alandonovan | ac23acb | 2020-06-11 17:56:15 -0400 | [diff] [blame] | 1 | // Copyright 2020 The Bazel 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 starlarkjson defines utilities for converting Starlark values |
| 6 | // to/from JSON strings. The most recent IETF standard for JSON is |
| 7 | // https://www.ietf.org/rfc/rfc7159.txt. |
| 8 | package starlarkjson // import "go.starlark.net/starlarkjson" |
| 9 | |
| 10 | import ( |
| 11 | "bytes" |
| 12 | "encoding/json" |
| 13 | "fmt" |
| 14 | "log" |
| 15 | "math" |
| 16 | "math/big" |
| 17 | "sort" |
| 18 | "strconv" |
| 19 | "strings" |
| 20 | "unicode/utf8" |
| 21 | |
| 22 | "go.starlark.net/starlark" |
| 23 | "go.starlark.net/starlarkstruct" |
| 24 | ) |
| 25 | |
| 26 | // Module json is a Starlark module of JSON-related functions. |
| 27 | // |
| 28 | // json = module( |
| 29 | // encode, |
| 30 | // decode, |
| 31 | // indent, |
| 32 | // ) |
| 33 | // |
| 34 | // def encode(x): |
| 35 | // |
| 36 | // The encode function accepts one required positional argument, |
| 37 | // which it converts to JSON by cases: |
| 38 | // - A Starlark value that implements Go's standard json.Marshal |
| 39 | // interface defines its own JSON encoding. |
| 40 | // - None, True, and False are converted to null, true, and false, respectively. |
| 41 | // - Starlark int values, no matter how large, are encoded as decimal integers. |
| 42 | // Some decoders may not be able to decode very large integers. |
| 43 | // - Starlark float values are encoded using decimal point notation, |
| 44 | // even if the value is an integer. |
| 45 | // It is an error to encode a non-finite floating-point value. |
| 46 | // - Starlark strings are encoded as JSON strings, using UTF-16 escapes. |
| 47 | // - a Starlark IterableMapping (e.g. dict) is encoded as a JSON object. |
| 48 | // It is an error if any key is not a string. |
| 49 | // - any other Starlark Iterable (e.g. list, tuple) is encoded as a JSON array. |
| 50 | // - a Starlark HasAttrs (e.g. struct) is encoded as a JSON object. |
| 51 | // It an application-defined type matches more than one the cases describe above, |
| 52 | // (e.g. it implements both Iterable and HasFields), the first case takes precedence. |
| 53 | // Encoding any other value yields an error. |
| 54 | // |
| 55 | // def decode(x): |
| 56 | // |
| 57 | // The decode function accepts one positional parameter, a JSON string. |
| 58 | // It returns the Starlark value that the string denotes. |
| 59 | // - Numbers are parsed as int or float, depending on whether they |
| 60 | // contain a decimal point. |
| 61 | // - JSON objects are parsed as new unfrozen Starlark dicts. |
| 62 | // - JSON arrays are parsed as new unfrozen Starlark lists. |
| 63 | // Decoding fails if x is not a valid JSON string. |
| 64 | // |
| 65 | // def indent(str, *, prefix="", indent="\t"): |
| 66 | // |
| 67 | // The indent function pretty-prints a valid JSON encoding, |
| 68 | // and returns a string containing the indented form. |
| 69 | // It accepts one required positional parameter, the JSON string, |
| 70 | // and two optional keyword-only string parameters, prefix and indent, |
