cvss: implement encoding.TextAppender

This implements the new interface, redefines `encoding.TextMarshaler`
using it, and adds benchmarks.

The goal is to have `AppendText` do no unavoidable allocations, meaning
that if the input slice has sufficient capacity there are no additional
heap allocations to populate it.

Signed-off-by: Hank Donnay <hdonnay@redhat.com>
Change-Id: I50157434eb16073a14edb2b4bec253966a6a6964

Author: hdonnay

toolkit/types/cvss/append_bench_test.go

@@ -0,0 +1,133 @@
+package cvss
+
+import (
+	"math/rand/v2"
+	"testing"
+)
+
+func BenchmarkAppend(b *testing.B) {
+	b.Run("V2", benchAppendV2)
+	b.Run("V3", benchAppendV3)
+	b.Run("V4", benchAppendV4)
+}
+
+func benchAppendV4(b *testing.B) {
+	buf := make([]byte, 0, 1024) // Is it cheating to oversize this?
+	benchOne := func(b *testing.B, vec []byte) {
+		b.Helper()
+		b.Attr("input", string(vec))
+		var v V4
+		if err := v.UnmarshalText(vec); err != nil {
+			b.Fatal(err)
+		}
+		var err error
+		var x []byte
+		b.ReportAllocs()
+
+		for b.Loop() {
+			x, err = v.AppendText(buf)
+			if err != nil {
+				b.Error(err)
+			}
+			_ = x
+		}
+	}
+
+	b.Run("List", func(b *testing.B) {
+		vecs := loadVectorList(b, `v4_roundtrip.list`)
+		todo := make([][]byte, 10)
+		for i := range todo {
+			todo[i] = vecs[rand.N(len(vecs))]
+		}
+		for _, vec := range todo {
+			b.Run("", func(b *testing.B) { benchOne(b, vec) })
+		}
+	})
+	// Each of the following test one fixture plucked from the Spec's examples.
+	b.Run("B", func(b *testing.B) {
+		benchOne(b, []byte("CVSS:4.0/AV:A/AC:H/AT:P/PR:L/UI:P/VC:H/VI:H/VA:H/SC:L/SI:L/SA:L"))
+	})
+	b.Run("BT", func(b *testing.B) {
+		benchOne(b, []byte("CVSS:4.0/AV:A/AC:H/AT:P/PR:L/UI:P/VC:H/VI:H/VA:H/SC:L/SI:L/SA:L/E:P"))
+	})
+	b.Run("BE", func(b *testing.B) {
+		benchOne(b, []byte("CVSS:4.0/AV:L/AC:H/AT:N/PR:N/UI:A/VC:N/VI:N/VA:L/SC:H/SI:H/SA:H/CR:H/IR:H/AR:M/MAV:N/MAC:L/MAT:P/MPR:L/MUI:A/MVC:N/MVI:H/MVA:L/MSC:L/MSI:S/MSA:H"))
+	})
+	b.Run("BTES", func(b *testing.B) {
+		benchOne(b, []byte("CVSS:4.0/AV:N/AC:L/AT:N/PR:H/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N/E:U/CR:L/IR:X/AR:L/MAV:A/MAC:H/MAT:N/MPR:N/MUI:P/MVC:X/MVI:N/MVA:H/MSC:N/MSI:L/MSA:S/S:N/AU:N/R:I/V:C/RE:H/U:Green"))
+	})
+}
+
+func benchAppendV3(b *testing.B) {
+	buf := make([]byte, 0, 1024) // Is it cheating to oversize this?
+	benchOne := func(b *testing.B, vec []byte) {
+		b.Helper()
+		b.Attr("input", string(vec))
+		var v V3
+		if err := v.UnmarshalText(vec); err != nil {
+			b.Fatal(err)
+		}
+		var err error
+		var x []byte
+		b.ReportAllocs()
+
+		for b.Loop() {
+			x, err = v.AppendText(buf)
+			if err != nil {
+				b.Error(err)
+			}
+			_ = x
+		}
+	}
+
+	b.Run("List", func(b *testing.B) {
+		vecs := loadVectorList(b, `v31_score.list`)
+		todo := make([][]byte, 10)
+		for i := range todo {
+			todo[i] = vecs[rand.N(len(vecs))]
+		}
+		for _, vec := range todo {
+			b.Run("", func(b *testing.B) { benchOne(b, vec) })
+		}
+	})
+	b.Run("Heartbleed", func(b *testing.B) {
+		benchOne(b, []byte("CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N"))
+	})
+}
+
+func benchAppendV2(b *testing.B) {
+	buf := make([]byte, 0, 1024) // Is it cheating to oversize this?
+	benchOne := func(b *testing.B, vec []byte) {
+		b.Helper()
+		b.Attr("input", string(vec))
+		var v V2
+		if err := v.UnmarshalText(vec); err != nil {
+			b.Fatal(err)
+		}
+		var err error
+		var x []byte
+		b.ReportAllocs()
+
+		for b.Loop() {
+			x, err = v.AppendText(buf)
+			if err != nil {
+				b.Error(err)
+			}
+			_ = x
+		}
+	}
+
+	b.Run("List", func(b *testing.B) {
+		vecs := loadVectorList(b, `v2_score.list`)
+		todo := make([][]byte, 10)
+		for i := range todo {
+			todo[i] = vecs[rand.N(len(vecs))]
+		}
+		for _, vec := range todo {
+			b.Run("", func(b *testing.B) { benchOne(b, vec) })
+		}
+	})
+	b.Run("Heartbleed", func(b *testing.B) {
+		benchOne(b, []byte("AV:N/AC:L/Au:N/C:P/I:N/A:N"))
+	})
+}

toolkit/types/cvss/cvss.go

@@ -59,10 +59,21 @@ var internalDoc = struct{}{}
 // ErrMalformedVector is reported when a vector is invalid in some way.
 var ErrMalformedVector = errors.New("malformed vector")
 
