add assignments name=value and unset

- name=value (one or more) without a command sets variables of the shell
- name=value before a command sets them only in its environment (for
  builtins only while they run)
- an assignment needs an unquoted name and =; the value is not split
  and not expanded as wildcard or brace pattern, ~ is expanded after =
  and :
- the status of an assignment is the one of its last $(...)
- new builtin unset

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

README.md

@@ -12,6 +12,7 @@ Built-in commands:
   quit [code]   - Quit shell with status code (default: $?)
   set VAR=val   - Set environment variable
   export VAR=val - Export environment variable
+  unset VAR     - Remove a variable
   source file   - Run the commands of file in this shell (also: . file)
   jobs          - List background and stopped jobs
   fg [%n]       - Continue job n (default: current job) in the foreground
@@ -44,6 +45,9 @@ Features:
     GOSH_COLORS='user=1;33:path=36' (keys user, root, path, error; values
     are ANSI SGR codes, empty = no color); disabled with $NO_COLOR,
     TERM=dumb or when output is not a terminal
+  - Variables: name=value sets a variable (like set), name=value command
+    sets it only for that command; the value is not split or expanded as
+    a pattern
   - Environment variable expansion ($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
@@ -150,10 +154,10 @@ done
 
 for i in {1..5}; do echo "Durchlauf $i"; done
 
-set n=3
+n=3
 while [ $n -gt 0 ]; do
   echo $n
-  set n=$(expr $n - 1)
+  n=$(expr $n - 1)
 done
 
 until ping -c1 -W1 server >/dev/null; do sleep 1; done
@@ -175,7 +179,28 @@ until ping -c1 -W1 server >/dev/null; do sleep 1; done
 - 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 $n + 1)`.
+- Nicht unterstützt: Kontrollstrukturen in Pipes (`for …; done | sort`), mit Umleitung (`done > datei`) oder im Hintergrund (`done &`), außerdem `case`, Funktionen und `read`. Arithmetik geht z. B. mit `n=$(expr $n + 1)`.
+
+### Variablen
+
+```bash
+name=wert                 # Variable setzen (wie set name=wert)
+a=1 b=2                   # mehrere auf einmal
+gruss="Hallo Welt"        # Leerzeichen im Wert brauchen Anführungszeichen
+heute=$(date +%F)         # Ausgabe eines Befehls
+PATH=~/bin:$PATH          # ~ wird auch nach = und : ersetzt
+LANG=C sort datei         # nur für diesen einen Befehl
+unset name                # Variable entfernen
+echo $name ${name:-standard}
+```
+
+- Eine Zuweisung ist ein Wort `name=wert` am Anfang eines Befehls, bei dem `name` und `=` ohne Anführungszeichen stehen und `name` aus Buchstaben, Ziffern und `_` besteht (nicht mit einer Ziffer beginnend). `"x=1"`, `x\=1` und `echo x=1` sind keine Zuweisungen.
+- Steht danach kein Befehl, wird die Variable in der Shell gesetzt. Wie bei `set` und `export` sind in gosh alle Variablen auch für gestartete Programme sichtbar (es gibt keine reinen Shell-Variablen).
+- Steht ein Befehl danach (`LANG=C sort datei`), gilt die Zuweisung nur für diesen Befehl, die Shell behält den alten Wert.
+- Der Wert wird nicht in Wörter geteilt und nicht als Wildcard oder `{a,b}` expandiert: `x=$(ls)` speichert die ganze Ausgabe, `x=*.go` speichert `*.go`. Variablen, `$(...)` und `~` (am Anfang, nach `=` und nach `:`) werden ersetzt.
+- `$?` ist nach einer Zuweisung der Status des letzten `$(...)` darin, sonst 0.
+- Variablen in den Argumenten werden ersetzt, bevor die Zuweisung davor gilt: `x=1 echo $x` gibt nicht `1` aus, sondern meldet `x: unbound variable` (wenn `x` vorher nicht gesetzt war).
+- In einer Pipeline (`x=1 | cat`) ändert eine Zuweisung die Shell nicht.
 
