perf(validation): optimize field overlap check

CHANGELOG.md

@@ -3,6 +3,9 @@
 [Unreleased]
 
 * [BUGFIX] Reject object, interface, and input object type definitions that declare zero fields/input values (spec compliance).
+* [IMPROVEMENT] Optimize overlapping field validation to avoid quadratic memory blowups on large sibling field lists.
+* [FEATURE] Add configurable safety valve for overlapping field comparison count with `OverlapValidationLimit(n)` schema option (0 disables the cap). When exceeded validation aborts early with rule `OverlapValidationLimitExceeded`. Disabled by default.
+* [TEST] Add benchmarks & randomized overlap stress test for mixed field/fragment patterns.
 
 [v1.7.0](https://github.com/graph-gophers/graphql-go/releases/tag/v1.7.0) Release v1.7.0
 

README.md

@@ -153,6 +153,7 @@ schema := graphql.MustParseSchema(sdl, &RootResolver{}, nil)
 - `PanicHandler(panicHandler errors.PanicHandler)` is used to transform panics into errors during query execution. It defaults to `errors.DefaultPanicHandler`.
 - `DisableIntrospection()` disables introspection queries.
 - `DisableFieldSelections()` disables capturing child field selections used by helper APIs (see below).
+- `OverlapValidationLimit(n int)` sets a hard cap on examined overlap pairs during validation; exceeding it emits `OverlapValidationLimitExceeded` error.
 
 ### Field Selection Inspection Helpers
 

graphql.go

@@ -86,6 +86,7 @@ type Schema struct {
 	subscribeResolverTimeout time.Duration
 	useFieldResolvers        bool
 	disableFieldSelections   bool
+	overlapPairLimit         int
 }
 
 // AST returns the abstract syntax tree of the GraphQL schema definition.
@@ -152,6 +153,14 @@ func MaxQueryLength(n int) SchemaOpt {
 	}
 }
 
+// OverlapValidationLimit caps the number of overlapping selection pairs that will be examined
+// during validation of a single operation (including fragments). A value of 0 disables the cap.
+// When the cap is exceeded validation aborts early with an error (rule: OverlapValidationLimitExceeded)
+// to protect against maliciously constructed queries designed to exhaust memory/CPU.
+func OverlapValidationLimit(n int) SchemaOpt {
+	return func(s *Schema) { s.overlapPairLimit = n }
+}
+
 // Tracer is used to trace queries and fields. It defaults to [noop.Tracer].
 func Tracer(t tracer.Tracer) SchemaOpt {
 	return func(s *Schema) {
@@ -247,7 +256,7 @@ func (s *Schema) ValidateWithVariables(queryString string, variables map[string]
 		return []*errors.QueryError{errors.Errorf("executable document must contain at least one operation")}
 	}
 
-	return validation.Validate(s.schema, doc, variables, s.maxDepth)
+	return validation.Validate(s.schema, doc, variables, s.maxDepth, s.overlapPairLimit)
 }
 
 // Exec executes the given query with the schema's resolver. It panics if the schema was created
@@ -270,7 +279,7 @@ func (s *Schema) exec(ctx context.Context, queryString string, operationName str
 	}
 
 	validationFinish := s.validationTracer.TraceValidation(ctx)
-	errs := validation.Validate(s.schema, doc, variables, s.maxDepth)
+	errs := validation.Validate(s.schema, doc, variables, s.maxDepth, s.overlapPairLimit)
 	validationFinish(errs)
 	if len(errs) != 0 {
 		return &Response{Errors: errs}

