Draw a dotted token's head once per expansion

2026-08-28 15:03:23 +02:00
parent 5263c866c4
commit 12100239f5
4 changed files with 158 additions and 44 deletions
@@ -196,35 +196,27 @@ func TestPathThroughChoice(t *testing.T) {
 	}
 }

-// TestPathKeyIsUnambiguous pins the one shape that can collide: a field literally
-// named "a.b" in one variant and a field "a" holding "b" in another both spell the
-// dot path "a.b". Neither variant may inherit the other's, so the path must fail
-// the same way every call and must not be advertised.
+// TestPathKeyIsUnambiguous pins that the one shape which could collide cannot be
+// written: a field literally named "a.b" and a field "a" holding "b" would both
+// spell "a.b", so a dotted field name is rejected at New and a dot means a path
+// wherever it appears.
 func TestPathKeyIsUnambiguous(t *testing.T) {
 	dir := writeData(t, map[string]string{
 		"cat": `[{"format":"{a.b}","a.b":["1"]},{"format":"{a}","a":{"format":"{b}","b":["2"]}}]`,
 	})
-	f := newFakes(t, dir, WithSeed(1))
-	if slices.Contains(f.List(), "cat.a.b") {
-		t.Errorf("List() advertises cat.a.b, which only one variant carries")
+	_, err := New([]string{dir})
+	if err == nil || !strings.Contains(err.Error(), `field "a.b" contains a dot`) {
+		t.Fatalf("New = %v, want the dotted field name rejected", err)
 	}
-	var first string
-	for i := 0; i < 200; i++ {
-		_, err := f.Fake("cat.a.b")
-		if err == nil {
-			t.Fatal("Fake(cat.a.b) = nil error, want the same failure every call")
-		}
-		if i == 0 {
-			first = err.Error()
-		} else if err.Error() != first {
-			t.Fatalf("cat.a.b error varies:\n  %s\n  %s", first, err.Error())
-		}
+	// The same data without the dotted key is fine, and the path resolves.
+	f := newFakes(t, writeData(t, map[string]string{
+		"cat": `{"format":"{a.b}","a":{"format":"{b}","b":["2"]}}`,
+	}), WithSeed(1))
+	if !slices.Contains(f.List(), "cat.a.b") {
+		t.Error("List() omits cat.a.b, which the data carries")
 	}
-	// Each variant still renders through its own token.
-	for i := 0; i < 50; i++ {
-		if got := fake(t, f, "cat"); got != "1" && got != "2" {
-			t.Fatalf("cat = %q, want 1 or 2", got)
-		}
+	if got := fake(t, f, "cat"); got != "2" {
+		t.Fatalf("cat = %q, want 2", got)
 	}
 }

@@ -46,6 +46,10 @@ type template struct {
 	separator string
 	ops       []op // format compiled once (see compileOps); what expand walks
 	grow      int  // minimum output size, to size the render buffer
+	// bound is the fields the format addresses by dotted path, each drawn once per
+	// expansion so two tokens read one row (see boundHeads and expand). nil when
+	// the format takes no path.
+	bound map[string]bool
 }

 func (*template) isNode() {}
@@ -144,6 +148,9 @@ func compileTemplate(m map[string]any) (node, error) {
 		if isRef(k) {
 			return nil, fmt.Errorf("field %q starts with %q, which is reserved for {..path} bindings", k, refPrefix)
 		}
+		if strings.Contains(k, ".") {
+			return nil, fmt.Errorf("field %q contains a dot, which separates the segments of a path into a field", k)
+		}
 		n, err := compile(m[k])
 		if err != nil {
 			return nil, fmt.Errorf("field %q: %w", k, err)
@@ -153,10 +160,40 @@ func compileTemplate(m map[string]any) (node, error) {
 	if err := checkTokens(format, t.fields); err != nil {
 		return nil, err
 	}
-	t.ops, t.grow = compileOps(format)
+	t.ops, t.grow, t.bound = compileOps(format)
 	return t, nil
 }