 ### Befehlsersetzung
 
@@ -273,7 +298,7 @@ $ ls -d */            # nur Verzeichnisse
 - Versteckte Dateien (mit `.` am Anfang) werden nur gefunden, wenn das Muster selbst mit `.` beginnt: `*` findet `.bashrc` nicht, `.*rc` schon.
 - Passt keine Datei, bleibt das Wort unverändert stehen (wie in `sh`), `ls *.xyz` meldet dann also `ls` den Fehler.
 - In Anführungszeichen oder mit `\` davor sind die Zeichen normale Zeichen: `echo "*"`, `echo '*.go'`, `echo \*`.
-- Werte von Variablen werden nicht als Muster behandelt (anders als in `sh`): Nach `set P=*.go` gibt `echo $P` wörtlich `*.go` aus. So kann ein Variablenwert nie unbemerkt auf viele Dateien passen. `echo $DIR/*.go` funktioniert dagegen, weil das `*` direkt in der Zeile steht.
+- Werte von Variablen werden nicht als Muster behandelt (anders als in `sh`): Nach `P='*.go'` gibt `echo $P` wörtlich `*.go` aus. So kann ein Variablenwert nie unbemerkt auf viele Dateien passen. `echo $DIR/*.go` funktioniert dagegen, weil das `*` direkt in der Zeile steht.
 - Bei einer Umleitung muss das Muster genau eine Datei treffen (`> *.log`), sonst meldet gosh `ambiguous redirect`.
 
