errors.go

     1  // Copyright 2020 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  //go:build goexperiment.jsonv2
     6  
     7  package json
     8  
     9  import (
    10  	"cmp"
    11  	"errors"
    12  	"fmt"
    13  	"io"
    14  	"reflect"
    15  	"strconv"
    16  	"strings"
    17  	"sync"
    18  
    19  	"encoding/json/internal/jsonflags"
    20  	"encoding/json/internal/jsonopts"
    21  	"encoding/json/internal/jsonwire"
    22  	"encoding/json/jsontext"
    23  )
    24  
    25  // ErrUnknownName indicates that a JSON object member could not be
    26  // unmarshaled because the name is not known to the target Go struct.
    27  // This error is directly wrapped within a [SemanticError] when produced.
    28  //
    29  // The name of an unknown JSON object member can be extracted as:
    30  //
    31  //	err := ...
    32  //	serr, ok := errors.AsType[*json.SemanticError](err)
    33  //	if ok && serr.Err == json.ErrUnknownName {
    34  //		ptr := serr.JSONPointer // JSON pointer to unknown name
    35  //		name := ptr.LastToken() // unknown name itself
    36  //		...
    37  //	}
    38  //
    39  // This error is only returned if [RejectUnknownMembers] is true.
    40  var ErrUnknownName = errors.New("unknown object member name")
    41  
    42  const errorPrefix = "json: "
    43  
    44  func isSemanticError(err error) bool {
    45  	_, ok := err.(*SemanticError)
    46  	return ok
    47  }
    48  
    49  func isSyntacticError(err error) bool {
    50  	_, ok := err.(*jsontext.SyntacticError)
    51  	return ok
    52  }
    53  
    54  // isFatalError reports whether this error must terminate arshaling.
    55  // All errors are considered fatal unless operating under
    56  // [jsonflags.ReportErrorsWithLegacySemantics] in which case only
    57  // syntactic errors and I/O errors are considered fatal.
    58  func isFatalError(err error, flags jsonflags.Flags) bool {
    59  	return !flags.Get(jsonflags.ReportErrorsWithLegacySemantics) ||
    60  		isSyntacticError(err) || export.IsIOError(err)
    61  }
    62  
    63  // SemanticError describes an error determining the meaning
    64  // of JSON data as Go data or vice-versa.
    65  //
    66  // If a [Marshaler], [MarshalerTo], [Unmarshaler], or [UnmarshalerFrom] method
    67  // returns a SemanticError when called by the [json] package,
    68  // then the ByteOffset, JSONPointer, and GoType fields are automatically
    69  // populated by the calling context if they are the zero value.
    70  //
    71  // The contents of this error as produced by this package may change over time.
    72  type SemanticError struct {
    73  	requireKeyedLiterals
    74  	nonComparable
    75  
    76  	action string // either "marshal" or "unmarshal"
    77  
    78  	// ByteOffset indicates that an error occurred at or after this byte offset.
    79  	ByteOffset int64
    80  	// JSONPointer indicates that an error occurred within this JSON value
    81  	// as indicated using the JSON Pointer notation (see RFC 6901).
    82  	JSONPointer jsontext.Pointer
    83  
    84  	// JSONKind is the JSON kind that could not be handled.
    85  	JSONKind jsontext.Kind // may be zero if unknown
    86  	// JSONValue is the JSON number or string that could not be unmarshaled.
    87  	// It is not populated during marshaling.
    88  	JSONValue jsontext.Value // may be nil if irrelevant or unknown
    89  	// GoType is the Go type that could not be handled.
    90  	GoType reflect.Type // may be nil if unknown
    91  
    92  	// Err is the underlying error.
    93  	Err error // may be nil
    94  }
    95  
    96  // coder is implemented by [jsontext.Encoder] or [jsontext.Decoder].
    97  type coder interface {
    98  	StackPointer() jsontext.Pointer
    99  	Options() Options
   100  }
   101  
   102  // newInvalidFormatError constructs a SemanticError because
   103  // the current type t cannot handle the provided options format.
   104  // This function must be called before producing or consuming the next value.
   105  //
   106  // If [jsonflags.ReportErrorsWithLegacySemantics] is specified,
   107  // then this automatically skips the next value when unmarshaling
   108  // to ensure that the value is fully consumed.
