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 asharling.
    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 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 wraps err in a SemanticError because
   103  // the current type t cannot handle the provided options format.
   104  // This error 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  	return &SemanticError{action: "unmarshal", GoType: t, Err: toUnexpectedEOF(err),
   163  		ByteOffset:  d.InputOffset() - int64(len(tokOrVal)),
   164  		JSONPointer: jsontext.Pointer(export.Decoder(d).AppendStackPointer(nil, -1)),
   165  		JSONKind:    jsontext.Value(tokOrVal).Kind()}
   166  }
   167  
   168  // newUnmarshalErrorAfter wraps err in a SemanticError assuming that d
   169  // is positioned right after the previous token or value, which caused an error.
   170  // It also stores a copy of the last JSON value if it is a string or number.
   171  func newUnmarshalErrorAfterWithValue(d *jsontext.Decoder, t reflect.Type, err error) error {
   172  	serr := newUnmarshalErrorAfter(d, t, err).(*SemanticError)
   173  	if serr.JSONKind == '"' || serr.JSONKind == '0' {
   174  		serr.JSONValue = jsontext.Value(export.Decoder(d).PreviousTokenOrValue()).Clone()
   175  	}
   176  	return serr
   177  }
   178  
   179  // newUnmarshalErrorAfterWithSkipping is like [newUnmarshalErrorAfter],
   180  // but automatically skips the remainder of the current value if
   181  // [jsonflags.ReportErrorsWithLegacySemantics] is specified.
   182  func newUnmarshalErrorAfterWithSkipping(d *jsontext.Decoder, t reflect.Type, err error) error {
   183  	err = newUnmarshalErrorAfter(d, t, err)
   184  	if export.Decoder(d).Flags.Get(jsonflags.ReportErrorsWithLegacySemantics) {
   185  		if err2 := export.Decoder(d).SkipValueRemainder(); err2 != nil {
   186  			return err2
   187  		}
   188  	}
   189  	return err
   190  }
   191  
   192  // newSemanticErrorWithPosition wraps err in a SemanticError assuming that
   193  // the error occurred at the provided depth, and length.
   194  // If err is already a SemanticError, then position information is only
   195  // injected if it is currently unpopulated.
   196  //
   197  // If the position is unpopulated, it is ambiguous where the error occurred
   198  // in the user code, whether it was before or after the current position.
   199  // For the byte offset, we assume that the error occurred before the last read
   200  // token or value when decoding, or before the next value when encoding.
   201  // For the JSON pointer, we point to the parent object or array unless
   202  // we can be certain that it happened with an object member.
   203  //
   204  // This is used to annotate errors returned by user-provided
   205  // v2 MarshalJSON or UnmarshalJSON methods or functions.
   206  func newSemanticErrorWithPosition(c coder, t reflect.Type, prevDepth int, prevLength int64, err error) error {
   207  	serr, _ := err.(*SemanticError)
   208  	if serr == nil {
   209  		serr = &SemanticError{Err: err}
   210  	}
   211  	serr.Err = toUnexpectedEOF(serr.Err)
   212  	var currDepth int
   213  	var currLength int64
   214  	var coderState interface{ AppendStackPointer([]byte, int) []byte }
   215  	var offset int64
   216  	switch c := c.(type) {
   217  	case *jsontext.Encoder:
   218  		e := export.Encoder(c)
   219  		serr.action = cmp.Or(serr.action, "marshal")
   220  		currDepth, currLength = e.Tokens.DepthLength()
   221  		offset = c.OutputOffset() + int64(export.Encoder(c).CountNextDelimWhitespace())
   222  		coderState = e
   223  	case *jsontext.Decoder:
   224  		d := export.Decoder(c)
   225  		serr.action = cmp.Or(serr.action, "unmarshal")
   226  		currDepth, currLength = d.Tokens.DepthLength()
   227  		tokOrVal := d.PreviousTokenOrValue()
   228  		offset = c.InputOffset() - int64(len(tokOrVal))
   229  		if (prevDepth == currDepth && prevLength == currLength) || len(tokOrVal) == 0 {
   230  			// If no Read method was called in the user-defined method or
   231  			// if the Peek method was called, then use the offset of the next value.
