1// Copyright (c) 2012 The Chromium Authors. All rights reserved. 2// Use of this source code is governed by a BSD-style license that can be 3// found in the LICENSE file. 4 5// FilePath is a container for pathnames stored in a platform's native string 6// type, providing containers for manipulation in according with the 7// platform's conventions for pathnames. It supports the following path 8// types: 9// 10// POSIX Windows 11// --------------- ---------------------------------- 12// Fundamental type char[] wchar_t[] 13// Encoding unspecified* UTF-16 14// Separator / \, tolerant of / 15// Drive letters no case-insensitive A-Z followed by : 16// Alternate root // (surprise!) \\, for UNC paths 17// 18// * The encoding need not be specified on POSIX systems, although some 19// POSIX-compliant systems do specify an encoding. Mac OS X uses UTF-8. 20// Chrome OS also uses UTF-8. 21// Linux does not specify an encoding, but in practice, the locale's 22// character set may be used. 23// 24// For more arcane bits of path trivia, see below. 25// 26// FilePath objects are intended to be used anywhere paths are. An 27// application may pass FilePath objects around internally, masking the 28// underlying differences between systems, only differing in implementation 29// where interfacing directly with the system. For example, a single 30// OpenFile(const FilePath &) function may be made available, allowing all 31// callers to operate without regard to the underlying implementation. On 32// POSIX-like platforms, OpenFile might wrap fopen, and on Windows, it might 33// wrap _wfopen_s, perhaps both by calling file_path.value().c_str(). This 34// allows each platform to pass pathnames around without requiring conversions 35// between encodings, which has an impact on performance, but more imporantly, 36// has an impact on correctness on platforms that do not have well-defined 37// encodings for pathnames. 38// 39// Several methods are available to perform common operations on a FilePath 40// object, such as determining the parent directory (DirName), isolating the 41// final path component (BaseName), and appending a relative pathname string 42// to an existing FilePath object (Append). These methods are highly 43// recommended over attempting to split and concatenate strings directly. 44// These methods are based purely on string manipulation and knowledge of 45// platform-specific pathname conventions, and do not consult the filesystem 46// at all, making them safe to use without fear of blocking on I/O operations. 47// These methods do not function as mutators but instead return distinct 48// instances of FilePath objects, and are therefore safe to use on const 49// objects. The objects themselves are safe to share between threads. 50// 51// To aid in initialization of FilePath objects from string literals, a 52// FILE_PATH_LITERAL macro is provided, which accounts for the difference 53// between char[]-based pathnames on POSIX systems and wchar_t[]-based 54// pathnames on Windows. 55// 56// As a precaution against premature truncation, paths can't contain NULs. 57// 58// Because a FilePath object should not be instantiated at the global scope, 59// instead, use a FilePath::CharType[] and initialize it with 60// FILE_PATH_LITERAL. At runtime, a FilePath object can be created from the 61// character array. Example: 62// 63// | const FilePath::CharType kLogFileName[] = FILE_PATH_LITERAL("log.txt"); 64// | 65// | void Function() { 66// | FilePath log_file_path(kLogFileName); 67// | [...] 68// | } 69// 70// WARNING: FilePaths should ALWAYS be displayed with LTR directionality, even 71// when the UI language is RTL. This means you always need to pass filepaths 72// through base::i18n::WrapPathWithLTRFormatting() before displaying it in the 73// RTL UI. 74// 75// This is a very common source of bugs, please try to keep this in mind. 76// 77// ARCANE BITS OF PATH TRIVIA 78// 79// - A double leading slash is actually part of the POSIX standard. Systems 80// are allowed to treat // as an alternate root, as Windows does for UNC 81// (network share) paths. Most POSIX systems don't do anything special 82// with two leading slashes, but FilePath handles this case properly 83// in case it ever comes across such a system. FilePath needs this support 84// for Windows UNC paths, anyway. 85// References: 86// The Open Group Base Specifications Issue 7, sections 3.267 ("Pathname") 87// and 4.12 ("Pathname Resolution"), available at: 88// http://www.