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 <functional>
108#include <iosfwd>
109#include <string>
110#include <vector>
111
112#include "base/base_export.h"
113#include "base/compiler_specific.h"
114#include "base/stl_util.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 PRFilePath (based on PRIuS and friends from
129// C99 and format_macros.h) like this:
130// base::StringPrintf("Path is %" PRFilePath ".\n", path.value().c_str());
131#if defined(OS_WIN)
132#define PRFilePath "ls"
133#elif defined(OS_POSIX) || defined(OS_FUCHSIA)
134#define PRFilePath "s"
135#endif // OS_WIN
136
137// Macros for string literal initialization of FilePath::CharType[].
138#if defined(OS_WIN)
139#define FILE_PATH_LITERAL(x) L##x
140#elif defined(OS_POSIX) || defined(OS_FUCHSIA)
141#define FILE_PATH_LITERAL(x) x
142#endif // OS_WIN
143
144namespace base {
145
146class Pickle;
147class PickleIterator;
148
149// An abstraction to isolate users from the differences between native
150// pathnames on different platforms.
151class BASE_EXPORT FilePath {
152 public:
153#if defined(OS_WIN)
154 // On Windows, for Unicode-aware applications, native pathnames are wchar_t
155 // arrays encoded in UTF-16.
156 typedef base::string16 StringType;
157#elif defined(OS_POSIX) || defined(OS_FUCHSIA)
158 // On most platforms, native pathnames are char arrays, and the encoding
159 // may or may not be specified. On Mac OS X, native pathnames are encoded
160 // in UTF-8.
161 typedef std::string StringType;
162#endif // OS_WIN
163
164 typedef BasicStringPiece<StringType> StringPieceType;
165 typedef StringType::value_type CharType;
166
167 // Null-terminated array of separators used to separate components in
168 // hierarchical paths. Each character in this array is a valid separator,
169 // but kSeparators[0] is treated as the canonical separator and will be used
170 // when composing pathnames.
171 static const CharType kSeparators[];
172
173 // base::size(kSeparators).
174 static const size_t kSeparatorsLength;
175
176 // A special path component meaning "this directory."
177 static const CharType kCurrentDirectory[];
178
179 // A special path component meaning "the parent directory."
180 static const CharType kParentDirectory[];
181
182 // The character used to identify a file extension.
183 static const CharType kExtensionSeparator;
184
185 FilePath();
186 FilePath(const FilePath& that);
187 explicit FilePath(StringPieceType path);
188 ~FilePath();
189 FilePath& operator=(const FilePath& that);
190
191 // Constructs FilePath with the contents of |that|, which is left in valid but
192 // unspecified state.
193 FilePath(FilePath&& that) noexcept;
194 // Replaces the contents with those of |that|, which is left in valid but
195 // unspecified state.
196 FilePath& operator=(FilePath&& that);
197
198 bool operator==(const FilePath& that) const;
199
200 bool operator!=(const FilePath& that) const;
201
202 // Required for some STL containers and operations
203 bool operator<(const FilePath& that) const {
204 return path_ < that.path_;
205 }
206
207 const StringType& value() const { return path_; }
208
209 bool empty() const { return path_.empty(); }
210
211 void clear() { path_.clear(); }
212
213 // Returns true if |character| is in kSeparators.
214 static bool IsSeparator(CharType character);
215
216 // Returns a vector of all of the components of the provided path. It is
217 // equivalent to calling DirName().value() on the path's root component,
218 // and BaseName().value() on each child component.
219 //
220 // To make sure this is lossless so we can differentiate absolute and
221 // relative paths, the root slash will be included even though no other
222 // slashes will be. The precise behavior is:
223 //
224 // Posix: "/foo/bar" -> [ "/", "foo", "bar" ]
225 // Windows: "C:\foo\bar" -> [ "C:", "\\", "foo", "bar" ]
226 void GetComponents(std::vector<FilePath::StringType>* components) const;
227
228 // Returns true if this FilePath is a parent or ancestor of the |child|.
229 // Absolute and relative paths are accepted i.e. /foo is a parent to /foo/bar,
230 // and foo is a parent to foo/bar. Any ancestor is considered a parent i.e. /a
231 // is a parent to both /a/b and /a/b/c. Does not convert paths to absolute,
232 // follow symlinks or directory navigation (e.g. ".."). A path is *NOT* its
233 // own parent.
