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// Paths can't contain NULs as a precaution agaist premature truncation. 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 3.266 ("Pathname") 87// and 4.12 ("Pathname Resolution"), available at: 88// http://www.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap03.html#tag_03_266 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#include <string> 107#include <vector> 108 109#include "base/base_export.h" 110#include "base/compiler_specific.h" 111#include "base/containers/hash_tables.h" 112#include "base/strings/string16.h" 113#include "base/strings/string_piece.h" // For implicit conversions. 114#include "build/build_config.h" 115 116// Windows-style drive letter support and pathname separator characters can be 117// enabled and disabled independently, to aid testing. These #defines are 118// here so that the same setting can be used in both the implementation and 119// in the unit test. 120#if defined(OS_WIN) 121#define FILE_PATH_USES_DRIVE_LETTERS 122#define FILE_PATH_USES_WIN_SEPARATORS 123#endif // OS_WIN 124 125class Pickle; 126class PickleIterator; 127 128namespace base { 129 130// An abstraction to isolate users from the differences between native 131// pathnames on different platforms. 132class BASE_EXPORT FilePath { 133 public: 134#if defined(OS_POSIX) 135 // On most platforms, native pathnames are char arrays, and the encoding 136 // may or may not be specified. On Mac OS X, native pathnames are encoded 137 // in UTF-8. 138 typedef std::string StringType; 139#elif defined(OS_WIN) 140 // On Windows, for Unicode-aware applications, native pathnames are wchar_t 141 // arrays encoded in UTF-16. 142 typedef std::wstring StringType; 143#endif // OS_WIN 144 145 typedef StringType::value_type CharType; 146 147 // Null-terminated array of separators used to separate components in 148 // hierarchical paths. Each character in this array is a valid separator, 149 // but kSeparators[0] is treated as the canonical separator and will be used 150 // when composing pathnames. 151 static const CharType kSeparators[]; 152 153 // arraysize(kSeparators). 154 static const size_t kSeparatorsLength; 155 156 // A special path component meaning "this directory." 157 static const CharType kCurrentDirectory[]; 158 159 // A special path component meaning "the parent directory." 160 static const CharType kParentDirectory[]; 161 162 // The character used to identify a file extension. 163 static const CharType kExtensionSeparator; 164 165 FilePath(); 166 FilePath(const FilePath& that); 167 explicit FilePath(const StringType& path); 168 ~FilePath(); 169 FilePath& operator=(const FilePath& that); 170 171 bool operator==(const FilePath& that) const; 172 173 bool operator!=(const FilePath& that) const; 174 175 // Required for some STL containers and operations 176 bool operator<(const FilePath& that) const { 177 return path_ < that.path_; 178 } 179 180 const StringType& value() const { return path_; } 181 182 bool empty() const { return path_.empty(); } 183 184 void clear() { path_.clear(); } 185 186 // Returns true if |character| is in kSeparators. 187 static bool IsSeparator(CharType character); 188 189 // Returns a vector of all of the components of the provided path. It is 190 // equivalent to calling DirName().value() on the path's root component, 191 // and BaseName().value() on each child component. 192 // 193 // To make sure this is lossless so we can differentiate absolute and 194 // relative paths, the root slash will be included even though no other 195 // slashes will be. The precise behavior is: 196 // 197 // Posix: "/foo/bar" -> [ "/", "foo", "bar" ] 198 // Windows: "C:\foo\bar" -> [ "C:", "\\", "foo", "bar" ] 199 void GetComponents(std::vector<FilePath::StringType>* components) const; 200 201 // Returns true if this FilePath is a strict parent of the |child|. Absolute 202 // and relative paths are accepted i.e. is /foo parent to /foo/bar and 203 // is foo parent to foo/bar. Does not convert paths to absolute, follow 204 // symlinks or directory navigation (e.g. ".."). A path is *NOT* its own 205 // parent. 