   109  func newInvalidFormatError(c coder, t reflect.Type) error {
   110  	err := fmt.Errorf("invalid format flag %q", c.Options().(*jsonopts.Struct).Format)
   111  	switch c := c.(type) {
   112  	case *jsontext.Encoder:
   113  		err = newMarshalErrorBefore(c, t, err)
   114  	case *jsontext.Decoder:
   115  		err = newUnmarshalErrorBeforeWithSkipping(c, t, err)
   116  	}
   117  	return err
   118  }
   119  
   120  // newMarshalErrorBefore wraps err in a SemanticError assuming that e
   121  // is positioned right before the next token or value, which causes an error.
   122  func newMarshalErrorBefore(e *jsontext.Encoder, t reflect.Type, err error) error {
   123  	return &SemanticError{action: "marshal", GoType: t, Err: toUnexpectedEOF(err),
   124  		ByteOffset:  e.OutputOffset() + int64(export.Encoder(e).CountNextDelimWhitespace()),
   125  		JSONPointer: jsontext.Pointer(export.Encoder(e).AppendStackPointer(nil, +1))}
   126  }
   127  
   128  // newUnmarshalErrorBefore wraps err in a SemanticError assuming that d
   129  // is positioned right before the next token or value, which causes an error.
   130  // It does not record the next JSON kind as this error is used to indicate
   131  // the receiving Go value is invalid to unmarshal into (and not a JSON error).
   132  // However, if [jsonflags.ReportErrorsWithLegacySemantics] is specified,
   133  // then it does record the next JSON kind for historical reporting reasons.
   134  func newUnmarshalErrorBefore(d *jsontext.Decoder, t reflect.Type, err error) error {
   135  	var k jsontext.Kind
   136  	if export.Decoder(d).Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
   137  		k = d.PeekKind()
   138  	}
   139  	return &SemanticError{action: "unmarshal", GoType: t, Err: toUnexpectedEOF(err),
   140  		ByteOffset:  d.InputOffset() + int64(export.Decoder(d).CountNextDelimWhitespace()),
   141  		JSONPointer: jsontext.Pointer(export.Decoder(d).AppendStackPointer(nil, +1)),
   142  		JSONKind:    k}
   143  }
   144  
   145  // newUnmarshalErrorBeforeWithSkipping is like [newUnmarshalErrorBefore],
   146  // but automatically skips the next value if
   147  // [jsonflags.ReportErrorsWithLegacySemantics] is specified.
   148  func newUnmarshalErrorBeforeWithSkipping(d *jsontext.Decoder, t reflect.Type, err error) error {
   149  	err = newUnmarshalErrorBefore(d, t, err)
   150  	if export.Decoder(d).Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
   151  		if err2 := export.Decoder(d).SkipValue(); err2 != nil {
   152  			return err2
   153  		}
   154  	}
   155  	return err
   156  }
   157  
   158  // newUnmarshalErrorAfter wraps err in a SemanticError assuming that d
   159  // is positioned right after the previous token or value, which caused an error.
   160  func newUnmarshalErrorAfter(d *jsontext.Decoder, t reflect.Type, err error) error {
   161  	tokOrVal := export.Decoder(d).PreviousTokenOrValue()
   162  	byteOffset := d.InputOffset() - int64(len(tokOrVal))
   163  	if export.Decoder(d).Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
   164  		// NOTE: In v1, the offset pointed to the end of the bad token or value.
   165  		if k := jsontext.Value(tokOrVal).Kind(); k == '[' || k == '{' {
   166  			byteOffset++ // add just the '[' or '{'
   167  		} else {
   168  			byteOffset += int64(len(tokOrVal))
   169  		}
   170  	}
   171  	return &SemanticError{action: "unmarshal", GoType: t, Err: toUnexpectedEOF(err),
   172  		ByteOffset:  byteOffset,
   173  		JSONPointer: jsontext.Pointer(export.Decoder(d).AppendStackPointer(nil, -1)),
   174  		JSONKind:    jsontext.Value(tokOrVal).Kind()}
   175  }
   176  
   177  // newUnmarshalErrorAfterWithValue wraps err in a SemanticError assuming that d
   178  // is positioned right after the previous token or value, which caused an error.
   179  // It also stores a copy of the last JSON value if it is a string or number.