| 71 | // that specify a prefix of each new line, and the unit of indentation. |
| 72 | // |
| 73 | var Module = &starlarkstruct.Module{ |
| 74 | Name: "json", |
| 75 | Members: starlark.StringDict{ |
| 76 | "encode": starlark.NewBuiltin("json.encode", encode), |
| 77 | "decode": starlark.NewBuiltin("json.decode", decode), |
| 78 | "indent": starlark.NewBuiltin("json.indent", indent), |
| 79 | }, |
| 80 | } |
| 81 | |
| 82 | func encode(thread *starlark.Thread, b *starlark.Builtin, args starlark.Tuple, kwargs []starlark.Tuple) (starlark.Value, error) { |
| 83 | var x starlark.Value |
| 84 | if err := starlark.UnpackPositionalArgs(b.Name(), args, kwargs, 1, &x); err != nil { |
| 85 | return nil, err |
| 86 | } |
| 87 | |
| 88 | buf := new(bytes.Buffer) |
| 89 | |
| 90 | var quoteSpace [128]byte |
| 91 | quote := func(s string) { |
| 92 | // Non-trivial escaping is handled by Go's encoding/json. |
| 93 | if isPrintableASCII(s) { |
| 94 | buf.Write(strconv.AppendQuote(quoteSpace[:0], s)) |
| 95 | } else { |
| 96 | // TODO(adonovan): opt: RFC 8259 mandates UTF-8 for JSON. |
| 97 | // Can we avoid this call? |
| 98 | data, _ := json.Marshal(s) |
| 99 | buf.Write(data) |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | var emit func(x starlark.Value) error |
| 104 | emit = func(x starlark.Value) error { |
| 105 | switch x := x.(type) { |
| 106 | case json.Marshaler: |
| 107 | // Application-defined starlark.Value types |
| 108 | // may define their own JSON encoding. |
| 109 | data, err := x.MarshalJSON() |
| 110 | if err != nil { |
| 111 | return err |
| 112 | } |
| 113 | buf.Write(data) |
| 114 | |
| 115 | case starlark.NoneType: |
| 116 | buf.WriteString("null") |
| 117 | |
| 118 | case starlark.Bool: |
| 119 | if x { |
| 120 | buf.WriteString("true") |
| 121 | } else { |
| 122 | buf.WriteString("false") |
| 123 | } |
| 124 | |
| 125 | case starlark.Int: |
| 126 | fmt.Fprint(buf, x) |
| 127 | |
| 128 | case starlark.Float: |
| 129 | if !isFinite(float64(x)) { |
| 130 | return fmt.Errorf("cannot encode non-finite float %v", x) |
| 131 | } |
| 132 | fmt.Fprintf(buf, "%g", x) // always contains a decimal point |
| 133 | |
| 134 | case starlark.String: |
| 135 | quote(string(x)) |
| 136 | |
| 137 | case starlark.IterableMapping: |
| 138 | // e.g. dict (must have string keys) |
| 139 | buf.WriteByte('{') |
alandonovan | e292e66 | 2020-11-13 16:44:10 -0500 | [diff] [blame] | 140 | items := x.Items() |
| 141 | for _, item := range items { |
| 142 | if _, ok := item[0].(starlark.String); !ok { |
| 143 | return fmt.Errorf("%s has %s key, want string", x.Type(), item[0].Type()) |
| 144 | } |
| 145 | } |
| 146 | sort.Slice(items, func(i, j int) bool { |
| 147 | return items[i][0].(starlark.String) < items[j][0].(starlark.String) |
| 148 | }) |
| 149 | for i, item := range items { |
alandonovan | ac23acb | 2020-06-11 17:56:15 -0400 | [diff] [blame] | 150 | if i > 0 { |
| 151 | buf.WriteByte(',') |
| 152 | } |
alandonovan | e292e66 | 2020-11-13 16:44:10 -0500 | [diff] [blame] | 153 | k, _ := starlark.AsString(item[0]) |
| 154 | quote(k) |
alandonovan | ac23acb | 2020-06-11 17:56:15 -0400 | [diff] [blame] | 155 | buf.WriteByte(':') |
alandonovan | e292e66 | 2020-11-13 16:44:10 -0500 | [diff] [blame] | 156 | if err := emit(item[1]); err != nil { |
