David K. Bainbridge | 215e024 | 2017-09-05 23:18:24 -0700 | [diff] [blame] | 1 | // Copyright 2011 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 | package xml |
| 6 | |
| 7 | import ( |
| 8 | "fmt" |
| 9 | "reflect" |
| 10 | "strings" |
| 11 | "sync" |
| 12 | ) |
| 13 | |
| 14 | // typeInfo holds details for the xml representation of a type. |
| 15 | type typeInfo struct { |
| 16 | xmlname *fieldInfo |
| 17 | fields []fieldInfo |
| 18 | } |
| 19 | |
| 20 | // fieldInfo holds details for the xml representation of a single field. |
| 21 | type fieldInfo struct { |
| 22 | idx []int |
| 23 | name string |
| 24 | xmlns string |
| 25 | flags fieldFlags |
| 26 | parents []string |
| 27 | } |
| 28 | |
| 29 | type fieldFlags int |
| 30 | |
| 31 | const ( |
| 32 | fElement fieldFlags = 1 << iota |
| 33 | fAttr |
| 34 | fCharData |
| 35 | fInnerXml |
| 36 | fComment |
| 37 | fAny |
| 38 | |
| 39 | fOmitEmpty |
| 40 | |
| 41 | fMode = fElement | fAttr | fCharData | fInnerXml | fComment | fAny |
| 42 | ) |
| 43 | |
| 44 | var tinfoMap = make(map[reflect.Type]*typeInfo) |
| 45 | var tinfoLock sync.RWMutex |
| 46 | |
| 47 | var nameType = reflect.TypeOf(Name{}) |
| 48 | |
| 49 | // getTypeInfo returns the typeInfo structure with details necessary |
| 50 | // for marshalling and unmarshalling typ. |
| 51 | func getTypeInfo(typ reflect.Type) (*typeInfo, error) { |
| 52 | tinfoLock.RLock() |
| 53 | tinfo, ok := tinfoMap[typ] |
| 54 | tinfoLock.RUnlock() |
| 55 | if ok { |
| 56 | return tinfo, nil |
| 57 | } |
| 58 | tinfo = &typeInfo{} |
| 59 | if typ.Kind() == reflect.Struct && typ != nameType { |
| 60 | n := typ.NumField() |
| 61 | for i := 0; i < n; i++ { |
| 62 | f := typ.Field(i) |
| 63 | if f.PkgPath != "" || f.Tag.Get("xml") == "-" { |
| 64 | continue // Private field |
| 65 | } |
| 66 | |
| 67 | // For embedded structs, embed its fields. |
| 68 | if f.Anonymous { |
| 69 | t := f.Type |
| 70 | if t.Kind() == reflect.Ptr { |
| 71 | t = t.Elem() |
| 72 | } |
| 73 | if t.Kind() == reflect.Struct { |
| 74 | inner, err := getTypeInfo(t) |
| 75 | if err != nil { |
| 76 | return nil, err |
| 77 | } |
| 78 | if tinfo.xmlname == nil { |
| 79 | tinfo.xmlname = inner.xmlname |
| 80 | } |
| 81 | for _, finfo := range inner.fields { |
| 82 | finfo.idx = append([]int{i}, finfo.idx...) |
| 83 | if err := addFieldInfo(typ, tinfo, &finfo); err != nil { |
| 84 | return nil, err |
| 85 | } |
| 86 | } |
| 87 | continue |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | finfo, err := structFieldInfo(typ, &f) |
| 92 | if err != nil { |
| 93 | return nil, err |
| 94 | } |
| 95 | |
| 96 | if f.Name == "XMLName" { |
| 97 | tinfo.xmlname = finfo |
| 98 | continue |
| 99 | } |
| 100 | |
| 101 | // Add the field if it doesn't conflict with other fields. |
| 102 | if err := addFieldInfo(typ, tinfo, finfo); err != nil { |
| 103 | return nil, err |
| 104 | } |
| 105 | } |
| 106 | } |
| 107 | tinfoLock.Lock() |
| 108 | tinfoMap[typ] = tinfo |
| 109 | tinfoLock.Unlock() |
| 110 | return tinfo, nil |
| 111 | } |
| 112 | |
| 113 | // structFieldInfo builds and returns a fieldInfo for f. |
| 114 | func structFieldInfo(typ reflect.Type, f *reflect.StructField) (*fieldInfo, error) { |