+// checkPath is the static twin of descend: it reports whether a dotted tail can
+// address a node whichever way the draw goes. A multi-variant choice must carry the
+// whole remaining path in the set every variant shares, so a path that validates
+// here resolves on every render, and a typo is a New-time error.
+func checkPath(n node, tail []string) error {
+	if len(tail) == 0 {
+		return nil
+	}
+	switch n := n.(type) {
+	case *template:
+		child, ok := n.fields[tail[0]]
+		if !ok {
+			return fmt.Errorf("no field %q", tail[0])
+		}
+		return checkPath(child, tail[1:])
+	case *choice:
+		if len(n.items) > 1 {
+			if want := strings.Join(tail, "."); !n.shared[want] {
+				return unreachableInChoice(n, want)
+			}
+			return nil // every variant carries it, so any draw resolves
+		}
+		return checkPath(n.items[0], tail)
+	case literal:
+		return fmt.Errorf("a plain string has no field %q", tail[0])
+	default:
+		return fmt.Errorf("cannot descend into %T at %q", n, tail[0])
+	}
+}
+
 // repeatOf reads a template's "repeat" (default 1): how many times its format
 // is rendered and concatenated. A present one must be a positive integer.
 func repeatOf(m map[string]any) (int, error) {
@@ -137,6 +137,12 @@ func classChar(s *session, c rune) byte {
 func expand(s *session, t *template) string {
 	var b strings.Builder
 	b.Grow(t.grow)
+	// One draw per bound head, held for this expansion only: a nested template and
+	// each repeat iteration get their own, since each is its own expansion.
+	var bound map[string]node
+	if len(t.bound) > 0 {
+		bound = make(map[string]node, len(t.bound))
+	}
 	for i := range t.ops {
 		o := &t.ops[i]
 		switch o.kind {
@@ -145,7 +151,7 @@ func expand(s *session, t *template) string {
 		case 'c':
 			b.WriteByte(classChar(s, o.r))
 		case 'f':
-			b.WriteString(resolve(s, o.names, t.fields))
+			b.WriteString(resolve(s, o.arms, t, bound))
 		case 'b':
 			b.WriteString(o.call(s, b.String(), t.fields)) // b.String() is the output so far
 		}
@@ -153,9 +159,38 @@ func expand(s *session, t *template) string {
 	return b.String()
 }

-// resolve renders one field alternation: the '|' arms, one picked at random. A
-// name is a sibling field or a {..path} reference, which linkRefs bound into
-// fields too; checkTokens and linkRefs guarantee both exist.
-func resolve(s *session, names []string, fields map[string]node) string {
-	return render(s, fields[names[s.IntN(len(names))]])
+// resolve renders one field alternation: the '|' alternatives, one picked at
+// random. An arm's key is a sibling field or a {..path} reference, which linkRefs
+// bound into fields too. A head the format addresses by dotted path is drawn once
+// and held in bound, so {place.postal-code} and {place.locality} read one row;
+// the arm's tail then walks into that draw. checkTokens, checkPath and linkRefs
+// prove every step, so this cannot fail.
+func resolve(s *session, arms []arm, t *template, bound map[string]node) string {
+	a := arms[s.IntN(len(arms))]
+	n := t.fields[a.key]
+	if t.bound[a.key] {
+		held, drew := bound[a.key]
+		if !drew {
+			held = drawn(s, n)
+			bound[a.key] = held
+		}
+		n = held
+	}
+	if len(a.tail) > 0 {
+		var err error
+		if n, err = descend(s, n, a.tail); err != nil {
+			panic(fmt.Sprintf("fakes: %s: %v", strings.Join(append([]string{a.key}, a.tail...), "."), err))
+		}
+	}
+	return render(s, n)
+}
+
+// drawn resolves a choice to one variant, so a bound head is a concrete node the
+// rest of the expansion shares. Nested choices unwrap too: a draw is one value, not
+// another set to pick from.
+func drawn(s *session, n node) node {
+	for c, ok := n.(*choice); ok; c, ok = n.(*choice) {
+		n = pick(s, c)
+	}
+	return n
 }
@@ -142,11 +142,16 @@ func checkTokens(format string, fields map[string]node) error {
 				}
 				continue // a root reference; its target is checked at New (see linkRefs)
 			}
-			if _, ok := fields[name]; !ok {
-				if isOption(name) {
-					return fmt.Errorf("token {%s}: %q is an option and can never be a field", t.body, name)
+			a := splitArm(name)
+			head, ok := fields[a.key]
+			if !ok {
+				if isOption(a.key) {
+					return fmt.Errorf("token {%s}: %q is an option and can never be a field", t.body, a.key)
 				}
-				return fmt.Errorf("token {%s}: no field %q", t.body, name)
+				return fmt.Errorf("token {%s}: no field %q", t.body, a.key)
+			}
+			if err := checkPath(head, a.tail); err != nil {
+				return fmt.Errorf("token {%s}: field %q: %w", t.body, a.key, err)
 			}
 		}
 		return nil
@@ -167,6 +172,38 @@ func fieldTokens(format string) []string {
 	return names
 }