 ### Pipes und Umleitungen

assign_test.go

@@ -0,0 +1,109 @@
+package main
+
+import (
+	"os"
+	"testing"
+)
+
+func TestTokenize_Assignments(t *testing.T) {
+	t.Setenv("HOME", "/home/u")
+	tests := []struct {
+		input  string
+		word   string
+		assign bool
+	}{
+		{"x=1", "x=1", true},
+		{"_a1=", "_a1=", true},
+		{`x="a b"`, "x=a b", true},
+		{"x=a=b", "x=a=b", true},
+		{"P=~/bin:~/go/bin", "P=/home/u/bin:/home/u/go/bin", true},
+		{"x=a~b", "x=a~b", true},
+		{"x=~", "x=/home/u", true},
+		// Not an assignment: quoted or escaped name or =, invalid name
+		{`"x=1"`, "x=1", false},
+		{`x"=1"`, "x=1", false},
+		{`x\=1`, "x=1", false},
+		{"1x=2", "1x=2", false},
+		{"=x", "=x", false},
+		{"a-b=1", "a-b=1", false},
+	}
+	for _, tc := range tests {
+		tokens, err := tokenize(tc.input)
+		if err != nil || len(tokens) != 1 {
+			t.Errorf("tokenize(%q) = %+v, %v", tc.input, tokens, err)
+			continue
+		}
+		if tokens[0].word != tc.word || tokens[0].assign != tc.assign {
+			t.Errorf("tokenize(%q) = %q (assign %v), want %q (assign %v)",
+				tc.input, tokens[0].word, tokens[0].assign, tc.word, tc.assign)
+		}
+	}
+
+	// The output of $(...) is not split in an assignment
+	tokens, _ := tokenize("x=$(echo a b) y")
+	if len(tokens) != 2 || tokens[0].word != "x=a b" || !tokens[0].assign {
+		t.Errorf("x=$(echo a b): %+v", tokens)
+	}
+}
+
+func TestExec_Assignments(t *testing.T) {
+	for _, name := range []string{"GOSH_A", "GOSH_B", "GOSH_C"} {
+		os.Unsetenv(name)
+		defer os.Unsetenv(name)
+	}
+	tests := []struct {
+		input string
+		want  string
+	}{
+		{"GOSH_A=1; echo $GOSH_A > out", "1\n"},
+		{"GOSH_A=1 GOSH_B=2; echo $GOSH_A$GOSH_B > out", "12\n"},
+		{`GOSH_A="a  b"; echo "$GOSH_A" > out`, "a  b\n"},
+		{"GOSH_A=$(printf 'a b\\nc'); echo \"$GOSH_A\" > out", "a b\nc\n"},
+		// No wildcard or brace expansion in the value
+		{"touch x.go; GOSH_A=*.go; echo \"$GOSH_A\" > out", "*.go\n"},
+		{"GOSH_A={a,b}; echo $GOSH_A > out", "{a,b}\n"},
+		// ... but as an argument, assignment-like words are normal words
+		{"echo GOSH_A={a,b} > out", "GOSH_A=a GOSH_A=b\n"},
+		// Before a command, only in its environment
+		{"GOSH_C=temp sh -c 'echo $GOSH_C' > out; echo ${GOSH_C:-unset} >> out", "temp\nunset\n"},
+		{"GOSH_A=old; GOSH_A=new sh -c 'echo $GOSH_A' > out; echo $GOSH_A >> out", "new\nold\n"},
+		// ... and before a builtin only while it runs
+		{"GOSH_C=temp pwd > /dev/null; echo ${GOSH_C:-unset} > out", "unset\n"},
+		// In a pipeline or with a command they do not change the shell
+		{"GOSH_C=x | cat; echo ${GOSH_C:-unset} > out", "unset\n"},
+		// Redirections are applied
+		{"GOSH_A=r > out", ""},
+		// In loops
+		{"for i in 1 2 3; do GOSH_A=$i; done; echo $GOSH_A > out", "3\n"},
+		{"GOSH_B=0; while [ $GOSH_B -lt 3 ]; do echo $GOSH_B >> out; GOSH_B=$(expr $GOSH_B + 1); done", "0\n1\n2\n"},
+		// Not assignments
+		{`echo "GOSH_A=1" GOSH_B=2 > out`, "GOSH_A=1 GOSH_B=2\n"},
+		// unset
+		{"GOSH_A=1; unset GOSH_A; echo ${GOSH_A:-gone} > out", "gone\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)
+		}
+	}
+
+	// The status of an assignment is the one of its last $(...)
+	for input, want := range map[string]int{
+		"GOSH_A=1":                       0,
+		"GOSH_A=$(false)":                1,
+		"GOSH_A=$(sh -c 'exit 3')":       3,
+		"GOSH_A=$(false) GOSH_B=$(true)": 0,
+	} {
+		runIn(t, input)
+		if lastStatus != want {
+			t.Errorf("%q: $? = %d, want %d", input, lastStatus, want)
+		}
+	}
+
+	if err := execInput("unset 1x"); err == nil || err.Error() != "unset: `1x': not a valid identifier" {
+		t.Errorf("unset 1x: %v", err)
+	}
+	if err := execInput(`"GOSH_A=1"`); err == nil || exitStatus(err) != 127 {
+		t.Errorf(`"GOSH_A=1" must be a command: %v`, err)
+	}
+}

brace.go

@@ -22,6 +22,7 @@ func expandBraceWords(input string) (string, error) {
 	var (
 		out           strings.Builder
 		afterRedirect bool // the next word is the target of a redirection
+		commandStart  = true
 	)
 
 	for i := 0; i < len(input); {
@@ -33,6 +34,8 @@ func expandBraceWords(input string) (string, error) {
 		case strings.IndexByte(";&|<>", ch) >= 0:
 			if ch == '<' || ch == '>' {
 				afterRedirect = true
+			} else {
+				commandStart = true
 			}
 			out.WriteByte(ch)
 			i++
@@ -42,6 +45,16 @@ func expandBraceWords(input string) (string, error) {
 			return out.String(), nil
 		default:
 			end := wordEnd(input, i)
+			if commandStart && !afterRedirect && isAssignment(input[i:end]) {
+				// name=value before the command: no brace expansion, like
+				// in bash
+				out.WriteString(input[i:end])
+				i = end
+				continue
+			}
+			if !afterRedirect {
+				commandStart = false
+			}
 			words, err := expandBraces(input[i:end])
 			if err != nil {
 				return "", err
@@ -57,6 +70,12 @@ func expandBraceWords(input string) (string, error) {
 	return out.String(), nil
 }
 
