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