234 bool IsParent(const FilePath& child) const;
235
236 // If IsParent(child) holds, appends to path (if non-NULL) the
237 // relative path to child and returns true. For example, if parent
238 // holds "/Users/johndoe/Library/Application Support", child holds
239 // "/Users/johndoe/Library/Application Support/Google/Chrome/Default", and
240 // *path holds "/Users/johndoe/Library/Caches", then after
241 // parent.AppendRelativePath(child, path) is called *path will hold
242 // "/Users/johndoe/Library/Caches/Google/Chrome/Default". Otherwise,
243 // returns false.
244 bool AppendRelativePath(const FilePath& child, FilePath* path) const;
245
246 // Returns a FilePath corresponding to the directory containing the path
247 // named by this object, stripping away the file component. If this object
248 // only contains one component, returns a FilePath identifying
249 // kCurrentDirectory. If this object already refers to the root directory,
250 // returns a FilePath identifying the root directory. Please note that this
251 // doesn't resolve directory navigation, e.g. the result for "../a" is "..".
252 FilePath DirName() const WARN_UNUSED_RESULT;
253
254 // Returns a FilePath corresponding to the last path component of this
255 // object, either a file or a directory. If this object already refers to
256 // the root directory, returns a FilePath identifying the root directory;
257 // this is the only situation in which BaseName will return an absolute path.
258 FilePath BaseName() const WARN_UNUSED_RESULT;
259
260 // Returns ".jpg" for path "C:\pics\jojo.jpg", or an empty string if
261 // the file has no extension. If non-empty, Extension() will always start
262 // with precisely one ".". The following code should always work regardless
263 // of the value of path. For common double-extensions like .tar.gz and
264 // .user.js, this method returns the combined extension. For a single
265 // component, use FinalExtension().
266 // new_path = path.RemoveExtension().value().append(path.Extension());
267 // ASSERT(new_path == path.value());
268 // NOTE: this is different from the original file_util implementation which
269 // returned the extension without a leading "." ("jpg" instead of ".jpg")
270 StringType Extension() const WARN_UNUSED_RESULT;
271
272 // Returns the path's file extension, as in Extension(), but will
273 // never return a double extension.
274 //
275 // TODO(davidben): Check all our extension-sensitive code to see if
276 // we can rename this to Extension() and the other to something like
277 // LongExtension(), defaulting to short extensions and leaving the
278 // long "extensions" to logic like base::GetUniquePathNumber().
279 StringType FinalExtension() const WARN_UNUSED_RESULT;
280
281 // Returns "C:\pics\jojo" for path "C:\pics\jojo.jpg"
282 // NOTE: this is slightly different from the similar file_util implementation
283 // which returned simply 'jojo'.
284 FilePath RemoveExtension() const WARN_UNUSED_RESULT;
285
286 // Removes the path's file extension, as in RemoveExtension(), but
287 // ignores double extensions.
288 FilePath RemoveFinalExtension() const WARN_UNUSED_RESULT;
289
290 // Inserts |suffix| after the file name portion of |path| but before the
291 // extension. Returns "" if BaseName() == "." or "..".
292 // Examples:
293 // path == "C:\pics\jojo.jpg" suffix == " (1)", returns "C:\pics\jojo (1).jpg"
294 // path == "jojo.jpg" suffix == " (1)", returns "jojo (1).jpg"
295 // path == "C:\pics\jojo" suffix == " (1)", returns "C:\pics\jojo (1)"
296 // path == "C:\pics.old\jojo" suffix == " (1)", returns "C:\pics.old\jojo (1)"
297 FilePath InsertBeforeExtension(
298 StringPieceType suffix) const WARN_UNUSED_RESULT;
299 FilePath InsertBeforeExtensionASCII(
300 StringPiece suffix) const WARN_UNUSED_RESULT;
301
302 // Adds |extension| to |file_name|. Returns the current FilePath if
303 // |extension| is empty. Returns "" if BaseName() == "." or "..".