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap03.html#tag_03_267 89// http://www.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap04.html#tag_04_12 90// 91// - Windows treats c:\\ the same way it treats \\. This was intended to 92// allow older applications that require drive letters to support UNC paths 93// like \\server\share\path, by permitting c:\\server\share\path as an 94// equivalent. Since the OS treats these paths specially, FilePath needs 95// to do the same. Since Windows can use either / or \ as the separator, 96// FilePath treats c://, c:\\, //, and \\ all equivalently. 97// Reference: 98// The Old New Thing, "Why is a drive letter permitted in front of UNC 99// paths (sometimes)?", available at: 100// http://blogs.msdn.com/oldnewthing/archive/2005/11/22/495740.aspx 101 102#ifndef BASE_FILES_FILE_PATH_H_ 103#define BASE_FILES_FILE_PATH_H_ 104 105#include <stddef.h> 106 107#include <iosfwd> 108#include <string> 109#include <vector> 110 111#include "base/base_export.h" 112#include "base/compiler_specific.h" 113#include "base/containers/hash_tables.h" 114#include "base/macros.h" 115#include "base/strings/string16.h" 116#include "base/strings/string_piece.h" 117#include "build/build_config.h" 118 119// Windows-style drive letter support and pathname separator characters can be 120// enabled and disabled independently, to aid testing. These #defines are 121// here so that the same setting can be used in both the implementation and 122// in the unit test. 123#if defined(OS_WIN) 124#define FILE_PATH_USES_DRIVE_LETTERS 125#define FILE_PATH_USES_WIN_SEPARATORS 126#endif // OS_WIN 127 128// To print path names portably use PRIsFP (based on PRIuS and friends from 129// C99 and format_macros.h) like this: 130// base::StringPrintf("Path is %" PRIsFP ".\n", path.value().c_str()); 131#if defined(OS_POSIX) 132#define PRIsFP "s" 133#elif defined(OS_WIN) 134#define PRIsFP "ls" 135#endif // OS_WIN 136 137namespace base { 138 139class Pickle; 140class PickleIterator; 141 142// An abstraction to isolate users from the differences between native 143// pathnames on different platforms. 144class BASE_EXPORT FilePath { 145 public: 146#if defined(OS_POSIX) 147 // On most platforms, native pathnames are char arrays, and the encoding 148 // may or may not be specified. On Mac OS X, native pathnames are encoded 149 // in UTF-8. 150 typedef std::string StringType; 151#elif defined(OS_WIN) 152 // On Windows, for Unicode-aware applications, native pathnames are wchar_t 153 // arrays encoded in UTF-16. 154 typedef std::wstring StringType; 155#endif // OS_WIN 156 157 typedef BasicStringPiece<StringType> StringPieceType; 158 typedef StringType::value_type CharType; 159 160 // Null-terminated array of separators used to separate components in 161 // hierarchical paths. Each character in this array is a valid separator, 162 // but kSeparators[0] is treated as the canonical separator and will be used 163 // when composing pathnames. 164 static const CharType kSeparators[]; 165 166 // arraysize(kSeparators). 167 static const size_t kSeparatorsLength; 168 169 // A special path component meaning "this directory." 170 static const CharType kCurrentDirectory[]; 171 172 // A special path component meaning "the parent directory." 173 static const CharType kParentDirectory[]; 174 175 // The character used to identify a file extension. 176 static const CharType kExtensionSeparator; 177 178 FilePath(); 179 FilePath(const FilePath& that); 180 explicit FilePath(StringPieceType path); 181 ~FilePath(); 182 FilePath& operator=(const FilePath& that); 183 184 bool operator==(const FilePath& that) const; 185 186 bool operator!=(const FilePath& that) const; 187 188 // Required for some STL containers and operations 189 bool operator<(const FilePath& that) const { 190 return path_ < that.path_; 191 } 192 193 const StringType& value() const { return path_; } 194 195 bool empty() const { return path_.empty(); } 196 197 void clear() { path_.clear(); } 198 199 // Returns true if |character| is in kSeparators. 200 static bool IsSeparator(CharType character); 201 202 // Returns a vector of all of the components of the provided path. It is 203 // equivalent to calling DirName().value() on the path's root component, 204 // and BaseName().value() on each child component. 