2 * Copyright 2016 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #include <folly/String.h>
19 #include <folly/Format.h>
20 #include <folly/ScopeGuard.h>
29 #include <glog/logging.h>
35 int stringAppendfImplHelper(char* buf,
40 va_copy(args_copy, args);
41 int bytes_used = vsnprintf(buf, bufsize, format, args_copy);
46 void stringAppendfImpl(std::string& output, const char* format, va_list args) {
47 // Very simple; first, try to avoid an allocation by using an inline
48 // buffer. If that fails to hold the output string, allocate one on
49 // the heap, use it instead.
51 // It is hard to guess the proper size of this buffer; some
52 // heuristics could be based on the number of format characters, or
53 // static analysis of a codebase. Or, we can just pick a number
54 // that seems big enough for simple cases (say, one line of text on
55 // a terminal) without being large enough to be concerning as a
57 std::array<char, 128> inline_buffer;
59 int bytes_used = stringAppendfImplHelper(
60 inline_buffer.data(), inline_buffer.size(), format, args);
62 throw std::runtime_error(to<std::string>(
63 "Invalid format string; snprintf returned negative "
64 "with format string: ",
68 if (static_cast<size_t>(bytes_used) < inline_buffer.size()) {
69 output.append(inline_buffer.data(), bytes_used);
73 // Couldn't fit. Heap allocate a buffer, oh well.
74 std::unique_ptr<char[]> heap_buffer(new char[bytes_used + 1]);
75 int final_bytes_used =
76 stringAppendfImplHelper(heap_buffer.get(), bytes_used + 1, format, args);
77 // The second call can take fewer bytes if, for example, we were printing a
78 // string buffer with null-terminating char using a width specifier -
79 // vsnprintf("%.*s", buf.size(), buf)
80 CHECK(bytes_used >= final_bytes_used);
82 // We don't keep the trailing '\0' in our output string
83 output.append(heap_buffer.get(), final_bytes_used);
88 std::string stringPrintf(const char* format, ...) {
94 return stringVPrintf(format, ap);
97 std::string stringVPrintf(const char* format, va_list ap) {
99 stringAppendfImpl(ret, format, ap);
103 // Basic declarations; allow for parameters of strings and string
104 // pieces to be specified.
105 std::string& stringAppendf(std::string* output, const char* format, ...) {
107 va_start(ap, format);
111 return stringVAppendf(output, format, ap);
114 std::string& stringVAppendf(std::string* output,
117 stringAppendfImpl(*output, format, ap);
121 void stringPrintf(std::string* output, const char* format, ...) {
123 va_start(ap, format);
127 return stringVPrintf(output, format, ap);
130 void stringVPrintf(std::string* output, const char* format, va_list ap) {
132 stringAppendfImpl(*output, format, ap);
137 struct PrettySuffix {
142 const PrettySuffix kPrettyTimeSuffixes[] = {
152 const PrettySuffix kPrettyBytesMetricSuffixes[] = {
161 const PrettySuffix kPrettyBytesBinarySuffixes[] = {
162 { "TB", int64_t(1) << 40 },
163 { "GB", int64_t(1) << 30 },
164 { "MB", int64_t(1) << 20 },
165 { "kB", int64_t(1) << 10 },
170 const PrettySuffix kPrettyBytesBinaryIECSuffixes[] = {
171 { "TiB", int64_t(1) << 40 },
172 { "GiB", int64_t(1) << 30 },
173 { "MiB", int64_t(1) << 20 },
174 { "KiB", int64_t(1) << 10 },
179 const PrettySuffix kPrettyUnitsMetricSuffixes[] = {
188 const PrettySuffix kPrettyUnitsBinarySuffixes[] = {
189 { "T", int64_t(1) << 40 },
190 { "G", int64_t(1) << 30 },
191 { "M", int64_t(1) << 20 },
192 { "k", int64_t(1) << 10 },
197 const PrettySuffix kPrettyUnitsBinaryIECSuffixes[] = {
198 { "Ti", int64_t(1) << 40 },
199 { "Gi", int64_t(1) << 30 },
200 { "Mi", int64_t(1) << 20 },
201 { "Ki", int64_t(1) << 10 },
206 const PrettySuffix kPrettySISuffixes[] = {
231 const PrettySuffix* const kPrettySuffixes[PRETTY_NUM_TYPES] = {
233 kPrettyBytesMetricSuffixes,
234 kPrettyBytesBinarySuffixes,
235 kPrettyBytesBinaryIECSuffixes,
236 kPrettyUnitsMetricSuffixes,
237 kPrettyUnitsBinarySuffixes,
238 kPrettyUnitsBinaryIECSuffixes,
244 std::string prettyPrint(double val, PrettyType type, bool addSpace) {
247 // pick the suffixes to use
249 assert(type < PRETTY_NUM_TYPES);
250 const PrettySuffix* suffixes = kPrettySuffixes[type];
252 // find the first suffix we're bigger than -- then use it
253 double abs_val = fabs(val);
254 for (int i = 0; suffixes[i].suffix; ++i) {
255 if (abs_val >= suffixes[i].val) {
256 snprintf(buf, sizeof buf, "%.4g%s%s",
257 (suffixes[i].val ? (val / suffixes[i].val)
259 (addSpace ? " " : ""),
261 return std::string(buf);
265 // no suffix, we've got a tiny value -- just print it in sci-notation
266 snprintf(buf, sizeof buf, "%.4g", val);
267 return std::string(buf);
271 //1) Benchmark & optimize
272 double prettyToDouble(folly::StringPiece *const prettyString,
273 const PrettyType type) {
274 double value = folly::to<double>(prettyString);
275 while (prettyString->size() > 0 && std::isspace(prettyString->front())) {
276 prettyString->advance(1); //Skipping spaces between number and suffix
278 const PrettySuffix* suffixes = kPrettySuffixes[type];
279 int longestPrefixLen = -1;
280 int bestPrefixId = -1;
281 for (int j = 0 ; suffixes[j].suffix; ++j) {
282 if (suffixes[j].suffix[0] == ' '){//Checking for " " -> number rule.