internal/validation/overlap_fuzz_test.go

@@ -0,0 +1,85 @@
+package validation_test
+
+import (
+	"math/rand"
+	"testing"
+	"time"
+
+	"github.com/graph-gophers/graphql-go/internal/query"
+	"github.com/graph-gophers/graphql-go/internal/schema"
+	v "github.com/graph-gophers/graphql-go/internal/validation"
+)
+
+// FuzzValidateOverlapMixed exercises the overlap validation logic with randomly generated queries
+// containing many sibling fields and fragment spreads to ensure it does not panic or explode in memory.
+// It uses a modest overlap pair cap to keep each iteration bounded.
+func FuzzValidateOverlapMixed(f *testing.F) {
+	baseQueries := []string{
+		"query{root{id}}",
+		"query Q{root{id name}}",
+	}
+	for _, q := range baseQueries {
+		f.Add(q)
+	}
+
+	s := schema.New()
+	_ = schema.Parse(s, `schema{query:Query} type Query{root: Thing} type Thing { id: ID name: String value: String }`, false)
+
+	randSource := rand.New(rand.NewSource(time.Now().UnixNano()))
+
+	f.Fuzz(func(t *testing.T, seed string) {
+		// Use hash of seed to deterministically generate but bound complexity.
+		r := rand.New(rand.NewSource(int64(len(seed)) + randSource.Int63()))
+		fieldCount := 50 + r.Intn(150) // 50-199
+		fragCount := 1 + r.Intn(5)
+
+		// Build fragments.
+		fragBodies := make([]string, fragCount)
+		for i := 0; i < fragCount; i++ {
+			// each fragment gets subset of fields
+			var body string
+			innerFields := 5 + r.Intn(20)
+			for j := 0; j < innerFields; j++ {
+				body += " f" + nameIdx(r.Intn(500)) + ":id"
+			}
+			fragBodies[i] = "fragment F" + nameIdx(i) + " on Thing{" + body + " }"
+		}
+
+		// Root selection
+		sel := "query{root{"
+		for i := 0; i < fieldCount; i++ {
+			sel += " a" + nameIdx(r.Intn(1000)) + ":id"
+		}
+		// Sprinkle fragment spreads
+		for i := 0; i < fragCount; i++ {
+			sel += " ...F" + nameIdx(i)
+		}
+		sel += "}}"
+		queryText := sel
+		for _, fb := range fragBodies {
+			queryText += fb
+		}
+
+		doc, err := query.Parse(queryText)
+		if err != nil {
+			return
+		} // parser fuzzing not our goal
+		if len(doc.Operations) == 0 {
+			return
+		}
+		// Use overlap limit to bound cost.
+		errs := v.Validate(s, doc, nil, 0, 10_000)
+		// Ensure no panic (implicit). Optionally sanity check: errors slice must not be ridiculously huge.
+		if len(errs) > 1000 {
+			t.Fatalf("too many errors: %d", len(errs))
+		}
+	})
+}
+
+func nameIdx(i int) string {
+	const letters = "abcdefghijklmnopqrstuvwxyz"
+	if i < len(letters) {
+		return string(letters[i])
+	}
+	return string(letters[i%len(letters)]) + nameIdx(i/len(letters))
+}

internal/validation/validate_max_depth_test.go

@@ -83,7 +83,7 @@ func (tc maxDepthTestCase) Run(t *testing.T, s *ast.Schema) {
 			t.Fatal(qErr)
 		}
 
-		errs := Validate(s, doc, nil, tc.depth)
+		errs := Validate(s, doc, nil, tc.depth, 0)
 		if len(tc.expectedErrors) > 0 {
 			if len(errs) > 0 {
 				for _, expected := range tc.expectedErrors {
@@ -489,7 +489,7 @@ func TestMaxDepthValidation(t *testing.T) {
 				t.Fatal(err)
 			}
 
-			context := newContext(s, doc, tc.maxDepth)
+			context := newContext(s, doc, tc.maxDepth, 0)
 			op := doc.Operations[0]
 
 			opc := &opContext{context: context, ops: doc.Operations}

internal/validation/validation.go

@@ -26,14 +26,17 @@ type fieldInfo struct {
 }
 
 type context struct {
-	schema           *ast.Schema
-	doc              *ast.ExecutableDefinition
-	errs             []*errors.QueryError
-	opErrs           map[*ast.OperationDefinition][]*errors.QueryError
-	usedVars         map[*ast.OperationDefinition]varSet
-	fieldMap         map[*ast.Field]fieldInfo
-	overlapValidated map[selectionPair]struct{}
-	maxDepth         int
+	schema               *ast.Schema
+	doc                  *ast.ExecutableDefinition
+	errs                 []*errors.QueryError
+	opErrs               map[*ast.OperationDefinition][]*errors.QueryError
+	usedVars             map[*ast.OperationDefinition]varSet
+	fieldMap             map[*ast.Field]fieldInfo
+	overlapValidated     map[selectionPair]struct{}
+	maxDepth             int
+	overlapPairLimit     int
+	overlapPairsObserved int
+	overlapLimitHit      bool
 }
 
 func (c *context) addErr(loc errors.Location, rule string, format string, a ...interface{}) {
@@ -53,7 +56,7 @@ type opContext struct {
 	ops []*ast.OperationDefinition
 }
 