| 157 | return fmt.Errorf("in %s key %s: %v", x.Type(), item[0], err) |
alandonovan | ac23acb | 2020-06-11 17:56:15 -0400 | [diff] [blame] | 158 | } |
| 159 | } |
| 160 | buf.WriteByte('}') |
| 161 | |
| 162 | case starlark.Iterable: |
| 163 | // e.g. tuple, list |
| 164 | buf.WriteByte('[') |
| 165 | iter := x.Iterate() |
| 166 | defer iter.Done() |
| 167 | var elem starlark.Value |
| 168 | for i := 0; iter.Next(&elem); i++ { |
| 169 | if i > 0 { |
| 170 | buf.WriteByte(',') |
| 171 | } |
| 172 | if err := emit(elem); err != nil { |
| 173 | return fmt.Errorf("at %s index %d: %v", x.Type(), i, err) |
| 174 | } |
| 175 | } |
| 176 | buf.WriteByte(']') |
| 177 | |
| 178 | case starlark.HasAttrs: |
| 179 | // e.g. struct |
| 180 | buf.WriteByte('{') |
| 181 | var names []string |
| 182 | names = append(names, x.AttrNames()...) |
| 183 | sort.Strings(names) |
| 184 | for i, name := range names { |
| 185 | v, err := x.Attr(name) |
| 186 | if err != nil || v == nil { |
| 187 | log.Fatalf("internal error: dir(%s) includes %q but value has no .%s field", x.Type(), name, name) |
| 188 | } |
| 189 | if i > 0 { |
| 190 | buf.WriteByte(',') |
| 191 | } |
| 192 | quote(name) |
| 193 | buf.WriteByte(':') |
| 194 | if err := emit(v); err != nil { |
| 195 | return fmt.Errorf("in field .%s: %v", name, err) |
| 196 | } |
| 197 | } |
| 198 | buf.WriteByte('}') |
| 199 | |
| 200 | default: |
| 201 | return fmt.Errorf("cannot encode %s as JSON", x.Type()) |
| 202 | } |
| 203 | return nil |
| 204 | } |
| 205 | |
| 206 | if err := emit(x); err != nil { |
| 207 | return nil, fmt.Errorf("%s: %v", b.Name(), err) |
| 208 | } |
| 209 | return starlark.String(buf.String()), nil |
| 210 | } |
| 211 | |
| 212 | // isPrintableASCII reports whether s contains only printable ASCII. |
| 213 | func isPrintableASCII(s string) bool { |
| 214 | for i := 0; i < len(s); i++ { |
| 215 | b := s[i] |
| 216 | if b < 0x20 || b >= 0x80 { |
| 217 | return false |
| 218 | } |
| 219 | } |
| 220 | return true |
| 221 | } |
| 222 | |
| 223 | // isFinite reports whether f represents a finite rational value. |
| 224 | // It is equivalent to !math.IsNan(f) && !math.IsInf(f, 0). |
| 225 | func isFinite(f float64) bool { |
| 226 | return math.Abs(f) <= math.MaxFloat64 |
| 227 | } |
| 228 | |
| 229 | func indent(thread *starlark.Thread, b *starlark.Builtin, args starlark.Tuple, kwargs []starlark.Tuple) (starlark.Value, error) { |
| 230 | prefix, indent := "", "\t" // keyword-only |
| 231 | if err := starlark.UnpackArgs(b.Name(), nil, kwargs, |
| 232 | "prefix?", &prefix, |
| 233 | "indent?", &indent, |
| 234 | ); err != nil { |
| 235 | return nil, err |
| 236 | } |
| 237 | var str string // positional-only |
| 238 | if err := starlark.UnpackPositionalArgs(b.Name(), args, nil, 1, &str); err != nil { |
| 239 | return nil, err |
| 240 | } |
| 241 | |
| 242 | buf := new(bytes.Buffer) |
| 243 | if err := json.Indent(buf, []byte(str), prefix, indent); err != nil { |
| 244 | return nil, fmt.Errorf("%s: %v", b.Name(), err) |
| 245 | } |
| 246 | return starlark.String(buf.String()), nil |
| 247 | } |
| 248 | |