205 // 206 // To make sure this is lossless so we can differentiate absolute and 207 // relative paths, the root slash will be included even though no other 208 // slashes will be. The precise behavior is: 209 // 210 // Posix: "/foo/bar" -> [ "/", "foo", "bar" ] 211 // Windows: "C:\foo\bar" -> [ "C:", "\\", "foo", "bar" ] 212 void GetComponents(std::vector<FilePath::StringType>* components) const; 213 214 // Returns true if this FilePath is a strict parent of the |child|. Absolute 215 // and relative paths are accepted i.e. is /foo parent to /foo/bar and 216 // is foo parent to foo/bar. Does not convert paths to absolute, follow 217 // symlinks or directory navigation (e.g. ".."). A path is *NOT* its own 218 // parent. 219 bool IsParent(const FilePath& child) const; 220 221 // If IsParent(child) holds, appends to path (if non-NULL) the 222 // relative path to child and returns true. For example, if parent 223 // holds "/Users/johndoe/Library/Application Support", child holds 224 // "/Users/johndoe/Library/Application Support/Google/Chrome/Default", and 225 // *path holds "/Users/johndoe/Library/Caches", then after 226 // parent.AppendRelativePath(child, path) is called *path will hold 227 // "/Users/johndoe/Library/Caches/Google/Chrome/Default". Otherwise, 228 // returns false. 229 bool AppendRelativePath(const FilePath& child, FilePath* path) const; 230 231 // Returns a FilePath corresponding to the directory containing the path 232 // named by this object, stripping away the file component. If this object 233 // only contains one component, returns a FilePath identifying 234 // kCurrentDirectory. If this object already refers to the root directory, 235 // returns a FilePath identifying the root directory. 236 FilePath DirName() const WARN_UNUSED_RESULT; 237 238 // Returns a FilePath corresponding to the last path component of this 239 // object, either a file or a directory. If this object already refers to 240 // the root directory, returns a FilePath identifying the root directory; 241 // this is the only situation in which BaseName will return an absolute path. 242 FilePath BaseName() const WARN_UNUSED_RESULT; 243 244 // Returns ".jpg" for path "C:\pics\jojo.jpg", or an empty string if 245 // the file has no extension. If non-empty, Extension() will always start 246 // with precisely one ".". The following code should always work regardless 247 // of the value of path. For common double-extensions like .tar.gz and 248 // .user.js, this method returns the combined extension. For a single 249 // component, use FinalExtension(). 250 // new_path = path.RemoveExtension().value().append(path.Extension()); 251 // ASSERT(new_path == path.value()); 252 // NOTE: this is different from the original file_util implementation which 253 // returned the extension without a leading "." ("jpg" instead of ".jpg") 254 StringType Extension() const WARN_UNUSED_RESULT; 255 256 // Returns the path's file extension, as in Extension(), but will 257 // never return a double extension. 258 // 259 // TODO(davidben): Check all our extension-sensitive code to see if 260 // we can rename this to Extension() and the other to something like 261 // LongExtension(), defaulting to short extensions and leaving the 262 // long "extensions" to logic like base::GetUniquePathNumber(). 263 StringType FinalExtension() const WARN_UNUSED_RESULT; 264 265 // Returns "C:\pics\jojo" for path "C:\pics\jojo.jpg" 266 // NOTE: this is slightly different from the similar file_util implementation 267 // which returned simply 'jojo'. 268 FilePath RemoveExtension() const WARN_UNUSED_RESULT; 269 270 // Removes the path's file extension, as in RemoveExtension(), but 271 // ignores double extensions. 272 FilePath RemoveFinalExtension() const WARN_UNUSED_RESULT; 273 274 // Inserts |suffix| after the file name portion of |path| but before the 275 // extension. Returns "" if BaseName() == "." or "..". 276 // Examples: 277 // path == "C:\pics\jojo.jpg" suffix == " (1)", returns "C:\pics\jojo (1).jpg" 278 // path == "jojo.jpg" suffix == " (1)", returns "jojo (1).jpg" 279 // path == "C:\pics\jojo" suffix == " (1)", returns "C:\pics\jojo (1)" 280 // path == "C:\pics.old\jojo" suffix == " (1)", returns "C:\pics.old\jojo (1)" 281 FilePath InsertBeforeExtension( 282 StringPieceType suffix) const WARN_UNUSED_RESULT; 283 FilePath InsertBeforeExtensionASCII( 284 StringPiece suffix) const WARN_UNUSED_RESULT; 285 286 // Adds |extension| to |file_name|. Returns the current FilePath if 287 // |extension| is empty. Returns "" if BaseName() == "." or "..". 