+// isAssignment reports whether a word starts with name= (unquoted).
+func isAssignment(word string) bool {
+	name, _, ok := strings.Cut(word, "=")
+	return ok && isVarName(name)
+}
+
 // wordEnd returns the end of the word starting at i: the next unquoted blank
 // or operator character.
 func wordEnd(s string, i int) int {

main.go

@@ -121,6 +121,7 @@ type token struct {
 	word    string // the word, if op == ""
 	pattern string // the word as glob pattern (quoted characters escaped)
 	glob    bool   // the word contains unquoted * ? or [ and is expanded
+	assign  bool   // the word is an assignment name=value (unquoted name and =)
 	fd      int    // the redirected file descriptor (for redirections)
 	dup     int    // the target file descriptor of >& and <&
 }
@@ -136,7 +137,8 @@ type token struct {
 //   - $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
-//   - ~ and ~/... are expanded at the start of an unquoted word
+//   - ~ and ~/... are expanded at the start of an unquoted word, and after
+//     the = and : of an assignment (PATH=~/bin:~/go/bin)
 //   - unquoted |, <, >, >>, N<, N>, N>>, N>&M, N<&M, &> and &>> are operators
 //
 // Expanded values are never split into several words.
@@ -147,6 +149,9 @@ func tokenize(input string) ([]token, error) {
 		pat      strings.Builder // the word as glob pattern, quoted parts escaped
 		glob     bool            // the word contains unquoted * ? or [
 		inWord   bool            // a word has been started (even if it is still empty)
+		plain    = true          // the word so far has no quotes, escapes or expansions
+		assign   bool            // the word is an assignment name=value
+		tildeOK  bool            // ~ may be expanded here (after = or : of an assignment)
 		quote    byte
 		escaped  bool
 		rawStart = -1 // start of the current word in input
@@ -154,16 +159,17 @@ func tokenize(input string) ([]token, error) {
 
 	endWord := func() {
 		if inWord {
-			tokens = append(tokens, token{word: b.String(), pattern: pat.String(), glob: glob})
+			tokens = append(tokens, token{word: b.String(), pattern: pat.String(), glob: glob, assign: assign})
 		}
 		b.Reset()
 		pat.Reset()
-		glob = false
+		glob, assign, plain = false, false, true
 		inWord = false
 		rawStart = -1
 	}
 	// literal adds quoted, escaped or expanded text: it is never a pattern
 	literal := func(s string) {
+		plain = false
 		b.WriteString(s)
 		pat.WriteString(escapeGlob(s))
 	}
@@ -175,6 +181,8 @@ func tokenize(input string) ([]token, error) {
 
 	for i := 0; i < len(input); i++ {
 		ch := input[i]
+		afterAssignSep := tildeOK
+		tildeOK = false
 		switch {
 		case escaped:
 			if quote == '"' && !strings.ContainsRune("$\"\\`", rune(ch)) {
@@ -202,7 +210,8 @@ func tokenize(input string) ([]token, error) {
 			if err != nil {
 				return nil, err
 			}
-			if quote == '"' {
+			if quote == '"' || assign {
+				// Quoted or in an assignment, the output is one word
 				literal(out)
 				inWord = true
 			} else {
@@ -241,6 +250,7 @@ func tokenize(input string) ([]token, error) {
 			}
 		case ch == '"' || ch == '\'':
 			startRaw(i)
+			plain = false
 			quote = ch
 			inWord = true
 		case ch == ' ' || ch == '\t' || ch == '\n':
@@ -296,8 +306,8 @@ func tokenize(input string) ([]token, error) {
 				i = j - 1
 			}
 			tokens = append(tokens, t)
-		case ch == '~' && !inWord && b.Len() == 0 &&
-			(i+1 == len(input) || strings.IndexByte("/ \t\n|<>", input[i+1]) >= 0):
+		case ch == '~' && (!inWord && b.Len() == 0 || afterAssignSep) &&
+			(i+1 == len(input) || strings.IndexByte("/ \t\n|<>:", input[i+1]) >= 0):
 			startRaw(i)
 			home := getHomeDir()
 			if home == "" {
@@ -307,6 +317,14 @@ func tokenize(input string) ([]token, error) {
 			inWord = true
 		default:
 			startRaw(i)
+			switch {
+			case ch == '=' && plain && !assign && isVarName(b.String()):
+				// name=value; ~ may follow
+				assign, tildeOK = true, true
+			case ch == ':' && assign:
+				// PATH=~/bin:~/go/bin
+				tildeOK = true
+			}
 			b.WriteByte(ch)
 			switch {
 			case ch == '*' || ch == '?' || ch == '[':
@@ -742,7 +760,7 @@ type shellCompleter struct {
 
