libgo/go/runtime/string.go

   1 // Copyright 2014 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 runtime
   6
   7 import (
   8         "internal/bytealg"
   9         "runtime/internal/sys"
  10         "unsafe"
  11 )
  12
  13 // For gccgo, use go:linkname to export compiler-called functions.
  14 //
  15 //go:linkname concatstrings
  16 //go:linkname slicebytetostring
  17 //go:linkname slicebytetostringtmp
  18 //go:linkname stringtoslicebyte
  19 //go:linkname stringtoslicerune
  20 //go:linkname slicerunetostring
  21 //go:linkname intstring
  22 // Temporary for C code to call:
  23 //go:linkname gostringnocopy
  24 //go:linkname findnull
  25
  26 // The constant is known to the compiler.
  27 // There is no fundamental theory behind this number.
  28 const tmpStringBufSize = 32
  29
  30 type tmpBuf [tmpStringBufSize]byte
  31
  32 // concatstrings implements a Go string concatenation x+y+z+...
  33 // The operands are passed in the slice a.
  34 // If buf != nil, the compiler has determined that the result does not
  35 // escape the calling function, so the string data can be stored in buf
  36 // if small enough.
  37 func concatstrings(buf *tmpBuf, p *string, n int) string {
  38         var a []string
  39         *(*slice)(unsafe.Pointer(&a)) = slice{unsafe.Pointer(p), n, n}
  40         // idx := 0
  41         l := 0
  42         count := 0
  43         for _, x := range a {
  44                 n := len(x)
  45                 if n == 0 {
  46                         continue
  47                 }
  48                 if l+n < l {
  49                         throw("string concatenation too long")
  50                 }
  51                 l += n
  52                 count++
  53                 // idx = i
  54         }
  55         if count == 0 {
  56                 return ""
  57         }
  58
  59         // If there is just one string and either it is not on the stack
  60         // or our result does not escape the calling frame (buf != nil),
  61         // then we can return that string directly.
  62         // Commented out for gccgo--no implementation of stringDataOnStack.
  63         // if count == 1 && (buf != nil || !stringDataOnStack(a[idx])) {
  64         //      return a[idx]
  65         // }
  66         s, b := rawstringtmp(buf, l)
  67         for _, x := range a {
  68                 copy(b, x)
  69                 b = b[len(x):]
  70         }
  71         return s
  72 }
  73
  74 // slicebytetostring converts a byte slice to a string.
  75 // It is inserted by the compiler into generated code.
  76 // ptr is a pointer to the first element of the slice;
  77 // n is the length of the slice.
  78 // Buf is a fixed-size buffer for the result,
  79 // it is not nil if the result does not escape.
  80 func slicebytetostring(buf *tmpBuf, ptr *byte, n int) (str string) {
  81         if n == 0 {
  82                 // Turns out to be a relatively common case.
  83                 // Consider that you want to parse out data between parens in "foo()bar",
  84                 // you find the indices and convert the subslice to string.
  85                 return ""
  86         }
  87         if raceenabled {
  88                 racereadrangepc(unsafe.Pointer(ptr),
  89                         uintptr(n),
  90                         getcallerpc(),
  91                         funcPC(slicebytetostring))
  92         }
  93         if msanenabled {
  94                 msanread(unsafe.Pointer(ptr), uintptr(n))
  95         }
  96         if n == 1 {
  97                 p := unsafe.Pointer(&staticuint64s[*ptr])
  98                 if sys.BigEndian {
  99                         p = add(p, 7)
 100                 }
 101                 stringStructOf(&str).str = p
 102                 stringStructOf(&str).len = 1
 103                 return
 104         }
 105
 106         var p unsafe.Pointer
 107         if buf != nil && n <= len(buf) {
 108                 p = unsafe.Pointer(buf)
 109         } else {
 110                 p = mallocgc(uintptr(n), nil, false)
 111         }
 112         stringStructOf(&str).str = p
 113         stringStructOf(&str).len = n
 114         memmove(p, unsafe.Pointer(ptr), uintptr(n))
 115         return
 116 }
 117
 118 func rawstringtmp(buf *tmpBuf, l int) (s string, b []byte) {
 119         if buf != nil && l <= len(buf) {
 120                 b = buf[:l]
 121                 s = slicebytetostringtmp(&b[0], len(b))
 122         } else {
 123                 s, b = rawstring(l)
 124         }
 125         return
 126 }
 127
 128 // slicebytetostringtmp returns a "string" referring to the actual []byte bytes.
 129 //
 130 // Callers need to ensure that the returned string will not be used after
 131 // the calling goroutine modifies the original slice or synchronizes with
 132 // another goroutine.
 133 //
 134 // The function is only called when instrumenting
 135 // and otherwise intrinsified by the compiler.