283 if (longestPrefixLen == -1) {
284 longestPrefixLen = 0; //No characters to skip
287 } else if (prettyString->startsWith(suffixes[j].suffix)) {
288 int suffixLen = strlen(suffixes[j].suffix);
289 //We are looking for a longest suffix matching prefix of the string
290 //after numeric value. We need this in case suffixes have common prefix.
291 if (suffixLen > longestPrefixLen) {
292 longestPrefixLen = suffixLen;
297 if (bestPrefixId == -1) { //No valid suffix rule found
298 throw std::invalid_argument(folly::to<std::string>(
299 "Unable to parse suffix \"",
300 prettyString->toString(), "\""));
302 prettyString->advance(longestPrefixLen);
303 return suffixes[bestPrefixId].val ? value * suffixes[bestPrefixId].val :
307 double prettyToDouble(folly::StringPiece prettyString, const PrettyType type){
308 double result = prettyToDouble(&prettyString, type);
309 detail::enforceWhitespace(prettyString);
313 std::string hexDump(const void* ptr, size_t size) {
314 std::ostringstream os;
315 hexDump(ptr, size, std::ostream_iterator<StringPiece>(os, "\n"));
319 fbstring errnoStr(int err) {
320 int savedErrno = errno;
322 // Ensure that we reset errno upon exit.
323 auto guard(makeGuard([&] { errno = savedErrno; }));
330 // https://developer.apple.com/library/mac/documentation/Darwin/Reference/ManPages/man3/strerror_r.3.html
331 // http://www.kernel.org/doc/man-pages/online/pages/man3/strerror.3.html
332 #if defined(_WIN32) && (defined(__MINGW32__) || defined(_MSC_VER))
333 // mingw64 has no strerror_r, but Windows has strerror_s, which C11 added
334 // as well. So maybe we should use this across all platforms (together
335 // with strerrorlen_s). Note strerror_r and _s have swapped args.
336 int r = strerror_s(buf, sizeof(buf), err);
338 result = to<fbstring>(
339 "Unknown error ", err,
340 " (strerror_r failed with error ", errno, ")");
344 #elif defined(FOLLY_HAVE_XSI_STRERROR_R) || \
345 defined(__APPLE__) || defined(__ANDROID__)
346 // Using XSI-compatible strerror_r
347 int r = strerror_r(err, buf, sizeof(buf));
349 // OSX/FreeBSD use EINVAL and Linux uses -1 so just check for non-zero
351 result = to<fbstring>(
352 "Unknown error ", err,
353 " (strerror_r failed with error ", errno, ")");
358 // Using GNU strerror_r
359 result.assign(strerror_r(err, buf, sizeof(buf)));
367 void toLowerAscii8(char& c) {
368 // Branchless tolower, based on the input-rotating trick described
369 // at http://www.azillionmonkeys.com/qed/asmexample.html
371 // This algorithm depends on an observation: each uppercase
372 // ASCII character can be converted to its lowercase equivalent
375 // Step 1: Clear the high order bit. We'll deal with it in Step 5.
376 unsigned char rotated = c & 0x7f;
377 // Currently, the value of rotated, as a function of the original c is:
382 // Step 2: Add 0x25 (37)
384 // Now the value of rotated, as a function of the original c is:
387 // above 'Z': 128-164
389 // Step 3: clear the high order bit
395 // Step 4: Add 0x1a (26)
401 // At this point, note that only the uppercase letters have been
402 // transformed into values with the high order bit set (128 and above).
404 // Step 5: Shift the high order bit 2 spaces to the right: the spot
405 // where the only 1 bit in 0x20 is. But first, how we ignored the
406 // high order bit of the original c in step 1? If that bit was set,
407 // we may have just gotten a false match on a value in the range
408 // 128+'A' to 128+'Z'. To correct this, need to clear the high order
409 // bit of rotated if the high order bit of c is set. Since we don't
410 // care about the other bits in rotated, the easiest thing to do
411 // is invert all the bits in c and bitwise-and them with rotated.