   232  			offset = c.InputOffset() + int64(export.Decoder(c).CountNextDelimWhitespace())
   233  		}
   234  		coderState = d
   235  	}
   236  	serr.ByteOffset = cmp.Or(serr.ByteOffset, offset)
   237  	if serr.JSONPointer == "" {
   238  		where := 0 // default to ambiguous positioning
   239  		switch {
   240  		case prevDepth == currDepth && prevLength+0 == currLength:
   241  			where = +1
   242  		case prevDepth == currDepth && prevLength+1 == currLength:
   243  			where = -1
   244  		}
   245  		serr.JSONPointer = jsontext.Pointer(coderState.AppendStackPointer(nil, where))
   246  	}
   247  	serr.GoType = cmp.Or(serr.GoType, t)
   248  	return serr
   249  }
   250  
   251  // collapseSemanticErrors collapses double SemanticErrors at the outer levels
   252  // into a single SemanticError by preserving the inner error,
   253  // but prepending the ByteOffset and JSONPointer with the outer error.
   254  //
   255  // For example:
   256  //
   257  //	collapseSemanticErrors(&SemanticError{
   258  //		ByteOffset:  len64(`[0,{"alpha":[0,1,`),
   259  //		JSONPointer: "/1/alpha/2",
   260  //		GoType:      reflect.TypeFor[outerType](),
   261  //		Err: &SemanticError{
   262  //			ByteOffset:  len64(`{"foo":"bar","fizz":[0,`),
   263  //			JSONPointer: "/fizz/1",
   264  //			GoType:      reflect.TypeFor[innerType](),
   265  //			Err:         ...,
   266  //		},
   267  //	})
   268  //
   269  // results in:
   270  //
   271  //	&SemanticError{
   272  //		ByteOffset:  len64(`[0,{"alpha":[0,1,`) + len64(`{"foo":"bar","fizz":[0,`),
   273  //		JSONPointer: "/1/alpha/2" + "/fizz/1",
   274  //		GoType:      reflect.TypeFor[innerType](),
   275  //		Err:         ...,
   276  //	}
   277  //
   278  // This is used to annotate errors returned by user-provided
   279  // v1 MarshalJSON or UnmarshalJSON methods with precise position information
   280  // if they themselves happened to return a SemanticError.
   281  // Since MarshalJSON and UnmarshalJSON are not operating on the root JSON value,
   282  // their positioning must be relative to the nested JSON value
   283  // returned by UnmarshalJSON or passed to MarshalJSON.
   284  // Therefore, we can construct an absolute position by concatenating
   285  // the outer with the inner positions.
   286  //
   287  // Note that we do not use collapseSemanticErrors with user-provided functions
   288  // that take in an [jsontext.Encoder] or [jsontext.Decoder] since they contain
   289  // methods to report position relative to the root JSON value.
   290  // We assume user-constructed errors are correctly precise about position.
   291  func collapseSemanticErrors(err error) error {
   292  	if serr1, ok := err.(*SemanticError); ok {
   293  		if serr2, ok := serr1.Err.(*SemanticError); ok {
   294  			serr2.ByteOffset = serr1.ByteOffset + serr2.ByteOffset
   295  			serr2.JSONPointer = serr1.JSONPointer + serr2.JSONPointer
   296  			*serr1 = *serr2
   297  		}
   298  	}
   299  	return err
   300  }
   301  
   302  // errorModalVerb is a modal verb like "cannot" or "unable to".
   303  //
   304  // Once per process, Hyrum-proof the error message by deliberately
   305  // switching between equivalent renderings of the same error message.
   306  // The randomization is tied to the Hyrum-proofing already applied
   307  // on map iteration in Go.
   308  var errorModalVerb = sync.OnceValue(func() string {
   309  	for phrase := range map[string]struct{}{"cannot": {}, "unable to": {}} {
   310  		return phrase // use whichever phrase we get in the first iteration
   311  	}
   312  	return ""
   313  })
   314  
   315  func (e *SemanticError) Error() string {
   316  	var sb strings.Builder
   317  	sb.WriteString(errorPrefix)
   318  	sb.WriteString(errorModalVerb())
   319  
   320  	// Format action.
   321  	var preposition string
   322  	switch e.action {
   323  	case "marshal":
   324  		sb.WriteString(" marshal")
   325  		preposition = " from"
   326  	case "unmarshal":
   327  		sb.WriteString(" unmarshal")
   328  		preposition = " into"
   329  	default:
   330  		sb.WriteString(" handle")
   331  		preposition = " with"
   332  	}
   333  
   334  	// Format JSON kind.
