fix(csvx): read human null words as absent, unrecognized values as false

BoolishTimestamp defaulted unrecognized non-empty values to true: any
word it does not understand (a misspelling, "disabled", "banana")
behaved as on. Unrecognized values are now valid+false with the error
retained so callers can apply their own policy, and "disabled"/"disable"
are no longer true-like.

Both bool types now read the human "not supplied" set through one shared
isBoolNull sentinel: empty/whitespace-only, the exact PG dump sentinel
\N (case-sensitive), and the case-insensitive words null, \null, none,
unknown. Bool previously reached absent only by ParseBool failing; it
now matches BoolishTimestamp explicitly. Free-text types must not reuse
this set: most of these words are legitimate values there.

database/csvx/bool.go

@@ -34,6 +34,11 @@ func (b *Bool) Scan(value any) error {
 		b.Bool = v
 		b.Valid = true
 	case string:
+		if isBoolNull(v) {
+			b.Valid = false
+			b.Bool = false
+			return nil
+		}
 		parsed, err := strconv.ParseBool(strings.TrimSpace(v))
 		if err != nil {
 			b.Valid = false
@@ -61,7 +66,7 @@ func (b Bool) MarshalCSV() ([]byte, error) {
 
 func (b *Bool) UnmarshalCSV(data []byte) error {
 	s := strings.TrimSpace(string(data))
-	if s == "" {
+	if isBoolNull(s) {
 		b.Valid = false
 		b.Bool = false
 		return nil

database/csvx/boolish.go

@@ -17,19 +17,23 @@ type BoolishTimestamp struct {
 	Error     error
 }
 
-// UnmarshalCSV parses a human value. Unknown non-empty values are retained as
-// errors so the collection mapper can apply its own conservative policy.
+// UnmarshalCSV parses a human value. Recognized true-like values and dates
+// set Bool; recognized false-like values leave it unset. Unrecognized
+// non-empty values are conservatively false with the error retained so the
+// collection mapper can apply its own policy.
 func (v *BoolishTimestamp) UnmarshalCSV(data []byte) error {
 	*v = BoolishTimestamp{}
 	raw := strings.TrimSpace(string(data))
-	if raw == "" || strings.EqualFold(raw, "null") || raw == `\N` {
+	// Null sentinels (isBoolNull): absent means Valid=false. `\N` is the
+	// exact PG dump sentinel and case-sensitive; the word forms are not.
+	if isBoolNull(raw) {
 		return nil
 	}
 	v.Valid = true
 	switch strings.ToLower(raw) {
 	case "false", "f", "n", "no", "0":
 		return nil
-	case "true", "t", "y", "yes", "1", "disabled", "disable":
+	case "true", "t", "y", "yes", "1":
 		v.Bool = true
 		return nil
 	}
@@ -41,7 +45,9 @@ func (v *BoolishTimestamp) UnmarshalCSV(data []byte) error {
 			return nil
 		}
 	}
-	v.Bool = true
+	// Unrecognized: valid but false, error recorded (Bool must not default
+	// to true for values this type does not understand).
+	v.Bool = false
 	v.Error = fmt.Errorf("unrecognized boolish timestamp %q", raw)
 	return nil
 }