 136 //
 137 // Some internal compiler optimizations use this function.
 138 // - Used for m[T1{... Tn{..., string(k), ...} ...}] and m[string(k)]
 139 //   where k is []byte, T1 to Tn is a nesting of struct and array literals.
 140 // - Used for "<"+string(b)+">" concatenation where b is []byte.
 141 // - Used for string(b)=="foo" comparison where b is []byte.
 142 func slicebytetostringtmp(ptr *byte, n int) (str string) {
 143         if raceenabled && n > 0 {
 144                 racereadrangepc(unsafe.Pointer(ptr),
 145                         uintptr(n),
 146                         getcallerpc(),
 147                         funcPC(slicebytetostringtmp))
 148         }
 149         if msanenabled && n > 0 {
 150                 msanread(unsafe.Pointer(ptr), uintptr(n))
 151         }
 152         stringStructOf(&str).str = unsafe.Pointer(ptr)
 153         stringStructOf(&str).len = n
 154         return
 155 }
 156
 157 func stringtoslicebyte(buf *tmpBuf, s string) []byte {
 158         var b []byte
 159         if buf != nil && len(s) <= len(buf) {
 160                 *buf = tmpBuf{}
 161                 b = buf[:len(s)]
 162         } else {
 163                 b = rawbyteslice(len(s))
 164         }
 165         copy(b, s)
 166         return b
 167 }
 168
 169 func stringtoslicerune(buf *[tmpStringBufSize]rune, s string) []rune {
 170         // two passes.
 171         // unlike slicerunetostring, no race because strings are immutable.
 172         n := 0
 173         for range s {
 174                 n++
 175         }
 176
 177         var a []rune
 178         if buf != nil && n <= len(buf) {
 179                 *buf = [tmpStringBufSize]rune{}
 180                 a = buf[:n]
 181         } else {
 182                 a = rawruneslice(n)
 183         }
 184
 185         n = 0
 186         for _, r := range s {
 187                 a[n] = r
 188                 n++
 189         }
 190         return a
 191 }
 192
 193 func slicerunetostring(buf *tmpBuf, a []rune) string {
 194         if raceenabled && len(a) > 0 {
 195                 racereadrangepc(unsafe.Pointer(&a[0]),
 196                         uintptr(len(a))*unsafe.Sizeof(a[0]),
 197                         getcallerpc(),
 198                         funcPC(slicerunetostring))
 199         }
 200         if msanenabled && len(a) > 0 {
 201                 msanread(unsafe.Pointer(&a[0]), uintptr(len(a))*unsafe.Sizeof(a[0]))
 202         }
 203         var dum [4]byte
 204         size1 := 0
 205         for _, r := range a {
 206                 size1 += encoderune(dum[:], r)
 207         }
 208         s, b := rawstringtmp(buf, size1+3)
 209         size2 := 0
 210         for _, r := range a {
 211                 // check for race
 212                 if size2 >= size1 {
 213                         break
 214                 }
 215                 size2 += encoderune(b[size2:], r)
 216         }
 217         return s[:size2]
 218 }
 219
 220 type stringStruct struct {
 221         str unsafe.Pointer
 222         len int
 223 }
 224
 225 // Variant with *byte pointer type for DWARF debugging.
 226 type stringStructDWARF struct {
 227         str *byte
 228         len int
 229 }
 230
 231 func stringStructOf(sp *string) *stringStruct {
 232         return (*stringStruct)(unsafe.Pointer(sp))
 233 }
 234
 235 func intstring(buf *[4]byte, v int64) (s string) {
 236         var b []byte
 237         if buf != nil {
 238                 b = buf[:]
 239                 s = slicebytetostringtmp(&b[0], len(b))
 240         } else {
 241                 s, b = rawstring(4)
 242         }
 243         if int64(rune(v)) != v {
 244                 v = runeError
 245         }
 246         n := encoderune(b, rune(v))
 247         return s[:n]
 248 }
 249
 250 // rawstring allocates storage for a new string. The returned
 251 // string and byte slice both refer to the same storage.
 252 // The storage is not zeroed. Callers should use
 253 // b to set the string contents and then drop b.
 254 func rawstring(size int) (s string, b []byte) {
 255         p := mallocgc(uintptr(size), nil, false)
 256
 257         stringStructOf(&s).str = p
 258         stringStructOf(&s).len = size
 259
 260         *(*slice)(unsafe.Pointer(&b)) = slice{p, size, size}
 261
 262         return
 263 }
 264
 265 // rawbyteslice allocates a new byte slice. The byte slice is not zeroed.