   335  	switch e.JSONKind {
   336  	case 'n':
   337  		sb.WriteString(" JSON null")
   338  	case 'f', 't':
   339  		sb.WriteString(" JSON boolean")
   340  	case '"':
   341  		sb.WriteString(" JSON string")
   342  	case '0':
   343  		sb.WriteString(" JSON number")
   344  	case '{', '}':
   345  		sb.WriteString(" JSON object")
   346  	case '[', ']':
   347  		sb.WriteString(" JSON array")
   348  	default:
   349  		if e.action == "" {
   350  			preposition = ""
   351  		}
   352  	}
   353  	if len(e.JSONValue) > 0 && len(e.JSONValue) < 100 {
   354  		sb.WriteByte(' ')
   355  		sb.Write(e.JSONValue)
   356  	}
   357  
   358  	// Format Go type.
   359  	if e.GoType != nil {
   360  		typeString := e.GoType.String()
   361  		if len(typeString) > 100 {
   362  			// An excessively long type string most likely occurs for
   363  			// an anonymous struct declaration with many fields.
   364  			// Reduce the noise by just printing the kind,
   365  			// and optionally prepending it with the package name
   366  			// if the struct happens to include an unexported field.
   367  			typeString = e.GoType.Kind().String()
   368  			if e.GoType.Kind() == reflect.Struct && e.GoType.Name() == "" {
   369  				for i := range e.GoType.NumField() {
   370  					if pkgPath := e.GoType.Field(i).PkgPath; pkgPath != "" {
   371  						typeString = pkgPath[strings.LastIndexByte(pkgPath, '/')+len("/"):] + ".struct"
   372  						break
   373  					}
   374  				}
   375  			}
   376  		}
   377  		sb.WriteString(preposition)
   378  		sb.WriteString(" Go ")
   379  		sb.WriteString(typeString)
   380  	}
   381  
   382  	// Special handling for unknown names.
   383  	if e.Err == ErrUnknownName {
   384  		sb.WriteString(": ")
   385  		sb.WriteString(ErrUnknownName.Error())
   386  		sb.WriteString(" ")
   387  		sb.WriteString(strconv.Quote(e.JSONPointer.LastToken()))
   388  		if parent := e.JSONPointer.Parent(); parent != "" {
   389  			sb.WriteString(" within ")
   390  			sb.WriteString(strconv.Quote(jsonwire.TruncatePointer(string(parent), 100)))
   391  		}
   392  		return sb.String()
   393  	}
   394  
   395  	// Format where.
   396  	// Avoid printing if it overlaps with a wrapped SyntacticError.
   397  	switch serr, _ := e.Err.(*jsontext.SyntacticError); {
   398  	case e.JSONPointer != "":
   399  		if serr == nil || !e.JSONPointer.Contains(serr.JSONPointer) {
   400  			sb.WriteString(" within ")
   401  			sb.WriteString(strconv.Quote(jsonwire.TruncatePointer(string(e.JSONPointer), 100)))
   402  		}
   403  	case e.ByteOffset > 0:
   404  		if serr == nil || !(e.ByteOffset <= serr.ByteOffset) {
   405  			sb.WriteString(" after offset ")
   406  			sb.WriteString(strconv.FormatInt(e.ByteOffset, 10))
   407  		}
   408  	}
   409  
   410  	// Format underlying error.
   411  	if e.Err != nil {
   412  		errString := e.Err.Error()
   413  		if isSyntacticError(e.Err) {
   414  			errString = strings.TrimPrefix(errString, "jsontext: ")
   415  		}
   416  		sb.WriteString(": ")
   417  		sb.WriteString(errString)
   418  	}
   419  
   420  	return sb.String()
   421  }
   422  
   423  func (e *SemanticError) Unwrap() error {
   424  	return e.Err
   425  }
   426  
   427  func newDuplicateNameError(ptr jsontext.Pointer, quotedName []byte, offset int64) error {
   428  	if quotedName != nil {
   429  		name, _ := jsonwire.AppendUnquote(nil, quotedName)
   430  		ptr = ptr.AppendToken(string(name))
   431  	}
   432  	return &jsontext.SyntacticError{
   433  		ByteOffset:  offset,
   434  		JSONPointer: ptr,
   435  		Err:         jsontext.ErrDuplicateName,
   436  	}
   437  }
   438  
   439  // toUnexpectedEOF converts [io.EOF] to [io.ErrUnexpectedEOF].
   440  func toUnexpectedEOF(err error) error {
   441  	if err == io.EOF {
   442  		return io.ErrUnexpectedEOF
   443  	}
   444  	return err
   445  }
   446
View as plain text