 func newShellCompleter() *shellCompleter {
 	return &shellCompleter{
-		commands: []string{"cd", "pwd", "echo", "exit", "quit", "set", "export", "source", "jobs", "fg", "bg", "wait", "alias", "unalias", "break", "continue", "help"},
+		commands: []string{"cd", "pwd", "echo", "exit", "quit", "set", "export", "source", "jobs", "fg", "bg", "wait", "alias", "unalias", "break", "continue", "unset", "help"},
 	}
 }
 
@@ -1391,6 +1409,7 @@ type redirect struct {
 
 // stage is a simple command of a pipeline.
 type stage struct {
+	assigns   []string // name=value before the command name
 	args      []string
 	redirects []redirect
 }
@@ -1401,13 +1420,18 @@ func parsePipeline(tokens []token) ([]stage, error) {
 		stages []stage
 		cur    stage
 	)
-	isEmpty := func(s stage) bool { return len(s.args) == 0 && len(s.redirects) == 0 }
+	isEmpty := func(s stage) bool { return len(s.assigns) == 0 && len(s.args) == 0 && len(s.redirects) == 0 }
 
 	for i := 0; i < len(tokens); i++ {
 		t := tokens[i]
 		switch t.op {
 		case "":
+			if t.assign && len(cur.args) == 0 {
+				// The value is not split and not expanded as a pattern
+				cur.assigns = append(cur.assigns, t.word)
+			} else {
 				cur.args = append(cur.args, expandWord(t)...)
+			}
 		case "|":
 			if isEmpty(cur) {
 				return nil, syntaxError("|")
@@ -1526,6 +1550,20 @@ func startStage(st stage, stdin, stdout *os.File, j *job, inPipeline, background
 	fds := [3]*os.File{stdin, stdout, os.Stderr}
 	opened, err := applyRedirects(&fds, st.redirects)
 	owned = append(owned, opened...)
+	if err == nil && len(st.args) == 0 && len(st.assigns) > 0 {
+		// Assignments without a command set variables of the shell (not in
+		// a pipeline or in the background, where sh runs them in a
+		// subshell). The status is the one of the last $(...) in them.
+		if !inPipeline && !background {
+			for _, a := range st.assigns {
+				name, value, _ := strings.Cut(a, "=")
+				os.Setenv(name, value)
+			}
+		}
+		if substStatus != 0 {
+			err = statusError(substStatus)
+		}
+	}
 	if err != nil || len(st.args) == 0 {
 		// A command without arguments only creates or truncates the files
 		closeOwned()
@@ -1552,6 +1590,8 @@ func startStage(st stage, stdin, stdout *os.File, j *job, inPipeline, background
 			return err
 		}
 		if !inPipeline && !background {
+			// name=value before a builtin applies only while it runs
+			defer setTemporarily(st.assigns)()
 			return finishedProc(run())
 		}
 		p := &proc{done: make(chan error, 1)}
@@ -1561,6 +1601,11 @@ func startStage(st stage, stdin, stdout *os.File, j *job, inPipeline, background
 
