blob: 8205a4dd13889afa08b7414991d5528d7405251b [file] [log] [blame]
Matthias Andreas Benkard832a54e2019-01-29 09:27:38 +01001// Copyright 2013 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 json is forked from the Go standard library to enable us to find the
6// field of a struct that a given JSON key maps to.
7package json
8
9import (
10 "bytes"
11 "fmt"
12 "reflect"
13 "sort"
14 "strings"
15 "sync"
16 "unicode"
17 "unicode/utf8"
18)
19
20const (
21 patchStrategyTagKey = "patchStrategy"
22 patchMergeKeyTagKey = "patchMergeKey"
23)
24
25// Finds the patchStrategy and patchMergeKey struct tag fields on a given
26// struct field given the struct type and the JSON name of the field.
27// It returns field type, a slice of patch strategies, merge key and error.
28// TODO: fix the returned errors to be introspectable.
29func LookupPatchMetadataForStruct(t reflect.Type, jsonField string) (
30 elemType reflect.Type, patchStrategies []string, patchMergeKey string, e error) {
31 if t.Kind() == reflect.Ptr {
32 t = t.Elem()
33 }
34
35 if t.Kind() != reflect.Struct {
36 e = fmt.Errorf("merging an object in json but data type is not struct, instead is: %s",
37 t.Kind().String())
38 return
39 }
40 jf := []byte(jsonField)
41 // Find the field that the JSON library would use.
42 var f *field
43 fields := cachedTypeFields(t)
44 for i := range fields {
45 ff := &fields[i]
46 if bytes.Equal(ff.nameBytes, jf) {
47 f = ff
48 break
49 }
50 // Do case-insensitive comparison.
51 if f == nil && ff.equalFold(ff.nameBytes, jf) {
52 f = ff
53 }
54 }
55 if f != nil {
56 // Find the reflect.Value of the most preferential struct field.
57 tjf := t.Field(f.index[0])
58 // we must navigate down all the anonymously included structs in the chain
59 for i := 1; i < len(f.index); i++ {
60 tjf = tjf.Type.Field(f.index[i])
61 }
62 patchStrategy := tjf.Tag.Get(patchStrategyTagKey)
63 patchMergeKey = tjf.Tag.Get(patchMergeKeyTagKey)
64 patchStrategies = strings.Split(patchStrategy, ",")
65 elemType = tjf.Type
66 return
67 }
68 e = fmt.Errorf("unable to find api field in struct %s for the json field %q", t.Name(), jsonField)
69 return
70}
71
72// A field represents a single field found in a struct.
73type field struct {
74 name string
75 nameBytes []byte // []byte(name)
76 equalFold func(s, t []byte) bool // bytes.EqualFold or equivalent
77
78 tag bool
79 // index is the sequence of indexes from the containing type fields to this field.
80 // it is a slice because anonymous structs will need multiple navigation steps to correctly
81 // resolve the proper fields
82 index []int
83 typ reflect.Type
84 omitEmpty bool
85 quoted bool
86}
87
88func (f field) String() string {
89 return fmt.Sprintf("{name: %s, type: %v, tag: %v, index: %v, omitEmpty: %v, quoted: %v}", f.name, f.typ, f.tag, f.index, f.omitEmpty, f.quoted)
90}
91
92func fillField(f field) field {
93 f.nameBytes = []byte(f.name)
94 f.equalFold = foldFunc(f.nameBytes)
95 return f
96}
97
98// byName sorts field by name, breaking ties with depth,
99// then breaking ties with "name came from json tag", then
100// breaking ties with index sequence.
101type byName []field
102
103func (x byName) Len() int { return len(x) }
104
105func (x byName) Swap(i, j int) { x[i], x[j] = x[j], x[i] }
106
107func (x byName) Less(i, j int) bool {
108 if x[i].name != x[j].name {
109 return x[i].name < x[j].name
110 }
111 if len(x[i].index) != len(x[j].index) {
112 return len(x[i].index) < len(x[j].index)
113 }
114 if x[i].tag != x[j].tag {
115 return x[i].tag
116 }
117 return byIndex(x).Less(i, j)
118}
119
120// byIndex sorts field by index sequence.