+// arm is one alternative of a {a|b} token, split into the key naming the node in
+// a template's fields (a sibling field, or the whole "..path" string a reference is
+// bound under) and the tail of a dotted path into it. A non-empty tail is what makes
+// the arm a bound draw: its head is drawn once per expansion (see boundHeads).
+type arm struct {
+	key  string
+	tail []string
+}
+
+// splitArm splits one token alternative into key and tail. A reference keeps its
+// dots — linkRefs binds it whole — so only a sibling name reads as a path.
+func splitArm(name string) arm {
+	if isRef(name) {
+		return arm{key: name}
+	}
+	head, tail, dotted := strings.Cut(name, ".")
+	if !dotted {
+		return arm{key: name}
+	}
+	return arm{key: head, tail: strings.Split(tail, ".")}
+}
+
+// splitArms splits a token body's '|' alternatives.
+func splitArms(body string) []arm {
+	parts := strings.Split(body, "|")
+	arms := make([]arm, len(parts))
+	for i, p := range parts {
+		arms[i] = splitArm(p)
+	}
+	return arms
+}
+
 // callFn is a builtin bound to one call site: its args already parsed.
 type callFn func(s *session, emitted string, fields map[string]node) string

@@ -174,19 +211,23 @@ type callFn func(s *session, emitted string, fields map[string]node) string
 // alternation, or a builtin already bound to its args. compile builds these so
 // render never re-scans the format.
 type op struct {
-	kind  byte     // 'l' literal run, 'c' class char, 'f' field alternation, 'b' builtin
-	lit   string   // kind 'l'
-	r     rune     // kind 'c'
-	names []string // kind 'f': the '|' arms, split once
-	call  callFn
+	kind byte   // 'l' literal run, 'c' class char, 'f' field alternation, 'b' builtin
+	lit  string // kind 'l'
+	r    rune   // kind 'c'
+	arms []arm  // kind 'f': the '|' alternatives, split into key and path once
+	call callFn
 }

 // compileOps turns a format string into ops, and returns the smallest output it can
-// produce (literals plus one byte per class char) to size the render buffer. Call
-// checkTokens first: it is what proves the scan and every token are valid.
-func compileOps(format string) ([]op, int) {
+// produce (literals plus one byte per class char) to size the render buffer, plus
+// the fields the format addresses by dotted path — each drawn once per expansion,
+// so every token reading one sees the same row (see expand). bound is nil when the
+// format takes no path, so data that uses none carries no render-time cost.
+// Call checkTokens first: it is what proves the scan and every token are valid.
+func compileOps(format string) ([]op, int, map[string]bool) {
 	var ops []op
 	var lit strings.Builder
+	var bound map[string]bool
 	grow := 0
 	flush := func() {
 		if lit.Len() > 0 {
@@ -208,13 +249,22 @@ func compileOps(format string) ([]op, int) {
 			if name, args, ok := funcCall(t.body); ok {
 				ops = append(ops, op{kind: 'b', call: builtins[name].bind(args)})
 			} else {
-				ops = append(ops, op{kind: 'f', names: strings.Split(t.body, "|")})
+				arms := splitArms(t.body)
+				for _, a := range arms {
+					if len(a.tail) > 0 {
+						if bound == nil {
+							bound = map[string]bool{}
+						}
+						bound[a.key] = true
+					}
+				}
+				ops = append(ops, op{kind: 'f', arms: arms})
 			}
 		}
 		return nil
 	})
 	flush()
-	return ops, grow
+	return ops, grow, bound
 }

 // bind returns the closure for one call site, parsing args once via prep if present.