   180  func newUnmarshalErrorAfterWithValue(d *jsontext.Decoder, t reflect.Type, err error) error {
   181  	serr := newUnmarshalErrorAfter(d, t, err).(*SemanticError)
   182  	if serr.JSONKind == '"' || serr.JSONKind == '0' {
   183  		serr.JSONValue = jsontext.Value(export.Decoder(d).PreviousTokenOrValue()).Clone()
   184  	}
   185  	return serr
   186  }
   187  
   188  // newUnmarshalErrorAfterWithSkipping is like [newUnmarshalErrorAfter],
   189  // but automatically skips the remainder of the current value if
   190  // [jsonflags.ReportErrorsWithLegacySemantics] is specified.
   191  func newUnmarshalErrorAfterWithSkipping(d *jsontext.Decoder, t reflect.Type, err error) error {
   192  	err = newUnmarshalErrorAfter(d, t, err)
   193  	if export.Decoder(d).Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
   194  		if err2 := export.Decoder(d).SkipValueRemainder(); err2 != nil {
   195  			return err2
   196  		}
   197  	}
   198  	return err
   199  }
   200  
   201  // newSemanticErrorWithPosition wraps err in a SemanticError assuming that
   202  // the error occurred at the provided depth, and length.
   203  // If err is already a SemanticError, then position information is only
   204  // injected if it is currently unpopulated.
   205  //
   206  // If the position is unpopulated, it is ambiguous where the error occurred
   207  // in the user code, whether it was before or after the current position.
   208  // For the byte offset, we assume that the error occurred before the last read
   209  // token or value when decoding, or before the next value when encoding.
   210  // For the JSON pointer, we point to the parent object or array unless
   211  // we can be certain that it happened with an object member.
   212  //
   213  // This is used to annotate errors returned by user-provided
   214  // v2 MarshalJSON or UnmarshalJSON methods or functions.
   215  func newSemanticErrorWithPosition(c coder, t reflect.Type, prevDepth int, prevLength int64, err error) error {
   216  	serr, _ := err.(*SemanticError)
   217  	if serr == nil {
   218  		serr = &SemanticError{Err: err}
   219  	}
   220  	serr.Err = toUnexpectedEOF(serr.Err)
   221  	var currDepth int
   222  	var currLength int64
   223  	var coderState interface{ AppendStackPointer([]byte, int) []byte }
   224  	var offset int64
   225  	switch c := c.(type) {
   226  	case *jsontext.Encoder:
   227  		e := export.Encoder(c)
   228  		serr.action = cmp.Or(serr.action, "marshal")
   229  		currDepth, currLength = e.Tokens.DepthLength()
   230  		offset = c.OutputOffset() + int64(export.Encoder(c).CountNextDelimWhitespace())
   231  		coderState = e
   232  	case *jsontext.Decoder:
   233  		d := export.Decoder(c)
   234  		serr.action = cmp.Or(serr.action, "unmarshal")
   235  		currDepth, currLength = d.Tokens.DepthLength()
   236  		tokOrVal := d.PreviousTokenOrValue()
   237  		offset = c.InputOffset() - int64(len(tokOrVal))
   238  		if (prevDepth == currDepth && prevLength == currLength) || len(tokOrVal) == 0 {
   239  			// If no Read method was called in the user-defined method or
   240  			// if the Peek method was called, then use the offset of the next value.
   241  			offset = c.InputOffset() + int64(export.Decoder(c).CountNextDelimWhitespace())
   242  		}
   243  		coderState = d
   244  	}
   245  	serr.ByteOffset = cmp.Or(serr.ByteOffset, offset)
   246  	if serr.JSONPointer == "" {
   247  		where := 0 // default to ambiguous positioning
   248  		switch {
   249  		case prevDepth == currDepth && prevLength+0 == currLength:
   250  			where = +1
   251  		case prevDepth == currDepth && prevLength+1 == currLength:
   252  			where = -1
   253  		}
   254  		serr.JSONPointer = jsontext.Pointer(coderState.AppendStackPointer(nil, where))
   255  	}
   256  	serr.GoType = cmp.Or(serr.GoType, t)
   257  	return serr
   258  }
   259  
   260  // collapseSemanticErrors collapses double SemanticErrors at the outer levels
   261  // into a single SemanticError by preserving the inner error,
   262  // but prepending the ByteOffset and JSONPointer with the outer error.