| 115 | finfo := &fieldInfo{idx: f.Index} |
| 116 | |
| 117 | // Split the tag from the xml namespace if necessary. |
| 118 | tag := f.Tag.Get("xml") |
| 119 | if i := strings.Index(tag, " "); i >= 0 { |
| 120 | finfo.xmlns, tag = tag[:i], tag[i+1:] |
| 121 | } |
| 122 | |
| 123 | // Parse flags. |
| 124 | tokens := strings.Split(tag, ",") |
| 125 | if len(tokens) == 1 { |
| 126 | finfo.flags = fElement |
| 127 | } else { |
| 128 | tag = tokens[0] |
| 129 | for _, flag := range tokens[1:] { |
| 130 | switch flag { |
| 131 | case "attr": |
| 132 | finfo.flags |= fAttr |
| 133 | case "chardata": |
| 134 | finfo.flags |= fCharData |
| 135 | case "innerxml": |
| 136 | finfo.flags |= fInnerXml |
| 137 | case "comment": |
| 138 | finfo.flags |= fComment |
| 139 | case "any": |
| 140 | finfo.flags |= fAny |
| 141 | case "omitempty": |
| 142 | finfo.flags |= fOmitEmpty |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | // Validate the flags used. |
| 147 | valid := true |
| 148 | switch mode := finfo.flags & fMode; mode { |
| 149 | case 0: |
| 150 | finfo.flags |= fElement |
| 151 | case fAttr, fCharData, fInnerXml, fComment, fAny: |
| 152 | if f.Name == "XMLName" || tag != "" && mode != fAttr { |
| 153 | valid = false |
| 154 | } |
| 155 | default: |
| 156 | // This will also catch multiple modes in a single field. |
| 157 | valid = false |
| 158 | } |
| 159 | if finfo.flags&fMode == fAny { |
| 160 | finfo.flags |= fElement |
| 161 | } |
| 162 | if finfo.flags&fOmitEmpty != 0 && finfo.flags&(fElement|fAttr) == 0 { |
| 163 | valid = false |
| 164 | } |
| 165 | if !valid { |
| 166 | return nil, fmt.Errorf("xml: invalid tag in field %s of type %s: %q", |
| 167 | f.Name, typ, f.Tag.Get("xml")) |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | // Use of xmlns without a name is not allowed. |
| 172 | if finfo.xmlns != "" && tag == "" { |
| 173 | return nil, fmt.Errorf("xml: namespace without name in field %s of type %s: %q", |
| 174 | f.Name, typ, f.Tag.Get("xml")) |
| 175 | } |
| 176 | |
| 177 | if f.Name == "XMLName" { |
| 178 | // The XMLName field records the XML element name. Don't |
| 179 | // process it as usual because its name should default to |
| 180 | // empty rather than to the field name. |
| 181 | finfo.name = tag |
| 182 | return finfo, nil |
| 183 | } |
| 184 | |
| 185 | if tag == "" { |
| 186 | // If the name part of the tag is completely empty, get |
| 187 | // default from XMLName of underlying struct if feasible, |
| 188 | // or field name otherwise. |
| 189 | if xmlname := lookupXMLName(f.Type); xmlname != nil { |
| 190 | finfo.xmlns, finfo.name = xmlname.xmlns, xmlname.name |
| 191 | } else { |
| 192 | finfo.name = f.Name |
| 193 | } |
| 194 | return finfo, nil |
| 195 | } |
| 196 | |
| 197 | if finfo.xmlns == "" && finfo.flags&fAttr == 0 { |
| 198 | // If it's an element no namespace specified, get the default |
| 199 | // from the XMLName of enclosing struct if possible. |
| 200 | if xmlname := lookupXMLName(typ); xmlname != nil { |
| 201 | finfo.xmlns = xmlname.xmlns |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | // Prepare field name and parents. |