304 FilePath AddExtension(StringPieceType extension) const WARN_UNUSED_RESULT;
305
306 // Like above, but takes the extension as an ASCII string. See AppendASCII for
307 // details on how this is handled.
308 FilePath AddExtensionASCII(StringPiece extension) const WARN_UNUSED_RESULT;
309
310 // Replaces the extension of |file_name| with |extension|. If |file_name|
311 // does not have an extension, then |extension| is added. If |extension| is
312 // empty, then the extension is removed from |file_name|.
313 // Returns "" if BaseName() == "." or "..".
314 FilePath ReplaceExtension(StringPieceType extension) const WARN_UNUSED_RESULT;
315
316 // Returns true if the file path matches the specified extension. The test is
317 // case insensitive. Don't forget the leading period if appropriate.
318 bool MatchesExtension(StringPieceType extension) const;
319
320 // Returns a FilePath by appending a separator and the supplied path
321 // component to this object's path. Append takes care to avoid adding
322 // excessive separators if this object's path already ends with a separator.
323 // If this object's path is kCurrentDirectory, a new FilePath corresponding
324 // only to |component| is returned. |component| must be a relative path;
325 // it is an error to pass an absolute path.
326 FilePath Append(StringPieceType component) const WARN_UNUSED_RESULT;
327 FilePath Append(const FilePath& component) const WARN_UNUSED_RESULT;
328
329 // Although Windows StringType is std::wstring, since the encoding it uses for
330 // paths is well defined, it can handle ASCII path components as well.
331 // Mac uses UTF8, and since ASCII is a subset of that, it works there as well.
332 // On Linux, although it can use any 8-bit encoding for paths, we assume that
333 // ASCII is a valid subset, regardless of the encoding, since many operating
334 // system paths will always be ASCII.
335 FilePath AppendASCII(StringPiece component) const WARN_UNUSED_RESULT;
336
337 // Returns true if this FilePath contains an absolute path. On Windows, an
338 // absolute path begins with either a drive letter specification followed by
339 // a separator character, or with two separator characters. On POSIX
340 // platforms, an absolute path begins with a separator character.
341 bool IsAbsolute() const;
342
343 // Returns true if the patch ends with a path separator character.
344 bool EndsWithSeparator() const WARN_UNUSED_RESULT;
345
346 // Returns a copy of this FilePath that ends with a trailing separator. If
347 // the input path is empty, an empty FilePath will be returned.
348 FilePath AsEndingWithSeparator() const WARN_UNUSED_RESULT;
349
350 // Returns a copy of this FilePath that does not end with a trailing
351 // separator.
352 FilePath StripTrailingSeparators() const WARN_UNUSED_RESULT;
353
354 // Returns true if this FilePath contains an attempt to reference a parent
355 // directory (e.g. has a path component that is "..").
356 bool ReferencesParent() const;
357
358 // Return a Unicode human-readable version of this path.
359 // Warning: you can *not*, in general, go from a display name back to a real
360 // path. Only use this when displaying paths to users, not just when you
361 // want to stuff a string16 into some other API.
362 string16 LossyDisplayName() const;
363
364 // Return the path as ASCII, or the empty string if the path is not ASCII.
365 // This should only be used for cases where the FilePath is representing a
366 // known-ASCII filename.
367 std::string MaybeAsASCII() const;
368
369 // Return the path as UTF-8.
370 //
371 // This function is *unsafe* as there is no way to tell what encoding is
372 // used in file names on POSIX systems other than Mac and Chrome OS,
373 // although UTF-8 is practically used everywhere these days. To mitigate
374 // the encoding issue, this function internally calls
375 // SysNativeMBToWide() on POSIX systems other than Mac and Chrome OS,
376 // per assumption that the current locale's encoding is used in file
377 // names, but this isn't a perfect solution.
378 //
379 // Once it becomes safe to to stop caring about non-UTF-8 file names,
380 // the SysNativeMBToWide() hack will be removed from the code, along
381 // with "Unsafe" in the function name.
382 std::string AsUTF8Unsafe() const;
383
384 // Similar to AsUTF8Unsafe, but returns UTF-16 instead.
385 string16 AsUTF16Unsafe() const;
386
387 // Returns a FilePath object from a path name in UTF-8. This function
388 // should only be used for cases where you are sure that the input
389 // string is UTF-8.