| 249 | func decode(thread *starlark.Thread, b *starlark.Builtin, args starlark.Tuple, kwargs []starlark.Tuple) (_ starlark.Value, err error) { |
| 250 | var s string |
| 251 | if err := starlark.UnpackPositionalArgs(b.Name(), args, kwargs, 1, &s); err != nil { |
| 252 | return nil, err |
| 253 | } |
| 254 | |
| 255 | // The decoder necessarily makes certain representation choices |
| 256 | // such as list vs tuple, struct vs dict, int vs float. |
| 257 | // In principle, we could parameterize it to allow the caller to |
| 258 | // control the returned types, but there's no compelling need yet. |
| 259 | |
| 260 | // Use panic/recover with a distinguished type (failure) for error handling. |
| 261 | type failure string |
| 262 | fail := func(format string, args ...interface{}) { |
| 263 | panic(failure(fmt.Sprintf(format, args...))) |
| 264 | } |
| 265 | |
| 266 | i := 0 |
| 267 | |
| 268 | // skipSpace consumes leading spaces, and reports whether there is more input. |
| 269 | skipSpace := func() bool { |
| 270 | for ; i < len(s); i++ { |
| 271 | b := s[i] |
| 272 | if b != ' ' && b != '\t' && b != '\n' && b != '\r' { |
| 273 | return true |
| 274 | } |
| 275 | } |
| 276 | return false |
| 277 | } |
| 278 | |
| 279 | // next consumes leading spaces and returns the first non-space. |
| 280 | // It panics if at EOF. |
| 281 | next := func() byte { |
| 282 | if skipSpace() { |
| 283 | return s[i] |
| 284 | } |
| 285 | fail("unexpected end of file") |
| 286 | panic("unreachable") |
| 287 | } |
| 288 | |
| 289 | // parse returns the next JSON value from the input. |
| 290 | // It consumes leading but not trailing whitespace. |
| 291 | // It panics on error. |
| 292 | var parse func() starlark.Value |
| 293 | parse = func() starlark.Value { |
| 294 | b := next() |
| 295 | switch b { |
| 296 | case '"': |
| 297 | // string |
| 298 | |
| 299 | // Find end of quotation. |
| 300 | // Also, record whether trivial unquoting is safe. |
| 301 | // Non-trivial unquoting is handled by Go's encoding/json. |
| 302 | safe := true |
| 303 | closed := false |
| 304 | j := i + 1 |
| 305 | for ; j < len(s); j++ { |
| 306 | b := s[j] |
| 307 | if b == '\\' { |
| 308 | safe = false |
| 309 | j++ // skip x in \x |
| 310 | } else if b == '"' { |
| 311 | closed = true |
| 312 | j++ // skip '"' |
| 313 | break |
| 314 | } else if b >= utf8.RuneSelf { |
| 315 | safe = false |
| 316 | } |
| 317 | } |
| 318 | if !closed { |
| 319 | fail("unclosed string literal") |
| 320 | } |
| 321 | |
| 322 | r := s[i:j] |
| 323 | i = j |
| 324 | |
| 325 | // unquote |
| 326 | if safe { |
| 327 | r = r[1 : len(r)-1] |
| 328 | } else if err := json.Unmarshal([]byte(r), &r); err != nil { |
| 329 | fail("%s", err) |
| 330 | } |
| 331 | return starlark.String(r) |
| 332 | |
| 333 | case 'n': |
| 334 | if strings.HasPrefix(s[i:], "null") { |
| 335 | i += len("null") |
| 336 | return starlark.None |
| 337 | } |
| 338 | |
| 339 | case 't': |
| 340 | if strings.HasPrefix(s[i:], "true") { |
| 341 | i += len("true") |
| 342 | return starlark.True |
| 343 | } |
| 344 | |
| 345 | case 'f': |
| 346 | if strings.HasPrefix(s[i:], "false") { |
| 347 | i += len("false") |
| 348 | return starlark.False |
| 349 | } |
| 350 | |
| 351 | case '[': |
| 352 | // array |