-func newContext(s *ast.Schema, doc *ast.ExecutableDefinition, maxDepth int) *context {
+func newContext(s *ast.Schema, doc *ast.ExecutableDefinition, maxDepth int, overlapPairLimit int) *context {
 	return &context{
 		schema:           s,
 		doc:              doc,
@@ -62,11 +65,12 @@ func newContext(s *ast.Schema, doc *ast.ExecutableDefinition, maxDepth int) *con
 		fieldMap:         make(map[*ast.Field]fieldInfo),
 		overlapValidated: make(map[selectionPair]struct{}),
 		maxDepth:         maxDepth,
+		overlapPairLimit: overlapPairLimit,
 	}
 }
 
-func Validate(s *ast.Schema, doc *ast.ExecutableDefinition, variables map[string]interface{}, maxDepth int) []*errors.QueryError {
-	c := newContext(s, doc, maxDepth)
+func Validate(s *ast.Schema, doc *ast.ExecutableDefinition, variables map[string]interface{}, maxDepth int, overlapPairLimit int) []*errors.QueryError {
+	c := newContext(s, doc, maxDepth, overlapPairLimit)
 
 	opNames := make(nameSet, len(doc.Operations))
 	fragUsedBy := make(map[*ast.FragmentDefinition][]*ast.OperationDefinition)
@@ -303,13 +307,76 @@ func validateMaxDepth(c *opContext, sels []ast.Selection, visited map[*ast.Fragm
 }
 
 func validateSelectionSet(c *opContext, sels []ast.Selection, t ast.NamedType) {
+	if len(sels) == 0 {
+		return
+	}
+
+	// First pass: validate each selection and bucket fields by response name (alias or name).
+	fieldGroups := make(map[string][]ast.Selection)
+	var fragments []ast.Selection // fragment spreads & inline fragments
 	for _, sel := range sels {
+		if c.overlapLimitHit {
+			return
+		}
 		validateSelection(c, sel, t)
+		switch s := sel.(type) {
+		case *ast.Field:
+			name := s.Alias.Name
+			if name == "" {
+				name = s.Name.Name
+			}
+			fieldGroups[name] = append(fieldGroups[name], sel)
+		default:
+			fragments = append(fragments, sel)
+		}
 	}
 
-	for i, a := range sels {
-		for _, b := range sels[i+1:] {
-			c.validateOverlap(a, b, nil, nil)
+	// Compare fields only within same response name group (was O(n^2) across all fields previously).
+	for _, group := range fieldGroups {
+		if c.overlapLimitHit {
+			break
+		}
+		if len(group) < 2 {
+			continue
+		}
+		for i, a := range group {
+			if c.overlapLimitHit {
+				break
+			}
+			for _, b := range group[i+1:] {
+				if c.overlapLimitHit {
+					break
+				}
+				c.validateOverlap(a, b, nil, nil)
+			}
+		}
+	}
+
+	// Fragments can introduce any field names, so we must compare them with all fields and each other.
+	if len(fragments) > 0 && !c.overlapLimitHit {
+		// Flatten fields for fragment comparison.
+		var allFields []ast.Selection
+		for _, group := range fieldGroups {
+			allFields = append(allFields, group...)
+		}
+		for i, fa := range fragments {
+			if c.overlapLimitHit {
+				break
+			}
+			// Compare fragment with all fields
+			for _, fld := range allFields {
+				if c.overlapLimitHit {
+					break
+				}
+				c.validateOverlap(fa, fld, nil, nil)
+			}
+			// Compare fragment with following fragments
+			for _, fb := range fragments[i+1:] {
+				if c.overlapLimitHit {
+					break
+				}
+				c.validateOverlap(fa, fb, nil, nil)
+			}
 		}
 	}
 }
@@ -523,11 +590,38 @@ func (c *context) validateOverlap(a, b ast.Selection, reasons *[]string, locs *[
 		return
 	}
 