121type byIndex []field
122
123func (x byIndex) Len() int { return len(x) }
124
125func (x byIndex) Swap(i, j int) { x[i], x[j] = x[j], x[i] }
126
127func (x byIndex) Less(i, j int) bool {
128 for k, xik := range x[i].index {
129 if k >= len(x[j].index) {
130 return false
131 }
132 if xik != x[j].index[k] {
133 return xik < x[j].index[k]
134 }
135 }
136 return len(x[i].index) < len(x[j].index)
137}
138
139// typeFields returns a list of fields that JSON should recognize for the given type.
140// The algorithm is breadth-first search over the set of structs to include - the top struct
141// and then any reachable anonymous structs.
142func typeFields(t reflect.Type) []field {
143 // Anonymous fields to explore at the current level and the next.
144 current := []field{}
145 next := []field{{typ: t}}
146
147 // Count of queued names for current level and the next.
148 count := map[reflect.Type]int{}
149 nextCount := map[reflect.Type]int{}
150
151 // Types already visited at an earlier level.
152 visited := map[reflect.Type]bool{}
153
154 // Fields found.
155 var fields []field
156
157 for len(next) > 0 {
158 current, next = next, current[:0]
159 count, nextCount = nextCount, map[reflect.Type]int{}
160
161 for _, f := range current {
162 if visited[f.typ] {
163 continue
164 }
165 visited[f.typ] = true
166
167 // Scan f.typ for fields to include.
168 for i := 0; i < f.typ.NumField(); i++ {
169 sf := f.typ.Field(i)
170 if sf.PkgPath != "" { // unexported
171 continue
172 }
173 tag := sf.Tag.Get("json")
174 if tag == "-" {
175 continue
176 }
177 name, opts := parseTag(tag)
178 if !isValidTag(name) {
179 name = ""
180 }
181 index := make([]int, len(f.index)+1)
182 copy(index, f.index)
183 index[len(f.index)] = i
184
185 ft := sf.Type
186 if ft.Name() == "" && ft.Kind() == reflect.Ptr {
187 // Follow pointer.
188 ft = ft.Elem()
189 }
190
191 // Record found field and index sequence.
192 if name != "" || !sf.Anonymous || ft.Kind() != reflect.Struct {
193 tagged := name != ""
194 if name == "" {
195 name = sf.Name
196 }
197 fields = append(fields, fillField(field{
198 name: name,
199 tag: tagged,
200 index: index,
201 typ: ft,
202 omitEmpty: opts.Contains("omitempty"),
203 quoted: opts.Contains("string"),
204 }))
205 if count[f.typ] > 1 {
206 // If there were multiple instances, add a second,
207 // so that the annihilation code will see a duplicate.
208 // It only cares about the distinction between 1 or 2,
209 // so don't bother generating any more copies.
210 fields = append(fields, fields[len(fields)-1])
211 }
212 continue
213 }
214
215 // Record new anonymous struct to explore in next round.
216 nextCount[ft]++
217 if nextCount[ft] == 1 {
218 next = append(next, fillField(field{name: ft.Name(), index: index, typ: ft}))
219 }
220 }
221 }
222 }
223
224 sort.Sort(byName(fields))
225
226 // Delete all fields that are hidden by the Go rules for embedded fields,
227 // except that fields with JSON tags are promoted.
228
229 // The fields are sorted in primary order of name, secondary order
230 // of field index length. Loop over names; for each name, delete
231 // hidden fields by choosing the one dominant field that survives.
232 out := fields[:0]
233 for advance, i := 0, 0; i < len(fields); i += advance {
234 // One iteration per name.
235 // Find the sequence of fields with the name of this first field.