415 // Step 6: Apply a mask to clear everything except the 0x20 bit
419 // At this point, rotated is 0x20 if c is 'A'-'Z' and 0x00 otherwise
421 // Step 7: Add rotated to c
425 void toLowerAscii32(uint32_t& c) {
426 // Besides being branchless, the algorithm in toLowerAscii8() has another
427 // interesting property: None of the addition operations will cause
428 // an overflow in the 8-bit value. So we can pack four 8-bit values
429 // into a uint32_t and run each operation on all four values in parallel
430 // without having to use any CPU-specific SIMD instructions.
431 uint32_t rotated = c & uint32_t(0x7f7f7f7fL);
432 rotated += uint32_t(0x25252525L);
433 rotated &= uint32_t(0x7f7f7f7fL);
434 rotated += uint32_t(0x1a1a1a1aL);
436 // Step 5 involves a shift, so some bits will spill over from each
437 // 8-bit value into the next. But that's okay, because they're bits
438 // that will be cleared by the mask in step 6 anyway.
441 rotated &= uint32_t(0x20202020L);
445 void toLowerAscii64(uint64_t& c) {
446 // 64-bit version of toLower32
447 uint64_t rotated = c & uint64_t(0x7f7f7f7f7f7f7f7fL);
448 rotated += uint64_t(0x2525252525252525L);
449 rotated &= uint64_t(0x7f7f7f7f7f7f7f7fL);
450 rotated += uint64_t(0x1a1a1a1a1a1a1a1aL);
453 rotated &= uint64_t(0x2020202020202020L);
459 void toLowerAscii(char* str, size_t length) {
460 static const size_t kAlignMask64 = 7;
461 static const size_t kAlignMask32 = 3;
463 // Convert a character at a time until we reach an address that
464 // is at least 32-bit aligned
465 size_t n = (size_t)str;
467 n = std::min(n, length);
470 n = std::min(4 - n, length);
472 toLowerAscii8(str[offset]);
474 } while (offset < n);
477 n = (size_t)(str + offset);
479 if ((n != 0) && (offset + 4 <= length)) {
480 // The next address is 32-bit aligned but not 64-bit aligned.
481 // Convert the next 4 bytes in order to get to the 64-bit aligned
482 // part of the input.
483 toLowerAscii32(*(uint32_t*)(str + offset));
487 // Convert 8 characters at a time
488 while (offset + 8 <= length) {
489 toLowerAscii64(*(uint64_t*)(str + offset));
493 // Convert 4 characters at a time
494 while (offset + 4 <= length) {
495 toLowerAscii32(*(uint32_t*)(str + offset));
499 // Convert any characters remaining after the last 4-byte aligned group
500 while (offset < length) {
501 toLowerAscii8(str[offset]);
508 size_t hexDumpLine(const void* ptr, size_t offset, size_t size,
513 // (1+2)*16: hex bytes, each preceded by a space
514 // 1: space separating the two halves
521 const uint8_t* p = reinterpret_cast<const uint8_t*>(ptr) + offset;
522 size_t n = std::min(size - offset, size_t(16));
523 format("{:08x} ", offset).appendTo(line);
525 for (size_t i = 0; i < n; i++) {
529 format(" {:02x}", p[i]).appendTo(line);
532 // 3 spaces for each byte we're not printing, one separating the halves
534 line.append(3 * (16 - n) + (n <= 8), ' ');
537 for (size_t i = 0; i < n; i++) {
538 char c = (p[i] >= 32 && p[i] <= 126 ? static_cast<char>(p[i]) : '.');
541 line.append(16 - n, ' ');
543 DCHECK_EQ(line.size(), 78);
548 } // namespace detail
550 std::string stripLeftMargin(std::string s) {
551 std::vector<StringPiece> pieces;
552 split("\n", s, pieces);
553 auto piecer = range(pieces);
555 auto piece = (piecer.end() - 1);
556 auto needle = std::find_if(piece->begin(),
558 [](char c) { return c != ' ' && c != '\t'; });
559 if (needle == piece->end()) {
560 (piecer.end() - 1)->clear();
562 piece = piecer.begin();
563 needle = std::find_if(piece->begin(),
565 [](char c) { return c != ' ' && c != '\t'; });
566 if (needle == piece->end()) {
567 piecer.erase(piecer.begin(), piecer.begin() + 1);
570 const auto sentinel = std::numeric_limits<size_t>::max();
571 auto indent = sentinel;
572 size_t max_length = 0;
573 for (piece = piecer.begin(); piece != piecer.end(); piece++) {
574 needle = std::find_if(piece->begin(),
576 [](char c) { return c != ' ' && c != '\t'; });
577 if (needle != piece->end()) {
578 indent = std::min<size_t>(indent, needle - piece->begin());
580 max_length = std::max<size_t>(piece->size(), max_length);
583 indent = indent == sentinel ? max_length : indent;
584 for (piece = piecer.begin(); piece != piecer.end(); piece++) {
585 if (piece->size() < indent) {
588 piece->erase(piece->begin(), piece->begin() + indent);
591 return join("\n", piecer);
596 #ifdef FOLLY_DEFINED_DMGL
597 # undef FOLLY_DEFINED_DMGL
604 # undef DMGL_RET_POSTFIX