-// ErrValueUnset is used by [Vector.getString] implementations to signal a
-// metric's value is unset.
+// ErrValueUnset is used by [Vector] implementations to signal a metric's value
+// is unset.
 var errValueUnset = errors.New("unset")
 
+// ErrValueDefault is used by [Vector] implementations to signal a metric's value
+// is unset, but a default value was used for the requested operation.
+var errValueDefault = fmt.Errorf("default: %w", errValueUnset)
+
+// MarshalSize is the initial size of the backing slice for
+// [encoding.TextMarshaler] implementations.
+//
+// This was arrived at by trying sizes until [BenchmarkMarshal] reported a
+// single allocation for all but the longest V4 vectors.
+const marshalSize = 128
+
 // Value is a "packed" representation of the value of a metric.
 //
 // When possible, this is the first byte of the abbreviated form in the relevant
@@ -107,55 +118,56 @@ func Version(vec string) (v int) {
 	return v
 }
 
-// MarshalVector is a generic function to marshal vectors.
-//
-// The [Vector.getString] method is used here.
-func marshalVector[M Metric, V Vector[M]](prefix string, v V) ([]byte, error) {
-	text := append(make([]byte, 0, 64), prefix...) // Guess at an initial capacity.
+// AppendVector is a generic function to marshal vectors via appending to the
+// provided byte slice.
+func appendVector[M Metric, V Vector[M]](b []byte, prefix string, v V) ([]byte, error) {
+	b = append(b, prefix...)
 	var err error
-	// This is a rangefunc-style iterator.
-	v.groups(func(b [2]int) bool {
+	meta := v.meta()
+	g := meta.Groups
+	for s, e := 0, 1; e < len(g); s, e = s+1, e+1 {
 		var set bool
-		orig := len(text)
-		for i := b[0]; i < b[1]; i++ {
+		i, lim := g[s], g[e]
+		skipGroup := len(b)
+		for ; i < lim; i++ {
+			skipMetric := len(b)
 			m := M(i)
-			val, err := v.getString(m)
+
+			b = append(b, '/')
+			b, err = m.AppendText(b)
+			if err != nil {
+				return nil, fmt.Errorf("invalid cvss vector: %w", err)
+			}
+			b = append(b, ':')
+
+			b, err = v.appendValue(b, m)
 			switch {
 			case errors.Is(err, nil):
 				set = true
-			case errors.Is(err, errValueUnset) && val == "":
-				continue
+			case errors.Is(err, errValueDefault):
 			case errors.Is(err, errValueUnset):
+				b = b[:skipMetric]
 			default:
-				err = errors.New("invalid cvss vector")
-				return false
+				return nil, fmt.Errorf("invalid cvss vector: %w", err)
 			}
-
-			text = append(text, '/')
-			text = append(text, m.String()...)
-			text = append(text, ':')
-			text = append(text, val...)
 		}
 		if !set {
-			text = text[:orig]
+			b = b[:skipGroup]
 		}
-		return true
-	})
-	if err != nil {
-		return nil, err
 	}
-	// v2 hack
+	// v2 hack: remove the leading slash.
 	if prefix == "" {
-		text = text[1:]
+		b = b[1:]
 	}
-	return text, nil
+	return b, nil
 }
 
 // Metric is a CVSS metric.
 //
 // The set of types this describes is namespaced per-version.
 type Metric interface {
 	~int
+	encoding.TextAppender
 	fmt.Stringer
 
 	// Valid returns the concatenation of valid values for the metric.
@@ -166,6 +178,7 @@ type Metric interface {
 
 // Vector is a CVSS vector of any version.
 type Vector[M Metric] interface {
+	encoding.TextAppender
 	encoding.TextUnmarshaler
 	encoding.TextMarshaler
 	fmt.Stringer
@@ -181,20 +194,16 @@ type Vector[M Metric] interface {
 	// Environmental reports if the vector contains environmental metrics.
 	Environmental() bool
 
-	// GetString is a hook for returning the stringified version of the metric
-	// value. If the value is unset, implementations should return err
-	// [errValueUnset] rather than a specified default, as defaults are omitted
-	// from the string representation.
-	//
-	// CVSSv2 notably does not use names that are identifiable by a single byte,
-	// so they need to be packed and unpacked.
-	getString(M) (string, error)
+	// AppendValue is a hook for appending the stringified version of the metric
+	// value. If the value is unset, implementations should return the input
+	// slice and err == [errValueUnset] rather than a specified default, as
+	// defaults are omitted from the string representation.
+	appendValue([]byte, M) ([]byte, error)
 	// GetScore returns the "packed" value representation after any default
 	// rules are applied.
 	getScore(M) byte
-	// Groups is a rangefunc-style iterator returning the bounds for groups of metrics.
-	// For a returned value "b", it represents the interval "[b[0], b[1])".
-	groups(func([2]int) bool)
+	// Meta returns the static metadata for this vector.
+	meta() *vectorMetadata
 }
 
 var (
@@ -203,6 +212,14 @@ var (
 	_ Vector[V2Metric] = (*V2)(nil)
 )
 
+// VectorMetadata is static metadata about a vector.
+type vectorMetadata struct {
+	// Groups is a slice of boundaries for the groups of the vector.
+	//
+	// The pairs of ints are [lower, upper).
+	Groups []int
+}
+
 // Qualitative is the "Qualitative Severity" of a Vector.
 type Qualitative int