206 bool IsParent(const FilePath& child) const; 207 208 // If IsParent(child) holds, appends to path (if non-NULL) the 209 // relative path to child and returns true. For example, if parent 210 // holds "/Users/johndoe/Library/Application Support", child holds 211 // "/Users/johndoe/Library/Application Support/Google/Chrome/Default", and 212 // *path holds "/Users/johndoe/Library/Caches", then after 213 // parent.AppendRelativePath(child, path) is called *path will hold 214 // "/Users/johndoe/Library/Caches/Google/Chrome/Default". Otherwise, 215 // returns false. 216 bool AppendRelativePath(const FilePath& child, FilePath* path) const; 217 218 // Returns a FilePath corresponding to the directory containing the path 219 // named by this object, stripping away the file component. If this object 220 // only contains one component, returns a FilePath identifying 221 // kCurrentDirectory. If this object already refers to the root directory, 222 // returns a FilePath identifying the root directory. 223 FilePath DirName() const WARN_UNUSED_RESULT; 224 225 // Returns a FilePath corresponding to the last path component of this 226 // object, either a file or a directory. If this object already refers to 227 // the root directory, returns a FilePath identifying the root directory; 228 // this is the only situation in which BaseName will return an absolute path. 229 FilePath BaseName() const WARN_UNUSED_RESULT; 230 231 // Returns ".jpg" for path "C:\pics\jojo.jpg", or an empty string if 232 // the file has no extension. If non-empty, Extension() will always start 233 // with precisely one ".". The following code should always work regardless 234 // of the value of path. For common double-extensions like .tar.gz and 235 // .user.js, this method returns the combined extension. For a single 236 // component, use FinalExtension(). 237 // new_path = path.RemoveExtension().value().append(path.Extension()); 238 // ASSERT(new_path == path.value()); 239 // NOTE: this is different from the original file_util implementation which 240 // returned the extension without a leading "." ("jpg" instead of ".jpg") 241 StringType Extension() const; 242 243 // Returns the path's file extension, as in Extension(), but will 244 // never return a double extension. 245 // 246 // TODO(davidben): Check all our extension-sensitive code to see if 247 // we can rename this to Extension() and the other to something like 248 // LongExtension(), defaulting to short extensions and leaving the 249 // long "extensions" to logic like base::GetUniquePathNumber(). 250 StringType FinalExtension() const; 251 252 // Returns "C:\pics\jojo" for path "C:\pics\jojo.jpg" 253 // NOTE: this is slightly different from the similar file_util implementation 254 // which returned simply 'jojo'. 255 FilePath RemoveExtension() const WARN_UNUSED_RESULT; 256 257 // Removes the path's file extension, as in RemoveExtension(), but 258 // ignores double extensions. 259 FilePath RemoveFinalExtension() const WARN_UNUSED_RESULT; 260 261 // Inserts |suffix| after the file name portion of |path| but before the 262 // extension. Returns "" if BaseName() == "." or "..". 263 // Examples: 264 // path == "C:\pics\jojo.jpg" suffix == " (1)", returns "C:\pics\jojo (1).jpg" 265 // path == "jojo.jpg" suffix == " (1)", returns "jojo (1).jpg" 266 // path == "C:\pics\jojo" suffix == " (1)", returns "C:\pics\jojo (1)" 267 // path == "C:\pics.old\jojo" suffix == " (1)", returns "C:\pics.old\jojo (1)" 268 FilePath InsertBeforeExtension( 269 const StringType& suffix) const WARN_UNUSED_RESULT; 270 FilePath InsertBeforeExtensionASCII( 271 const base::StringPiece& suffix) const WARN_UNUSED_RESULT; 272 273 // Adds |extension| to |file_name|. Returns the current FilePath if 274 // |extension| is empty. Returns "" if BaseName() == "." or "..". 275 FilePath AddExtension( 276 const StringType& extension) const WARN_UNUSED_RESULT; 277 278 // Replaces the extension of |file_name| with |extension|. If |file_name| 279 // does not have an extension, then |extension| is added. If |extension| is 280 // empty, then the extension is removed from |file_name|. 281 // Returns "" if BaseName() == "." or "..". 282 FilePath ReplaceExtension( 283 const StringType& extension) const WARN_UNUSED_RESULT; 284 285 // Returns true if the file path matches the specified extension. The test is 286 // case insensitive. Don't forget the leading period if appropriate. 