288 FilePath AddExtension(StringPieceType extension) const WARN_UNUSED_RESULT; 289 290 // Replaces the extension of |file_name| with |extension|. If |file_name| 291 // does not have an extension, then |extension| is added. If |extension| is 292 // empty, then the extension is removed from |file_name|. 293 // Returns "" if BaseName() == "." or "..". 294 FilePath ReplaceExtension(StringPieceType extension) const WARN_UNUSED_RESULT; 295 296 // Returns true if the file path matches the specified extension. The test is 297 // case insensitive. Don't forget the leading period if appropriate. 298 bool MatchesExtension(StringPieceType extension) const; 299 300 // Returns a FilePath by appending a separator and the supplied path 301 // component to this object's path. Append takes care to avoid adding 302 // excessive separators if this object's path already ends with a separator. 303 // If this object's path is kCurrentDirectory, a new FilePath corresponding 304 // only to |component| is returned. |component| must be a relative path; 305 // it is an error to pass an absolute path. 306 FilePath Append(StringPieceType component) const WARN_UNUSED_RESULT; 307 FilePath Append(const FilePath& component) const WARN_UNUSED_RESULT; 308 309 // Although Windows StringType is std::wstring, since the encoding it uses for 310 // paths is well defined, it can handle ASCII path components as well. 311 // Mac uses UTF8, and since ASCII is a subset of that, it works there as well. 312 // On Linux, although it can use any 8-bit encoding for paths, we assume that 313 // ASCII is a valid subset, regardless of the encoding, since many operating 314 // system paths will always be ASCII. 315 FilePath AppendASCII(StringPiece component) const WARN_UNUSED_RESULT; 316 317 // Returns true if this FilePath contains an absolute path. On Windows, an 318 // absolute path begins with either a drive letter specification followed by 319 // a separator character, or with two separator characters. On POSIX 320 // platforms, an absolute path begins with a separator character. 321 bool IsAbsolute() const; 322 323 // Returns true if the patch ends with a path separator character. 324 bool EndsWithSeparator() const WARN_UNUSED_RESULT; 325 326 // Returns a copy of this FilePath that ends with a trailing separator. If 327 // the input path is empty, an empty FilePath will be returned. 328 FilePath AsEndingWithSeparator() const WARN_UNUSED_RESULT; 329 330 // Returns a copy of this FilePath that does not end with a trailing 331 // separator. 332 FilePath StripTrailingSeparators() const WARN_UNUSED_RESULT; 333 334 // Returns true if this FilePath contains an attempt to reference a parent 335 // directory (e.g. has a path component that is ".."). 336 bool ReferencesParent() const; 337 338 // Return a Unicode human-readable version of this path. 339 // Warning: you can *not*, in general, go from a display name back to a real 340 // path. Only use this when displaying paths to users, not just when you 341 // want to stuff a string16 into some other API. 342 string16 LossyDisplayName() const; 343 344 // Return the path as ASCII, or the empty string if the path is not ASCII. 345 // This should only be used for cases where the FilePath is representing a 346 // known-ASCII filename. 347 std::string MaybeAsASCII() const; 348 349 // Return the path as UTF-8. 350 // 351 // This function is *unsafe* as there is no way to tell what encoding is 352 // used in file names on POSIX systems other than Mac and Chrome OS, 353 // although UTF-8 is practically used everywhere these days. To mitigate 354 // the encoding issue, this function internally calls 355 // SysNativeMBToWide() on POSIX systems other than Mac and Chrome OS, 356 // per assumption that the current locale's encoding is used in file 357 // names, but this isn't a perfect solution. 358 // 359 // Once it becomes safe to to stop caring about non-UTF-8 file names, 360 // the SysNativeMBToWide() hack will be removed from the code, along 361 // with "Unsafe" in the function name. 362 std::string AsUTF8Unsafe() const; 363 364 // Similar to AsUTF8Unsafe, but returns UTF-16 instead. 365 string16 AsUTF16Unsafe() const; 366 367 // Returns a FilePath object from a path name in UTF-8. This function 368 // should only be used for cases where you are sure that the input 369 // string is UTF-8. 370 // 371 // Like AsUTF8Unsafe(), this function is unsafe. This function 372 // internally calls SysWideToNativeMB() on POSIX systems other than Mac 373 // and Chrome OS, to mitigate the encoding issue. See the comment at 374 // AsUTF8Unsafe() for details. 