-	if _, ok := c.overlapValidated[selectionPair{a, b}]; ok {
+	// Optimisation 1: store only one direction of the pair to halve memory and lookups.
+	pa := reflect.ValueOf(a).Pointer()
+	pb := reflect.ValueOf(b).Pointer()
+	if pb < pa { // canonical ordering
+		a, b = b, a
+	}
+	key := selectionPair{a: a, b: b}
+	if _, ok := c.overlapValidated[key]; ok {
 		return
 	}
-	c.overlapValidated[selectionPair{a, b}] = struct{}{}
-	c.overlapValidated[selectionPair{b, a}] = struct{}{}
+	c.overlapValidated[key] = struct{}{}
+
+	if c.overlapPairLimit > 0 && !c.overlapLimitHit {
+		c.overlapPairsObserved++
+		if c.overlapPairsObserved > c.overlapPairLimit {
+			c.overlapLimitHit = true
+			// determine a representative location for error reporting
+			var loc errors.Location
+			switch sel := a.(type) {
+			case *ast.Field:
+				loc = sel.Alias.Loc
+			case *ast.InlineFragment:
+				loc = sel.Loc
+			case *ast.FragmentSpread:
+				loc = sel.Loc
+			default:
+				// leave zero value
+			}
+			c.addErr(loc, "OverlapValidationLimitExceeded", "Overlapping field validation aborted after examining %d pairs (limit %d). Consider restructuring the query or increasing the limit.", c.overlapPairsObserved-1, c.overlapPairLimit)
+			return
+		}
+	}
 
 	switch a := a.(type) {
 	case *ast.Field:
@@ -608,11 +702,54 @@ func (c *context) validateFieldOverlap(a, b *ast.Field) ([]string, []errors.Loca
 
 	var reasons []string
 	var locs []errors.Location
+
+	// Fast-path: if either side has no subselections we are done.
+	if len(a.SelectionSet) == 0 || len(b.SelectionSet) == 0 {
+		return nil, nil
+	}
+
+	// Optimisation 2: avoid O(m*n) cartesian product for large sibling lists with mostly
+	// distinct response names (common & exploitable for DoS). Instead, index B's field
+	// selections by response name (alias/name). For each field in A we only compare
+	// against fields in B with the same response name plus all fragment spreads / inline
+	// fragments (which can expand to any field names and must be compared exhaustively).
+	bFieldIndex := make(map[string][]ast.Selection, len(b.SelectionSet))
+	var bNonField []ast.Selection
+	for _, bs := range b.SelectionSet {
+		if f, ok := bs.(*ast.Field); ok {
+			name := f.Alias.Name
+			if name == "" { // alias may be empty, fall back to field name
+				name = f.Name.Name
+			}
+			bFieldIndex[name] = append(bFieldIndex[name], bs)
+			continue
+		}
+		bNonField = append(bNonField, bs)
+	}
+
 	for _, a2 := range a.SelectionSet {
+		if af, ok := a2.(*ast.Field); ok {
+			name := af.Alias.Name
+			if name == "" {
+				name = af.Name.Name
+			}
+			// Compare only against same-name fields + all non-field selections.
+			if matches := bFieldIndex[name]; len(matches) != 0 {
+				for _, bMatch := range matches {
+					c.validateOverlap(a2, bMatch, &reasons, &locs)
+				}
+			}
+			for _, bnf := range bNonField {
+				c.validateOverlap(a2, bnf, &reasons, &locs)
+			}
+			continue
+		}
+		// For fragments / inline fragments we still need to compare against every selection in B.
 		for _, b2 := range b.SelectionSet {
 			c.validateOverlap(a2, b2, &reasons, &locs)
 		}
 	}
+
 	return reasons, locs
 }
 

internal/validation/validation_test.go

@@ -88,7 +88,7 @@ func TestValidate(t *testing.T) {
 			if err != nil {
 				t.Fatalf("failed to parse query: %s", err)
 			}
-			errs := validation.Validate(schemas[test.Schema], d, test.Vars, 0)
+			errs := validation.Validate(schemas[test.Schema], d, test.Vars, 0, 0)
 			got := []*errors.QueryError{}
 			for _, err := range errs {
 				if err.Rule == test.Rule {