287 bool MatchesExtension(const StringType& extension) const; 288 289 // Returns a FilePath by appending a separator and the supplied path 290 // component to this object's path. Append takes care to avoid adding 291 // excessive separators if this object's path already ends with a separator. 292 // If this object's path is kCurrentDirectory, a new FilePath corresponding 293 // only to |component| is returned. |component| must be a relative path; 294 // it is an error to pass an absolute path. 295 FilePath Append(const StringType& component) const WARN_UNUSED_RESULT; 296 FilePath Append(const FilePath& component) const WARN_UNUSED_RESULT; 297 298 // Although Windows StringType is std::wstring, since the encoding it uses for 299 // paths is well defined, it can handle ASCII path components as well. 300 // Mac uses UTF8, and since ASCII is a subset of that, it works there as well. 301 // On Linux, although it can use any 8-bit encoding for paths, we assume that 302 // ASCII is a valid subset, regardless of the encoding, since many operating 303 // system paths will always be ASCII. 304 FilePath AppendASCII(const base::StringPiece& component) 305 const WARN_UNUSED_RESULT; 306 307 // Returns true if this FilePath contains an absolute path. On Windows, an 308 // absolute path begins with either a drive letter specification followed by 309 // a separator character, or with two separator characters. On POSIX 310 // platforms, an absolute path begins with a separator character. 311 bool IsAbsolute() const; 312 313 // Returns true if the patch ends with a path separator character. 314 bool EndsWithSeparator() const WARN_UNUSED_RESULT; 315 316 // Returns a copy of this FilePath that ends with a trailing separator. If 317 // the input path is empty, an empty FilePath will be returned. 318 FilePath AsEndingWithSeparator() const WARN_UNUSED_RESULT; 319 320 // Returns a copy of this FilePath that does not end with a trailing 321 // separator. 322 FilePath StripTrailingSeparators() const WARN_UNUSED_RESULT; 323 324 // Returns true if this FilePath contains an attempt to reference a parent 325 // directory (e.g. has a path component that is ".."). 326 bool ReferencesParent() const; 327 328 // Return a Unicode human-readable version of this path. 329 // Warning: you can *not*, in general, go from a display name back to a real 330 // path. Only use this when displaying paths to users, not just when you 331 // want to stuff a string16 into some other API. 332 string16 LossyDisplayName() const; 333 334 // Return the path as ASCII, or the empty string if the path is not ASCII. 335 // This should only be used for cases where the FilePath is representing a 336 // known-ASCII filename. 337 std::string MaybeAsASCII() const; 338 339 // Return the path as UTF-8. 340 // 341 // This function is *unsafe* as there is no way to tell what encoding is 342 // used in file names on POSIX systems other than Mac and Chrome OS, 343 // although UTF-8 is practically used everywhere these days. To mitigate 344 // the encoding issue, this function internally calls 345 // SysNativeMBToWide() on POSIX systems other than Mac and Chrome OS, 346 // per assumption that the current locale's encoding is used in file 347 // names, but this isn't a perfect solution. 348 // 349 // Once it becomes safe to to stop caring about non-UTF-8 file names, 350 // the SysNativeMBToWide() hack will be removed from the code, along 351 // with "Unsafe" in the function name. 352 std::string AsUTF8Unsafe() const; 353 354 // Similar to AsUTF8Unsafe, but returns UTF-16 instead. 355 string16 AsUTF16Unsafe() const; 356 357 // Returns a FilePath object from a path name in UTF-8. This function 358 // should only be used for cases where you are sure that the input 359 // string is UTF-8. 360 // 361 // Like AsUTF8Unsafe(), this function is unsafe. This function 362 // internally calls SysWideToNativeMB() on POSIX systems other than Mac 363 // and Chrome OS, to mitigate the encoding issue. See the comment at 364 // AsUTF8Unsafe() for details. 365 static FilePath FromUTF8Unsafe(const std::string& utf8); 366 367 // Similar to FromUTF8Unsafe, but accepts UTF-16 instead. 