| 206 | parents := strings.Split(tag, ">") |
| 207 | if parents[0] == "" { |
| 208 | parents[0] = f.Name |
| 209 | } |
| 210 | if parents[len(parents)-1] == "" { |
| 211 | return nil, fmt.Errorf("xml: trailing '>' in field %s of type %s", f.Name, typ) |
| 212 | } |
| 213 | finfo.name = parents[len(parents)-1] |
| 214 | if len(parents) > 1 { |
| 215 | if (finfo.flags & fElement) == 0 { |
| 216 | return nil, fmt.Errorf("xml: %s chain not valid with %s flag", tag, strings.Join(tokens[1:], ",")) |
| 217 | } |
| 218 | finfo.parents = parents[:len(parents)-1] |
| 219 | } |
| 220 | |
| 221 | // If the field type has an XMLName field, the names must match |
| 222 | // so that the behavior of both marshalling and unmarshalling |
| 223 | // is straightforward and unambiguous. |
| 224 | if finfo.flags&fElement != 0 { |
| 225 | ftyp := f.Type |
| 226 | xmlname := lookupXMLName(ftyp) |
| 227 | if xmlname != nil && xmlname.name != finfo.name { |
| 228 | return nil, fmt.Errorf("xml: name %q in tag of %s.%s conflicts with name %q in %s.XMLName", |
| 229 | finfo.name, typ, f.Name, xmlname.name, ftyp) |
| 230 | } |
| 231 | } |
| 232 | return finfo, nil |
| 233 | } |
| 234 | |
| 235 | // lookupXMLName returns the fieldInfo for typ's XMLName field |
| 236 | // in case it exists and has a valid xml field tag, otherwise |
| 237 | // it returns nil. |
| 238 | func lookupXMLName(typ reflect.Type) (xmlname *fieldInfo) { |
| 239 | for typ.Kind() == reflect.Ptr { |
| 240 | typ = typ.Elem() |
| 241 | } |
| 242 | if typ.Kind() != reflect.Struct { |
| 243 | return nil |
| 244 | } |
| 245 | for i, n := 0, typ.NumField(); i < n; i++ { |
| 246 | f := typ.Field(i) |
| 247 | if f.Name != "XMLName" { |
| 248 | continue |
| 249 | } |
| 250 | finfo, err := structFieldInfo(typ, &f) |
| 251 | if finfo.name != "" && err == nil { |
| 252 | return finfo |
| 253 | } |
| 254 | // Also consider errors as a non-existent field tag |
| 255 | // and let getTypeInfo itself report the error. |
| 256 | break |
| 257 | } |
| 258 | return nil |
| 259 | } |
| 260 | |
| 261 | func min(a, b int) int { |
| 262 | if a <= b { |
| 263 | return a |
| 264 | } |
| 265 | return b |
| 266 | } |
| 267 | |
| 268 | // addFieldInfo adds finfo to tinfo.fields if there are no |
| 269 | // conflicts, or if conflicts arise from previous fields that were |
| 270 | // obtained from deeper embedded structures than finfo. In the latter |
| 271 | // case, the conflicting entries are dropped. |
| 272 | // A conflict occurs when the path (parent + name) to a field is |
| 273 | // itself a prefix of another path, or when two paths match exactly. |
| 274 | // It is okay for field paths to share a common, shorter prefix. |
| 275 | func addFieldInfo(typ reflect.Type, tinfo *typeInfo, newf *fieldInfo) error { |
| 276 | var conflicts []int |
| 277 | Loop: |
| 278 | // First, figure all conflicts. Most working code will have none. |
| 279 | for i := range tinfo.fields { |
| 280 | oldf := &tinfo.fields[i] |
| 281 | if oldf.flags&fMode != newf.flags&fMode { |
| 282 | continue |
| 283 | } |
| 284 | if oldf.xmlns != "" && newf.xmlns != "" && oldf.xmlns != newf.xmlns { |
| 285 | continue |