375 static FilePath FromUTF8Unsafe(const std::string& utf8); 376 377 // Similar to FromUTF8Unsafe, but accepts UTF-16 instead. 378 static FilePath FromUTF16Unsafe(const string16& utf16); 379 380 void WriteToPickle(Pickle* pickle) const; 381 bool ReadFromPickle(PickleIterator* iter); 382 383 // Normalize all path separators to backslash on Windows 384 // (if FILE_PATH_USES_WIN_SEPARATORS is true), or do nothing on POSIX systems. 385 FilePath NormalizePathSeparators() const; 386 387 // Normalize all path separattors to given type on Windows 388 // (if FILE_PATH_USES_WIN_SEPARATORS is true), or do nothing on POSIX systems. 389 FilePath NormalizePathSeparatorsTo(CharType separator) const; 390 391 // Compare two strings in the same way the file system does. 392 // Note that these always ignore case, even on file systems that are case- 393 // sensitive. If case-sensitive comparison is ever needed, add corresponding 394 // methods here. 395 // The methods are written as a static method so that they can also be used 396 // on parts of a file path, e.g., just the extension. 397 // CompareIgnoreCase() returns -1, 0 or 1 for less-than, equal-to and 398 // greater-than respectively. 399 static int CompareIgnoreCase(StringPieceType string1, 400 StringPieceType string2); 401 static bool CompareEqualIgnoreCase(StringPieceType string1, 402 StringPieceType string2) { 403 return CompareIgnoreCase(string1, string2) == 0; 404 } 405 static bool CompareLessIgnoreCase(StringPieceType string1, 406 StringPieceType string2) { 407 return CompareIgnoreCase(string1, string2) < 0; 408 } 409 410#if defined(OS_MACOSX) 411 // Returns the string in the special canonical decomposed form as defined for 412 // HFS, which is close to, but not quite, decomposition form D. See 413 // http://developer.apple.com/mac/library/technotes/tn/tn1150.html#UnicodeSubtleties 414 // for further comments. 415 // Returns the epmty string if the conversion failed. 416 static StringType GetHFSDecomposedForm(StringPieceType string); 417 418 // Special UTF-8 version of FastUnicodeCompare. Cf: 419 // http://developer.apple.com/mac/library/technotes/tn/tn1150.html#StringComparisonAlgorithm 420 // IMPORTANT: The input strings must be in the special HFS decomposed form! 421 // (cf. above GetHFSDecomposedForm method) 422 static int HFSFastUnicodeCompare(StringPieceType string1, 423 StringPieceType string2); 424#endif 425 426#if defined(OS_ANDROID) 427 // On android, file selection dialog can return a file with content uri 428 // scheme(starting with content://). Content uri needs to be opened with 429 // ContentResolver to guarantee that the app has appropriate permissions 430 // to access it. 431 // Returns true if the path is a content uri, or false otherwise. 432 bool IsContentUri() const; 433#endif 434 435 private: 436 // Remove trailing separators from this object. If the path is absolute, it 437 // will never be stripped any more than to refer to the absolute root 438 // directory, so "////" will become "/", not "". A leading pair of 439 // separators is never stripped, to support alternate roots. This is used to 440 // support UNC paths on Windows. 441 void StripTrailingSeparatorsInternal(); 442 443 StringType path_; 444}; 445 446// This is required by googletest to print a readable output on test failures. 447// This is declared here for use in gtest-based unit tests but is defined in 448// the test_support_base target. Depend on that to use this in your unit test. 449// This should not be used in production code - call ToString() instead. 450void PrintTo(const FilePath& path, std::ostream* out); 451 452} // namespace base 453 454// Macros for string literal initialization of FilePath::CharType[], and for 455// using a FilePath::CharType[] in a printf-style format string. 456#if defined(OS_POSIX) 457#define FILE_PATH_LITERAL(x) x 458#define PRFilePath "s" 459#elif defined(OS_WIN) 460#define FILE_PATH_LITERAL(x) L ## x 461#define PRFilePath "ls" 462#endif // OS_WIN 463 464// Provide a hash function so that hash_sets and maps can contain FilePath 465// objects. 466namespace BASE_HASH_NAMESPACE { 467 468template<> 469struct hash<base::FilePath> { 470 size_t operator()(const base::FilePath& f) const { 471 return hash<base::FilePath::StringType>()(f.value()); 472 } 473}; 474 475} // namespace BASE_HASH_NAMESPACE 476 477#endif // BASE_FILES_FILE_PATH_H_ 478