   263  //
   264  // For example:
   265  //
   266  //	collapseSemanticErrors(&SemanticError{
   267  //		ByteOffset:  len64(`[0,{"alpha":[0,1,`),
   268  //		JSONPointer: "/1/alpha/2",
   269  //		GoType:      reflect.TypeFor[outerType](),
   270  //		Err: &SemanticError{
   271  //			ByteOffset:  len64(`{"foo":"bar","fizz":[0,`),
   272  //			JSONPointer: "/fizz/1",
   273  //			GoType:      reflect.TypeFor[innerType](),
   274  //			Err:         ...,
   275  //		},
   276  //	})
   277  //
   278  // results in:
   279  //
   280  //	&SemanticError{
   281  //		ByteOffset:  len64(`[0,{"alpha":[0,1,`) + len64(`{"foo":"bar","fizz":[0,`),
   282  //		JSONPointer: "/1/alpha/2" + "/fizz/1",
   283  //		GoType:      reflect.TypeFor[innerType](),
   284  //		Err:         ...,
   285  //	}
   286  //
   287  // This is used to annotate errors returned by user-provided
   288  // v1 MarshalJSON or UnmarshalJSON methods with precise position information
   289  // if they themselves happened to return a SemanticError.
   290  // Since MarshalJSON and UnmarshalJSON are not operating on the root JSON value,
   291  // their positioning must be relative to the nested JSON value
   292  // returned by UnmarshalJSON or passed to MarshalJSON.
   293  // Therefore, we can construct an absolute position by concatenating
   294  // the outer with the inner positions.
   295  //
   296  // Note that we do not use collapseSemanticErrors with user-provided functions
   297  // that take in an [jsontext.Encoder] or [jsontext.Decoder] since they contain
   298  // methods to report position relative to the root JSON value.
   299  // We assume user-constructed errors are correctly precise about position.
   300  func collapseSemanticErrors(err error) error {
   301  	if serr1, ok := err.(*SemanticError); ok {
   302  		if serr2, ok := serr1.Err.(*SemanticError); ok {
   303  			serr2.ByteOffset = serr1.ByteOffset + serr2.ByteOffset
   304  			serr2.JSONPointer = serr1.JSONPointer + serr2.JSONPointer
   305  			*serr1 = *serr2
   306  		}
   307  	}
   308  	return err
   309  }
   310  
   311  func wrapErrUnsupported(err error, what string) error {
   312  	if errors.Is(err, errors.ErrUnsupported) {
   313  		return errors.New(what + " may not return errors.ErrUnsupported")
   314  	}
   315  	return err
   316  }
   317  
   318  // errorModalVerb is a modal verb like "cannot" or "unable to".
   319  //
   320  // Once per process, Hyrum-proof the error message by deliberately
   321  // switching between equivalent renderings of the same error message.
   322  // The randomization is tied to the Hyrum-proofing already applied
   323  // on map iteration in Go.
   324  var errorModalVerb = sync.OnceValue(func() string {
   325  	for phrase := range map[string]struct{}{"cannot": {}, "unable to": {}} {
   326  		return phrase // use whichever phrase we get in the first iteration
   327  	}
   328  	return ""
   329  })
   330  
   331  func (e *SemanticError) Error() string {
   332  	var sb strings.Builder
   333  	sb.WriteString(errorPrefix)
   334  	sb.WriteString(errorModalVerb())
   335  
   336  	// Format action.
   337  	var preposition string
   338  	switch e.action {
   339  	case "marshal":
   340  		sb.WriteString(" marshal")
   341  		preposition = " from"
   342  	case "unmarshal":
   343  		sb.WriteString(" unmarshal")
   344  		preposition = " into"
   345  	default:
   346  		sb.WriteString(" handle")
   347  		preposition = " with"
   348  	}
   349  
   350  	// Format JSON kind.