| 286 | } |
| 287 | minl := min(len(newf.parents), len(oldf.parents)) |
| 288 | for p := 0; p < minl; p++ { |
| 289 | if oldf.parents[p] != newf.parents[p] { |
| 290 | continue Loop |
| 291 | } |
| 292 | } |
| 293 | if len(oldf.parents) > len(newf.parents) { |
| 294 | if oldf.parents[len(newf.parents)] == newf.name { |
| 295 | conflicts = append(conflicts, i) |
| 296 | } |
| 297 | } else if len(oldf.parents) < len(newf.parents) { |
| 298 | if newf.parents[len(oldf.parents)] == oldf.name { |
| 299 | conflicts = append(conflicts, i) |
| 300 | } |
| 301 | } else { |
| 302 | if newf.name == oldf.name { |
| 303 | conflicts = append(conflicts, i) |
| 304 | } |
| 305 | } |
| 306 | } |
| 307 | // Without conflicts, add the new field and return. |
| 308 | if conflicts == nil { |
| 309 | tinfo.fields = append(tinfo.fields, *newf) |
| 310 | return nil |
| 311 | } |
| 312 | |
| 313 | // If any conflict is shallower, ignore the new field. |
| 314 | // This matches the Go field resolution on embedding. |
| 315 | for _, i := range conflicts { |
| 316 | if len(tinfo.fields[i].idx) < len(newf.idx) { |
| 317 | return nil |
| 318 | } |
| 319 | } |
| 320 | |
| 321 | // Otherwise, if any of them is at the same depth level, it's an error. |
| 322 | for _, i := range conflicts { |
| 323 | oldf := &tinfo.fields[i] |
| 324 | if len(oldf.idx) == len(newf.idx) { |
| 325 | f1 := typ.FieldByIndex(oldf.idx) |
| 326 | f2 := typ.FieldByIndex(newf.idx) |
| 327 | return &TagPathError{typ, f1.Name, f1.Tag.Get("xml"), f2.Name, f2.Tag.Get("xml")} |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | // Otherwise, the new field is shallower, and thus takes precedence, |
| 332 | // so drop the conflicting fields from tinfo and append the new one. |
| 333 | for c := len(conflicts) - 1; c >= 0; c-- { |
| 334 | i := conflicts[c] |
| 335 | copy(tinfo.fields[i:], tinfo.fields[i+1:]) |
| 336 | tinfo.fields = tinfo.fields[:len(tinfo.fields)-1] |
| 337 | } |
| 338 | tinfo.fields = append(tinfo.fields, *newf) |
| 339 | return nil |
| 340 | } |
| 341 | |
| 342 | // A TagPathError represents an error in the unmarshalling process |
| 343 | // caused by the use of field tags with conflicting paths. |
| 344 | type TagPathError struct { |
| 345 | Struct reflect.Type |
| 346 | Field1, Tag1 string |
| 347 | Field2, Tag2 string |
| 348 | } |
| 349 | |
| 350 | func (e *TagPathError) Error() string { |
| 351 | return fmt.Sprintf("%s field %q with tag %q conflicts with field %q with tag %q", e.Struct, e.Field1, e.Tag1, e.Field2, e.Tag2) |
| 352 | } |
| 353 | |
| 354 | // value returns v's field value corresponding to finfo. |
| 355 | // It's equivalent to v.FieldByIndex(finfo.idx), but initializes |
| 356 | // and dereferences pointers as necessary. |
| 357 | func (finfo *fieldInfo) value(v reflect.Value) reflect.Value { |
| 358 | for i, x := range finfo.idx { |
| 359 | if i > 0 { |
| 360 | t := v.Type() |
| 361 | if t.Kind() == reflect.Ptr && t.Elem().Kind() == reflect.Struct { |
| 362 | if v.IsNil() { |
| 363 | v.Set(reflect.New(v.Type().Elem())) |
| 364 | } |
| 365 | v = v.Elem() |
| 366 | } |
| 367 | } |
| 368 | v = v.Field(x) |
| 369 | } |
| 370 | return v |
| 371 | } |