database/csvx/csvx_test.go

@@ -204,11 +204,45 @@ func TestBoolishTimestamp(t *testing.T) {
 	}
 
 	// False-like
-	if err := v.UnmarshalCSV([]byte("no")); err != nil {
+	for _, falseish := range []string{"no", "0", "FALSE"} {
+		if err := v.UnmarshalCSV([]byte(falseish)); err != nil {
+			t.Fatalf("UnmarshalCSV(%q): %v", falseish, err)
+		}
+		if !v.Valid || v.Bool || v.Error != nil {
+			t.Fatalf("%q: expected valid+false, got %+v", falseish, v)
+		}
+	}
+
+	// An unrecognized word must not be treated as true: it is valid but
+	// false with an error recorded. The Bool-family null words are sentinels
+	// (covered by TestIsBoolNull), so use words outside that set.
+	for _, word := range []string{"random", "disabled"} {
+		if err := v.UnmarshalCSV([]byte(word)); err != nil {
+			t.Fatalf("UnmarshalCSV(%q): %v", word, err)
+		}
+		if !v.Valid || v.Bool || v.Error == nil {
+			t.Fatalf("%q: expected valid+false+error, got %+v", word, v)
+		}
+	}
+
+	// Empty / null sentinels: `\N` exact; word forms any case. The full
+	// set is covered by TestIsBoolNull.
+	for _, sent := range []string{"", "null", `\N`} {
+		if err := v.UnmarshalCSV([]byte(sent)); err != nil {
+			t.Fatalf("UnmarshalCSV(%q): %v", sent, err)
+		}
+		if v.Valid || v.Bool || v.Error != nil {
+			t.Fatalf("%q: expected absent, got %+v", sent, v)
+		}
+	}
+
+	// `\n` (lowercase) is NOT the PG sentinel: unrecognized, so valid +
+	// false + error.
+	if err := v.UnmarshalCSV([]byte(`\n`)); err != nil {
 		t.Fatalf("UnmarshalCSV: %v", err)
 	}
-	if !v.Valid || v.Bool {
-		t.Fatalf("expected valid+false, got %+v", v)
+	if !v.Valid || v.Bool || v.Error == nil {
+		t.Fatalf(`\n: expected valid+false+error, got %+v`, v)
 	}
 
 	// Empty

database/csvx/null.go

@@ -0,0 +1,22 @@
+package csvx
+
+import "strings"
+
+// isBoolNull reports whether a human-edited cell means "not supplied":
+// empty/whitespace-only, the exact PG dump sentinel `\N` (case-sensitive),
+// or a case-insensitive match of null, \null, none, or unknown.
+//
+// Bool and BoolishTimestamp read such cells as absent. Free-text types
+// must not reuse this set wholesale: most of these words are legitimate
+// text values in non-boolean contexts.
+func isBoolNull(raw string) bool {
+	s := strings.TrimSpace(raw)
+	if s == "" || s == `\N` {
+		return true
+	}
+	switch strings.ToLower(s) {
+	case "null", `\null`, "none", "unknown":
+		return true
+	}
+	return false
+}

database/csvx/null_test.go

@@ -0,0 +1,54 @@
+package csvx
+
+import "testing"
+
+func TestIsBoolNull(t *testing.T) {
+	for _, null := range []string{"", "  ", "null", "NULL", "Null", " null ", `\null`, `\NULL`, `\N`, "none", "None", "unknown", "UNKNOWN"} {
+		if !isBoolNull(null) {
+			t.Errorf("isBoolNull(%q) = false, want true", null)
+		}
+	}
+	// Everything else is a value, not a null.
+	for _, value := range []string{`\n`, `N`, `N\`, "none of the above", "nullable", "no", "off", "na", "disable", "disabled", "NULLITY"} {
+		if isBoolNull(value) {
+			t.Errorf("isBoolNull(%q) = true, want false", value)
+		}
+	}
+}
+
+func TestBoolCSVNullSentinels(t *testing.T) {
+	// Bool must read the same null sentinels as BoolishTimestamp (isBoolNull)
+	// explicitly rather than relying on ParseBool failing: the PG dump
+	// sentinel and a representative word.
+	for _, null := range []string{`\N`, "none"} {
+		var b Bool
+		if err := b.UnmarshalCSV([]byte(null)); err != nil {
+			t.Fatalf("UnmarshalCSV(%q): %v", null, err)
+		}
+		if b.Valid || b.Bool {
+			t.Errorf("UnmarshalCSV(%q) = %+v, want absent", null, b)
+		}
+		var b2 Bool
+		if err := b2.Scan(null); err != nil {
+			t.Fatalf("Scan(%q): %v", null, err)
+		}
+		if b2.Valid || b2.Bool {
+			t.Errorf("Scan(%q) = %+v, want absent", null, b2)
+		}
+	}
+	// `\n` (lowercase) is not the sentinel: it parses as unrecognized.
+	var b Bool
+	if err := b.Scan(`\n`); err != nil {
+		t.Fatalf("Scan: %v", err)
+	}
+	if b.Valid || b.Bool {
+		t.Errorf(`Scan(\n) = %+v, want absent`, b)
+	}
+	// Real values still parse; a non-sentinel word is unrecognized.
+	if err := b.UnmarshalCSV([]byte("true")); err != nil {
+		t.Fatalf("UnmarshalCSV(true): %v", err)
+	}
+	if !b.Valid || !b.Bool {
+		t.Errorf("UnmarshalCSV(true) = %+v", b)
+	}
+}