236 fi := fields[i]
237 name := fi.name
238 for advance = 1; i+advance < len(fields); advance++ {
239 fj := fields[i+advance]
240 if fj.name != name {
241 break
242 }
243 }
244 if advance == 1 { // Only one field with this name
245 out = append(out, fi)
246 continue
247 }
248 dominant, ok := dominantField(fields[i : i+advance])
249 if ok {
250 out = append(out, dominant)
251 }
252 }
253
254 fields = out
255 sort.Sort(byIndex(fields))
256
257 return fields
258}
259
260// dominantField looks through the fields, all of which are known to
261// have the same name, to find the single field that dominates the
262// others using Go's embedding rules, modified by the presence of
263// JSON tags. If there are multiple top-level fields, the boolean
264// will be false: This condition is an error in Go and we skip all
265// the fields.
266func dominantField(fields []field) (field, bool) {
267 // The fields are sorted in increasing index-length order. The winner
268 // must therefore be one with the shortest index length. Drop all
269 // longer entries, which is easy: just truncate the slice.
270 length := len(fields[0].index)
271 tagged := -1 // Index of first tagged field.
272 for i, f := range fields {
273 if len(f.index) > length {
274 fields = fields[:i]
275 break
276 }
277 if f.tag {
278 if tagged >= 0 {
279 // Multiple tagged fields at the same level: conflict.
280 // Return no field.
281 return field{}, false
282 }
283 tagged = i
284 }
285 }
286 if tagged >= 0 {
287 return fields[tagged], true
288 }
289 // All remaining fields have the same length. If there's more than one,
290 // we have a conflict (two fields named "X" at the same level) and we
291 // return no field.
292 if len(fields) > 1 {
293 return field{}, false
294 }
295 return fields[0], true
296}
297
298var fieldCache struct {
299 sync.RWMutex
300 m map[reflect.Type][]field
301}
302
303// cachedTypeFields is like typeFields but uses a cache to avoid repeated work.
304func cachedTypeFields(t reflect.Type) []field {
305 fieldCache.RLock()
306 f := fieldCache.m[t]
307 fieldCache.RUnlock()
308 if f != nil {
309 return f
310 }
311
312 // Compute fields without lock.
313 // Might duplicate effort but won't hold other computations back.
314 f = typeFields(t)
315 if f == nil {
316 f = []field{}
317 }
318
319 fieldCache.Lock()
320 if fieldCache.m == nil {
321 fieldCache.m = map[reflect.Type][]field{}
322 }
323 fieldCache.m[t] = f
324 fieldCache.Unlock()
325 return f
326}
327
328func isValidTag(s string) bool {
329 if s == "" {
330 return false
331 }
332 for _, c := range s {
333 switch {
334 case strings.ContainsRune("!#$%&()*+-./:<=>?@[]^_{|}~ ", c):
335 // Backslash and quote chars are reserved, but
336 // otherwise any punctuation chars are allowed
337 // in a tag name.
338 default:
339 if !unicode.IsLetter(c) && !unicode.IsDigit(c) {
340 return false
341 }
342 }
343 }
344 return true
345}
346
347const (
348 caseMask = ^byte(0x20) // Mask to ignore case in ASCII.
349 kelvin = '\u212a'
350 smallLongEss = '\u017f'
351)
352
353// foldFunc returns one of four different case folding equivalence
354// functions, from most general (and slow) to fastest:
355//
356// 1) bytes.EqualFold, if the key s contains any non-ASCII UTF-8
357// 2) equalFoldRight, if s contains special folding ASCII ('k', 'K', 's', 'S')
358// 3) asciiEqualFold, no special, but includes non-letters (including _)
359// 4) simpleLetterEqualFold, no specials, no non-letters.
360//
361// The letters S and K are special because they map to 3 runes, not just 2:
362// * S maps to s and to U+017F 'ſ' Latin small letter long s
363// * k maps to K and to U+212A 'K' Kelvin sign
364// See http://play.golang.org/p/tTxjOc0OGo
365//
366// The returned function is specialized for matching against s and
367// should only be given s. It's not curried for performance reasons.