368 static FilePath FromUTF16Unsafe(const string16& utf16); 369 370 void WriteToPickle(Pickle* pickle) const; 371 bool ReadFromPickle(PickleIterator* iter); 372 373 // Normalize all path separators to backslash on Windows 374 // (if FILE_PATH_USES_WIN_SEPARATORS is true), or do nothing on POSIX systems. 375 FilePath NormalizePathSeparators() const; 376 377 // Normalize all path separattors to given type on Windows 378 // (if FILE_PATH_USES_WIN_SEPARATORS is true), or do nothing on POSIX systems. 379 FilePath NormalizePathSeparatorsTo(CharType separator) const; 380 381 // Compare two strings in the same way the file system does. 382 // Note that these always ignore case, even on file systems that are case- 383 // sensitive. If case-sensitive comparison is ever needed, add corresponding 384 // methods here. 385 // The methods are written as a static method so that they can also be used 386 // on parts of a file path, e.g., just the extension. 387 // CompareIgnoreCase() returns -1, 0 or 1 for less-than, equal-to and 388 // greater-than respectively. 389 static int CompareIgnoreCase(const StringType& string1, 390 const StringType& string2); 391 static bool CompareEqualIgnoreCase(const StringType& string1, 392 const StringType& string2) { 393 return CompareIgnoreCase(string1, string2) == 0; 394 } 395 static bool CompareLessIgnoreCase(const StringType& string1, 396 const StringType& string2) { 397 return CompareIgnoreCase(string1, string2) < 0; 398 } 399 400#if defined(OS_MACOSX) 401 // Returns the string in the special canonical decomposed form as defined for 402 // HFS, which is close to, but not quite, decomposition form D. See 403 // http://developer.apple.com/mac/library/technotes/tn/tn1150.html#UnicodeSubtleties 404 // for further comments. 405 // Returns the epmty string if the conversion failed. 406 static StringType GetHFSDecomposedForm(const FilePath::StringType& string); 407 408 // Special UTF-8 version of FastUnicodeCompare. Cf: 409 // http://developer.apple.com/mac/library/technotes/tn/tn1150.html#StringComparisonAlgorithm 410 // IMPORTANT: The input strings must be in the special HFS decomposed form! 411 // (cf. above GetHFSDecomposedForm method) 412 static int HFSFastUnicodeCompare(const StringType& string1, 413 const StringType& string2); 414#endif 415 416#if defined(OS_ANDROID) 417 // On android, file selection dialog can return a file with content uri 418 // scheme(starting with content://). Content uri needs to be opened with 419 // ContentResolver to guarantee that the app has appropriate permissions 420 // to access it. 421 // Returns true if the path is a content uri, or false otherwise. 422 bool IsContentUri() const; 423#endif 424 425 private: 426 // Remove trailing separators from this object. If the path is absolute, it 427 // will never be stripped any more than to refer to the absolute root 428 // directory, so "////" will become "/", not "". A leading pair of 429 // separators is never stripped, to support alternate roots. This is used to 430 // support UNC paths on Windows. 431 void StripTrailingSeparatorsInternal(); 432 433 StringType path_; 434}; 435 436} // namespace base 437 438// This is required by googletest to print a readable output on test failures. 439BASE_EXPORT extern void PrintTo(const base::FilePath& path, std::ostream* out); 440 441// Macros for string literal initialization of FilePath::CharType[], and for 442// using a FilePath::CharType[] in a printf-style format string. 443#if defined(OS_POSIX) 444#define FILE_PATH_LITERAL(x) x 445#define PRFilePath "s" 446#define PRFilePathLiteral "%s" 447#elif defined(OS_WIN) 448#define FILE_PATH_LITERAL(x) L ## x 449#define PRFilePath "ls" 450#define PRFilePathLiteral L"%ls" 451#endif // OS_WIN 452 453// Provide a hash function so that hash_sets and maps can contain FilePath 454// objects. 455namespace BASE_HASH_NAMESPACE { 456#if defined(COMPILER_GCC) 457 458template<> 459struct hash<base::FilePath> { 460 size_t operator()(const base::FilePath& f) const { 461 return hash<base::FilePath::StringType>()(f.value()); 462 } 463}; 464 465#elif defined(COMPILER_MSVC) 466 467inline size_t hash_value(const base::FilePath& f) { 468 return hash_value(f.value()); 469} 470 471#endif // COMPILER 472 473} // namespace BASE_HASH_NAMESPACE 474 475#endif // BASE_FILES_FILE_PATH_H_ 476