390 //
391 // Like AsUTF8Unsafe(), this function is unsafe. This function
392 // internally calls SysWideToNativeMB() on POSIX systems other than Mac
393 // and Chrome OS, to mitigate the encoding issue. See the comment at
394 // AsUTF8Unsafe() for details.
395 static FilePath FromUTF8Unsafe(StringPiece utf8);
396
397 // Similar to FromUTF8Unsafe, but accepts UTF-16 instead.
398 static FilePath FromUTF16Unsafe(StringPiece16 utf16);
399
400 void WriteToPickle(Pickle* pickle) const;
401 bool ReadFromPickle(PickleIterator* iter);
402
403 // Normalize all path separators to backslash on Windows
404 // (if FILE_PATH_USES_WIN_SEPARATORS is true), or do nothing on POSIX systems.
405 FilePath NormalizePathSeparators() const;
406
407 // Normalize all path separattors to given type on Windows
408 // (if FILE_PATH_USES_WIN_SEPARATORS is true), or do nothing on POSIX systems.
409 FilePath NormalizePathSeparatorsTo(CharType separator) const;
410
411 // Compare two strings in the same way the file system does.
412 // Note that these always ignore case, even on file systems that are case-
413 // sensitive. If case-sensitive comparison is ever needed, add corresponding
414 // methods here.
415 // The methods are written as a static method so that they can also be used
416 // on parts of a file path, e.g., just the extension.
417 // CompareIgnoreCase() returns -1, 0 or 1 for less-than, equal-to and
418 // greater-than respectively.
419 static int CompareIgnoreCase(StringPieceType string1,
420 StringPieceType string2);
421 static bool CompareEqualIgnoreCase(StringPieceType string1,
422 StringPieceType string2) {
423 return CompareIgnoreCase(string1, string2) == 0;
424 }
425 static bool CompareLessIgnoreCase(StringPieceType string1,
426 StringPieceType string2) {
427 return CompareIgnoreCase(string1, string2) < 0;
428 }
429
430#if defined(OS_MACOSX)
431 // Returns the string in the special canonical decomposed form as defined for
432 // HFS, which is close to, but not quite, decomposition form D. See
433 // http://developer.apple.com/mac/library/technotes/tn/tn1150.html#UnicodeSubtleties
434 // for further comments.
435 // Returns the epmty string if the conversion failed.
436 static StringType GetHFSDecomposedForm(StringPieceType string);
437
438 // Special UTF-8 version of FastUnicodeCompare. Cf:
439 // http://developer.apple.com/mac/library/technotes/tn/tn1150.html#StringComparisonAlgorithm
440 // IMPORTANT: The input strings must be in the special HFS decomposed form!
441 // (cf. above GetHFSDecomposedForm method)
442 static int HFSFastUnicodeCompare(StringPieceType string1,
443 StringPieceType string2);
444#endif
445
446#if defined(OS_ANDROID)
447 // On android, file selection dialog can return a file with content uri
448 // scheme(starting with content://). Content uri needs to be opened with
449 // ContentResolver to guarantee that the app has appropriate permissions
450 // to access it.
451 // Returns true if the path is a content uri, or false otherwise.
452 bool IsContentUri() const;
453#endif
454
455 private:
456 // Remove trailing separators from this object. If the path is absolute, it
457 // will never be stripped any more than to refer to the absolute root
458 // directory, so "////" will become "/", not "". A leading pair of
459 // separators is never stripped, to support alternate roots. This is used to
460 // support UNC paths on Windows.
461 void StripTrailingSeparatorsInternal();
462
463 StringType path_;
464};
465
466BASE_EXPORT std::ostream& operator<<(std::ostream& out,
467 const FilePath& file_path);
468
469} // namespace base
470
471namespace std {
472
473template <>
474struct hash<base::FilePath> {
475 typedef base::FilePath argument_type;
476 typedef std::size_t result_type;
477 result_type operator()(argument_type const& f) const {
478 return hash<base::FilePath::StringType>()(f.value());
479 }
480};
481
482} // namespace std
483
484#endif // BASE_FILES_FILE_PATH_H_
485