 	cmd := exec.Command(st.args[0], st.args[1:]...)
 	cmd.Stdin, cmd.Stdout, cmd.Stderr = fds[0], fds[1], fds[2]
+	if len(st.assigns) > 0 {
+		// name=value before a command is only in its environment (the
+		// last value of a name wins)
+		cmd.Env = append(os.Environ(), st.assigns...)
+	}
 	if jobControl {
 		// All processes of the job share a process group: the one of the
 		// first process. In the foreground, the first process makes its
@@ -1593,6 +1638,31 @@ func startStage(st stage, stdin, stdout *os.File, j *job, inPipeline, background
 	return &proc{cmd: cmd}
 }
 
+// setTemporarily sets the variables of the assignments and returns a
+// function that restores the previous values.
+func setTemporarily(assigns []string) func() {
+	type saved struct {
+		name, value string
+		ok          bool
+	}
+	var restore []saved
+	for _, a := range assigns {
+		name, value, _ := strings.Cut(a, "=")
+		old, ok := os.LookupEnv(name)
+		restore = append(restore, saved{name, old, ok})
+		os.Setenv(name, value)
+	}
+	return func() {
+		for i := len(restore) - 1; i >= 0; i-- {
+			if r := restore[i]; r.ok {
+				os.Setenv(r.name, r.value)
+			} else {
+				os.Unsetenv(r.name)
+			}
+		}
+	}
+}
+
 // applyRedirects applies the redirections in order to fds (stdin, stdout,
 // stderr) and returns the files it opened.
 func applyRedirects(fds *[3]*os.File, redirects []redirect) ([]*os.File, error) {
@@ -1661,6 +1731,7 @@ var builtins = map[string]struct{ changesState bool }{
 	"alias":    {},
 	"unalias":  {changesState: true},
 	":":        {},
+	"unset":    {changesState: true},
 	"break":    {changesState: true},
 	"continue": {changesState: true},
 }
@@ -1745,6 +1816,20 @@ func runBuiltin(args []string, stdout io.Writer) error {
 	case ":":
 		// Does nothing, successfully (e.g. while :; do ...)
 		return nil
+	case "unset":
+		// unset NAME ... - remove variables
+		if len(args) < 2 {
+			return fmt.Errorf("usage: unset NAME ...")
+		}
+		var err error
+		for _, name := range args[1:] {
+			if !isVarName(name) {
+				err = fmt.Errorf("unset: `%s': not a valid identifier", sanitize(name))
+				continue
+			}
+			os.Unsetenv(name)
+		}
+		return err
 	case "break", "continue":
 		return builtinLoopControl(args)
 	case "alias":
@@ -1776,6 +1861,7 @@ Built-in commands:
   quit [code]   - Quit shell with status code (default: $?)
   set VAR=val   - Set environment variable
   export VAR=val - Export environment variable
+  unset VAR     - Remove a variable
   source file   - Run the commands of file in this shell (also: . file)
   jobs          - List background and stopped jobs
   fg [%n]       - Continue job n (default: current job) in the foreground
@@ -1808,6 +1894,9 @@ Features:
     GOSH_COLORS='user=1;33:path=36' (keys user, root, path, error; values
     are ANSI SGR codes, empty = no color); disabled with $NO_COLOR,
     TERM=dumb or when output is not a terminal
+  - Variables: name=value sets a variable (like set), name=value command
+    sets it only for that command; the value is not split or expanded as
+    a pattern
   - Environment variable expansion ($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

main_test.go

@@ -1674,7 +1674,7 @@ func TestTokenize_Operators(t *testing.T) {
 		}
 		// Patterns are tested in TestTokenize_Patterns
 		for i := range got {
-			got[i].pattern, got[i].glob = "", false
+			got[i].pattern, got[i].glob, got[i].assign = "", false, false
 		}
 		if !reflect.DeepEqual(got, tc.want) {
 			t.Errorf("tokenize(%q) = %+v, want %+v", tc.input, got, tc.want)