
+	// A command that consists only of a substitution has its status
+	execInput("$(sh -c 'exit 3')")
+	if lastStatus != 3 {
+		t.Errorf("$? = %d after $(exit 3), want 3", lastStatus)
+	}
+	// Otherwise the status of the command counts
+	execInput("echo $(false) > out")
+	if lastStatus != 0 {
+		t.Errorf("$? = %d after echo $(false), want 0", lastStatus)
+	}
+
+	for input, wantErr := range map[string]string{
+		"echo $(echo a":   "syntax error: unexpected end of input",
+		"echo $((1 + 2))": "invalid input: arithmetic expansion $((...)) is not supported",
+	} {
+		if err := execInput(input); err == nil || err.Error() != wantErr {
+			t.Errorf("%q: error %v, want %q", input, err, wantErr)
+		}
+	}
+}