368func foldFunc(s []byte) func(s, t []byte) bool {
369 nonLetter := false
370 special := false // special letter
371 for _, b := range s {
372 if b >= utf8.RuneSelf {
373 return bytes.EqualFold
374 }
375 upper := b & caseMask
376 if upper < 'A' || upper > 'Z' {
377 nonLetter = true
378 } else if upper == 'K' || upper == 'S' {
379 // See above for why these letters are special.
380 special = true
381 }
382 }
383 if special {
384 return equalFoldRight
385 }
386 if nonLetter {
387 return asciiEqualFold
388 }
389 return simpleLetterEqualFold
390}
391
392// equalFoldRight is a specialization of bytes.EqualFold when s is
393// known to be all ASCII (including punctuation), but contains an 's',
394// 'S', 'k', or 'K', requiring a Unicode fold on the bytes in t.
395// See comments on foldFunc.
396func equalFoldRight(s, t []byte) bool {
397 for _, sb := range s {
398 if len(t) == 0 {
399 return false
400 }
401 tb := t[0]
402 if tb < utf8.RuneSelf {
403 if sb != tb {
404 sbUpper := sb & caseMask
405 if 'A' <= sbUpper && sbUpper <= 'Z' {
406 if sbUpper != tb&caseMask {
407 return false
408 }
409 } else {
410 return false
411 }
412 }
413 t = t[1:]
414 continue
415 }
416 // sb is ASCII and t is not. t must be either kelvin
417 // sign or long s; sb must be s, S, k, or K.
418 tr, size := utf8.DecodeRune(t)
419 switch sb {
420 case 's', 'S':
421 if tr != smallLongEss {
422 return false
423 }
424 case 'k', 'K':
425 if tr != kelvin {
426 return false
427 }
428 default:
429 return false
430 }
431 t = t[size:]
432
433 }
434 if len(t) > 0 {
435 return false
436 }
437 return true
438}
439
440// asciiEqualFold is a specialization of bytes.EqualFold for use when
441// s is all ASCII (but may contain non-letters) and contains no
442// special-folding letters.
443// See comments on foldFunc.
444func asciiEqualFold(s, t []byte) bool {
445 if len(s) != len(t) {
446 return false
447 }
448 for i, sb := range s {
449 tb := t[i]
450 if sb == tb {
451 continue
452 }
453 if ('a' <= sb && sb <= 'z') || ('A' <= sb && sb <= 'Z') {
454 if sb&caseMask != tb&caseMask {
455 return false
456 }
457 } else {
458 return false
459 }
460 }
461 return true
462}
463
464// simpleLetterEqualFold is a specialization of bytes.EqualFold for
465// use when s is all ASCII letters (no underscores, etc) and also
466// doesn't contain 'k', 'K', 's', or 'S'.
467// See comments on foldFunc.
468func simpleLetterEqualFold(s, t []byte) bool {
469 if len(s) != len(t) {
470 return false
471 }
472 for i, b := range s {
473 if b&caseMask != t[i]&caseMask {
474 return false
475 }
476 }
477 return true
478}
479
480// tagOptions is the string following a comma in a struct field's "json"
481// tag, or the empty string. It does not include the leading comma.
482type tagOptions string
483
484// parseTag splits a struct field's json tag into its name and
485// comma-separated options.
486func parseTag(tag string) (string, tagOptions) {
487 if idx := strings.Index(tag, ","); idx != -1 {
488 return tag[:idx], tagOptions(tag[idx+1:])
489 }
490 return tag, tagOptions("")
491}
492
493// Contains reports whether a comma-separated list of options
494// contains a particular substr flag. substr must be surrounded by a
495// string boundary or commas.
496func (o tagOptions) Contains(optionName string) bool {
497 if len(o) == 0 {
498 return false
499 }
500 s := string(o)
501 for s != "" {
502 var next string
503 i := strings.Index(s, ",")
504 if i >= 0 {
505 s, next = s[:i], s[i+1:]
506 }
507 if s == optionName {
508 return true
509 }
510 s = next
511 }
512 return false
513}