| 353 | var elems []starlark.Value |
| 354 | |
| 355 | i++ // '[' |
| 356 | b = next() |
| 357 | if b != ']' { |
| 358 | for { |
| 359 | elem := parse() |
| 360 | elems = append(elems, elem) |
| 361 | b = next() |
| 362 | if b != ',' { |
| 363 | if b != ']' { |
| 364 | fail("got %q, want ',' or ']'", b) |
| 365 | } |
| 366 | break |
| 367 | } |
| 368 | i++ // ',' |
| 369 | } |
| 370 | } |
| 371 | i++ // ']' |
| 372 | return starlark.NewList(elems) |
| 373 | |
| 374 | case '{': |
| 375 | // object |
| 376 | dict := new(starlark.Dict) |
| 377 | |
| 378 | i++ // '{' |
| 379 | b = next() |
| 380 | if b != '}' { |
| 381 | for { |
| 382 | key := parse() |
| 383 | if _, ok := key.(starlark.String); !ok { |
| 384 | fail("got %s for object key, want string", key.Type()) |
| 385 | } |
| 386 | b = next() |
| 387 | if b != ':' { |
| 388 | fail("after object key, got %q, want ':' ", b) |
| 389 | } |
| 390 | i++ // ':' |
| 391 | value := parse() |
| 392 | dict.SetKey(key, value) // can't fail |
| 393 | b = next() |
| 394 | if b != ',' { |
| 395 | if b != '}' { |
| 396 | fail("in object, got %q, want ',' or '}'", b) |
| 397 | } |
| 398 | break |
| 399 | } |
| 400 | i++ // ',' |
| 401 | } |
| 402 | } |
| 403 | i++ // '}' |
| 404 | return dict |
| 405 | |
| 406 | default: |
| 407 | // number? |
| 408 | if isdigit(b) || b == '-' { |
| 409 | // scan literal. Allow [0-9+-eE.] for now. |
| 410 | float := false |
| 411 | var j int |
| 412 | for j = i + 1; j < len(s); j++ { |
| 413 | b = s[j] |
| 414 | if isdigit(b) { |
| 415 | // ok |
| 416 | } else if b == '.' || |
| 417 | b == 'e' || |
| 418 | b == 'E' || |
| 419 | b == '+' || |
| 420 | b == '-' { |
| 421 | float = true |
| 422 | } else { |
| 423 | break |
| 424 | } |
| 425 | } |
| 426 | num := s[i:j] |
| 427 | i = j |
| 428 | |
| 429 | // Unlike most C-like languages, |
| 430 | // JSON disallows a leading zero before a digit. |
| 431 | digits := num |
| 432 | if num[0] == '-' { |
| 433 | digits = num[1:] |
| 434 | } |
| 435 | if digits == "" || digits[0] == '0' && len(digits) > 1 && isdigit(digits[1]) { |
| 436 | fail("invalid number: %s", num) |
| 437 | } |
| 438 | |
| 439 | // parse literal |
| 440 | if float { |
| 441 | x, err := strconv.ParseFloat(num, 64) |
| 442 | if err != nil { |
| 443 | fail("invalid number: %s", num) |
| 444 | } |
| 445 | return starlark.Float(x) |
| 446 | } else { |
| 447 | x, ok := new(big.Int).SetString(num, 10) |
| 448 | if !ok { |
| 449 | fail("invalid number: %s", num) |
| 450 | } |
| 451 | return starlark.MakeBigInt(x) |
| 452 | } |
| 453 | } |
| 454 | } |
| 455 | fail("unexpected character %q", b) |
| 456 | panic("unreachable") |
| 457 | } |
| 458 | defer func() { |
| 459 | x := recover() |
| 460 | switch x := x.(type) { |
| 461 | case failure: |
| 462 | err = fmt.Errorf("json.decode: at offset %d, %s", i, x) |
| 463 | case nil: |
| 464 | // nop |
| 465 | default: |
| 466 | panic(x) // unexpected panic |
| 467 | } |
| 468 | }() |
| 469 | x := parse() |
| 470 | if skipSpace() { |
| 471 | fail("unexpected character %q after value", s[i]) |
| 472 | } |
| 473 | return x, nil |
| 474 | } |
| 475 | |
| 476 | func isdigit(b byte) bool { |
| 477 | return b >= '0' && b <= '9' |
| 478 | } |