 266 func rawbyteslice(size int) (b []byte) {
 267         cap := roundupsize(uintptr(size))
 268         p := mallocgc(cap, nil, false)
 269         if cap != uintptr(size) {
 270                 memclrNoHeapPointers(add(p, uintptr(size)), cap-uintptr(size))
 271         }
 272
 273         *(*slice)(unsafe.Pointer(&b)) = slice{p, size, int(cap)}
 274         return
 275 }
 276
 277 // rawruneslice allocates a new rune slice. The rune slice is not zeroed.
 278 func rawruneslice(size int) (b []rune) {
 279         if uintptr(size) > maxAlloc/4 {
 280                 throw("out of memory")
 281         }
 282         mem := roundupsize(uintptr(size) * 4)
 283         p := mallocgc(mem, nil, false)
 284         if mem != uintptr(size)*4 {
 285                 memclrNoHeapPointers(add(p, uintptr(size)*4), mem-uintptr(size)*4)
 286         }
 287
 288         *(*slice)(unsafe.Pointer(&b)) = slice{p, size, int(mem / 4)}
 289         return
 290 }
 291
 292 // used by cmd/cgo
 293 func gobytes(p *byte, n int) (b []byte) {
 294         if n == 0 {
 295                 return make([]byte, 0)
 296         }
 297
 298         if n < 0 || uintptr(n) > maxAlloc {
 299                 panic(errorString("gobytes: length out of range"))
 300         }
 301
 302         bp := mallocgc(uintptr(n), nil, false)
 303         memmove(bp, unsafe.Pointer(p), uintptr(n))
 304
 305         *(*slice)(unsafe.Pointer(&b)) = slice{bp, n, n}
 306         return
 307 }
 308
 309 // This is exported via linkname to assembly in syscall (for Plan9).
 310 //go:linkname gostring
 311 func gostring(p *byte) string {
 312         l := findnull(p)
 313         if l == 0 {
 314                 return ""
 315         }
 316         s, b := rawstring(l)
 317         memmove(unsafe.Pointer(&b[0]), unsafe.Pointer(p), uintptr(l))
 318         return s
 319 }
 320
 321 func gostringn(p *byte, l int) string {
 322         if l == 0 {
 323                 return ""
 324         }
 325         s, b := rawstring(l)
 326         memmove(unsafe.Pointer(&b[0]), unsafe.Pointer(p), uintptr(l))
 327         return s
 328 }
 329
 330 func index(s, t string) int {
 331         if len(t) == 0 {
 332                 return 0
 333         }
 334         for i := 0; i < len(s); i++ {
 335                 if s[i] == t[0] && hasPrefix(s[i:], t) {
 336                         return i
 337                 }
 338         }
 339         return -1
 340 }
 341
 342 func contains(s, t string) bool {
 343         return index(s, t) >= 0
 344 }
 345
 346 func hasPrefix(s, prefix string) bool {
 347         return len(s) >= len(prefix) && s[:len(prefix)] == prefix
 348 }
 349
 350 func hasSuffix(s, suffix string) bool {
 351         return len(s) >= len(suffix) && s[len(s)-len(suffix):] == suffix
 352 }
 353
 354 const (
 355         maxUint = ^uint(0)
 356         maxInt  = int(maxUint >> 1)
 357 )
 358
 359 // atoi parses an int from a string s.
 360 // The bool result reports whether s is a number
 361 // representable by a value of type int.
 362 func atoi(s string) (int, bool) {
 363         if s == "" {
 364                 return 0, false
 365         }
 366
 367         neg := false
 368         if s[0] == '-' {
 369                 neg = true
 370                 s = s[1:]
 371         }
 372
 373         un := uint(0)
 374         for i := 0; i < len(s); i++ {
 375                 c := s[i]
 376                 if c < '0' || c > '9' {
 377                         return 0, false
 378                 }
 379                 if un > maxUint/10 {
 380                         // overflow
 381                         return 0, false
 382                 }
 383                 un *= 10
 384                 un1 := un + uint(c) - '0'
 385                 if un1 < un {
 386                         // overflow
 387                         return 0, false
 388                 }
 389                 un = un1
 390         }
 391
 392         if !neg && un > uint(maxInt) {
 393                 return 0, false
 394         }
 395         if neg && un > uint(maxInt)+1 {
 396                 return 0, false
 397         }
 398
 399         n := int(un)
 400         if neg {
 401                 n = -n
 402         }
 403
 404         return n, true
 405 }
 406
 407 // atoi32 is like atoi but for integers
 408 // that fit into an int32.
 409 func atoi32(s string) (int32, bool) {
 410         if n, ok := atoi(s); n == int(int32(n)) {
 411                 return int32(n), ok
 412         }
 413         return 0, false
 414 }
 415
 416 //go:nosplit
 417 func findnull(s *byte) int {
 418         if s == nil {
 419                 return 0
 420         }
 421
 422         // Avoid IndexByteString on Plan 9 because it uses SSE instructions
 423         // on x86 machines, and those are classified as floating point instructions,
 424         // which are illegal in a note handler.