   351  	switch e.JSONKind {
   352  	case 'n':
   353  		sb.WriteString(" JSON null")
   354  	case 'f', 't':
   355  		sb.WriteString(" JSON boolean")
   356  	case '"':
   357  		sb.WriteString(" JSON string")
   358  	case '0':
   359  		sb.WriteString(" JSON number")
   360  	case '{', '}':
   361  		sb.WriteString(" JSON object")
   362  	case '[', ']':
   363  		sb.WriteString(" JSON array")
   364  	default:
   365  		if e.action == "" {
   366  			preposition = ""
   367  		}
   368  	}
   369  	if len(e.JSONValue) > 0 && len(e.JSONValue) < 100 {
   370  		sb.WriteByte(' ')
   371  		sb.Write(e.JSONValue)
   372  	}
   373  
   374  	// Format Go type.
   375  	if e.GoType != nil {
   376  		typeString := e.GoType.String()
   377  		if len(typeString) > 100 {
   378  			// An excessively long type string most likely occurs for
   379  			// an anonymous struct declaration with many fields.
   380  			// Reduce the noise by just printing the kind,
   381  			// and optionally prepending it with the package name
   382  			// if the struct happens to include an unexported field.
   383  			typeString = e.GoType.Kind().String()
   384  			if e.GoType.Kind() == reflect.Struct && e.GoType.Name() == "" {
   385  				for i := range e.GoType.NumField() {
   386  					if pkgPath := e.GoType.Field(i).PkgPath; pkgPath != "" {
   387  						typeString = pkgPath[strings.LastIndexByte(pkgPath, '/')+len("/"):] + ".struct"
   388  						break
   389  					}
   390  				}
   391  			}
   392  		}
   393  		sb.WriteString(preposition)
   394  		sb.WriteString(" Go ")
   395  		sb.WriteString(typeString)
   396  	}
   397  
   398  	// Special handling for unknown names.
   399  	if e.Err == ErrUnknownName {
   400  		sb.WriteString(": ")
   401  		sb.WriteString(ErrUnknownName.Error())
   402  		sb.WriteString(" ")
   403  		sb.WriteString(strconv.Quote(e.JSONPointer.LastToken()))
   404  		if parent := e.JSONPointer.Parent(); parent != "" {
   405  			sb.WriteString(" within ")
   406  			sb.WriteString(strconv.Quote(jsonwire.TruncatePointer(string(parent), 100)))
   407  		}
   408  		return sb.String()
   409  	}
   410  
   411  	// Format where.
   412  	// Avoid printing if it overlaps with a wrapped SyntacticError.
   413  	switch serr, _ := e.Err.(*jsontext.SyntacticError); {
   414  	case e.JSONPointer != "":
   415  		if serr == nil || !e.JSONPointer.Contains(serr.JSONPointer) {
   416  			sb.WriteString(" within ")
   417  			sb.WriteString(strconv.Quote(jsonwire.TruncatePointer(string(e.JSONPointer), 100)))
   418  		}
   419  	case e.ByteOffset > 0:
   420  		if serr == nil || !(e.ByteOffset <= serr.ByteOffset) {
   421  			sb.WriteString(" after offset ")
   422  			sb.WriteString(strconv.FormatInt(e.ByteOffset, 10))
   423  		}
   424  	}
   425  
   426  	// Format underlying error.
   427  	if e.Err != nil {
   428  		errString := e.Err.Error()
   429  		if isSyntacticError(e.Err) {
   430  			errString = strings.TrimPrefix(errString, "jsontext: ")
   431  		}
   432  		sb.WriteString(": ")
   433  		sb.WriteString(errString)
   434  	}
   435  
   436  	return sb.String()
   437  }
   438  
   439  func (e *SemanticError) Unwrap() error {
   440  	return e.Err
   441  }
   442  
   443  func newDuplicateNameError(ptr jsontext.Pointer, quotedName []byte, offset int64) error {
   444  	if quotedName != nil {
   445  		name, _ := jsonwire.AppendUnquote(nil, quotedName)
   446  		ptr = ptr.AppendToken(string(name))
   447  	}
   448  	return &jsontext.SyntacticError{
   449  		ByteOffset:  offset,
   450  		JSONPointer: ptr,
   451  		Err:         jsontext.ErrDuplicateName,
   452  	}
   453  }
   454  
   455  // toUnexpectedEOF converts [io.EOF] to [io.ErrUnexpectedEOF].
   456  func toUnexpectedEOF(err error) error {
   457  	if err == io.EOF {
   458  		return io.ErrUnexpectedEOF
   459  	}
   460  	return err
   461  }
   462
View as plain text