 425         if GOOS == "plan9" {
 426                 p := (*[maxAlloc/2 - 1]byte)(unsafe.Pointer(s))
 427                 l := 0
 428                 for p[l] != 0 {
 429                         l++
 430                 }
 431                 return l
 432         }
 433
 434         // pageSize is the unit we scan at a time looking for NULL.
 435         // It must be the minimum page size for any architecture Go
 436         // runs on. It's okay (just a minor performance loss) if the
 437         // actual system page size is larger than this value.
 438         const pageSize = 4096
 439
 440         offset := 0
 441         ptr := unsafe.Pointer(s)
 442         // IndexByteString uses wide reads, so we need to be careful
 443         // with page boundaries. Call IndexByteString on
 444         // [ptr, endOfPage) interval.
 445         safeLen := int(pageSize - uintptr(ptr)%pageSize)
 446
 447         for {
 448                 t := *(*string)(unsafe.Pointer(&stringStruct{ptr, safeLen}))
 449                 // Check one page at a time.
 450                 if i := bytealg.IndexByteString(t, 0); i != -1 {
 451                         return offset + i
 452                 }
 453                 // Move to next page
 454                 ptr = unsafe.Pointer(uintptr(ptr) + uintptr(safeLen))
 455                 offset += safeLen
 456                 safeLen = pageSize
 457         }
 458 }
 459
 460 func findnullw(s *uint16) int {
 461         if s == nil {
 462                 return 0
 463         }
 464         p := (*[maxAlloc/2/2 - 1]uint16)(unsafe.Pointer(s))
 465         l := 0
 466         for p[l] != 0 {
 467                 l++
 468         }
 469         return l
 470 }
 471
 472 //go:nosplit
 473 func gostringnocopy(str *byte) string {
 474         ss := stringStruct{str: unsafe.Pointer(str), len: findnull(str)}
 475         s := *(*string)(unsafe.Pointer(&ss))
 476         return s
 477 }
 478
 479 func gostringw(strw *uint16) string {
 480         var buf [8]byte
 481         str := (*[maxAlloc/2/2 - 1]uint16)(unsafe.Pointer(strw))
 482         n1 := 0
 483         for i := 0; str[i] != 0; i++ {
 484                 n1 += encoderune(buf[:], rune(str[i]))
 485         }
 486         s, b := rawstring(n1 + 4)
 487         n2 := 0
 488         for i := 0; str[i] != 0; i++ {
 489                 // check for race
 490                 if n2 >= n1 {
 491                         break
 492                 }
 493                 n2 += encoderune(b[n2:], rune(str[i]))
 494         }
 495         b[n2] = 0 // for luck
 496         return s[:n2]
 497 }
 498
 499 // These two functions are called by code generated by cgo -gccgo.
 500
 501 //go:linkname __go_byte_array_to_string __go_byte_array_to_string
 502 func __go_byte_array_to_string(p unsafe.Pointer, l int) string {
 503         if l == 0 {
 504                 return ""
 505         }
 506         s, c := rawstringtmp(nil, l)
 507         memmove(unsafe.Pointer(&c[0]), p, uintptr(l))
 508         return s
 509 }
 510
 511 //go:linkname __go_string_to_byte_array __go_string_to_byte_array
 512 func __go_string_to_byte_array(s string) []byte {
 513         return stringtoslicebyte(nil, s)
 514 }
 515
 516 // parseRelease parses a dot-separated version number. It follows the
 517 // semver syntax, but allows the minor and patch versions to be
 518 // elided.
 519 func parseRelease(rel string) (major, minor, patch int, ok bool) {
 520         // Strip anything after a dash or plus.
 521         for i := 0; i < len(rel); i++ {
 522                 if rel[i] == '-' || rel[i] == '+' {
 523                         rel = rel[:i]
 524                         break
 525                 }
 526         }
 527
 528         next := func() (int, bool) {
 529                 for i := 0; i < len(rel); i++ {
 530                         if rel[i] == '.' {
 531                                 ver, ok := atoi(rel[:i])
 532                                 rel = rel[i+1:]
 533                                 return ver, ok
 534                         }
 535                 }
 536                 ver, ok := atoi(rel)
 537                 rel = ""
 538                 return ver, ok
 539         }
 540         if major, ok = next(); !ok || rel == "" {
 541                 return
 542         }
 543         if minor, ok = next(); !ok || rel == "" {
 544                 return
 545         }
 546         patch, ok = next()
 547         return
 548 }