summaryrefslogtreecommitdiffstats
path: root/security/sandbox/chromium/base/strings/string_util.h
diff options
context:
space:
mode:
Diffstat (limited to 'security/sandbox/chromium/base/strings/string_util.h')
-rw-r--r--security/sandbox/chromium/base/strings/string_util.h461
1 files changed, 0 insertions, 461 deletions
diff --git a/security/sandbox/chromium/base/strings/string_util.h b/security/sandbox/chromium/base/strings/string_util.h
deleted file mode 100644
index e369f294d..000000000
--- a/security/sandbox/chromium/base/strings/string_util.h
+++ /dev/null
@@ -1,461 +0,0 @@
-// Copyright 2013 The Chromium Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-//
-// This file defines utility functions for working with strings.
-
-#ifndef BASE_STRINGS_STRING_UTIL_H_
-#define BASE_STRINGS_STRING_UTIL_H_
-
-#include <ctype.h>
-#include <stdarg.h> // va_list
-#include <stddef.h>
-#include <stdint.h>
-
-#include <string>
-#include <vector>
-
-#include "base/base_export.h"
-#include "base/compiler_specific.h"
-#include "base/strings/string16.h"
-#include "base/strings/string_piece.h" // For implicit conversions.
-#include "build/build_config.h"
-
-namespace base {
-
-// C standard-library functions that aren't cross-platform are provided as
-// "base::...", and their prototypes are listed below. These functions are
-// then implemented as inline calls to the platform-specific equivalents in the
-// platform-specific headers.
-
-// Wrapper for vsnprintf that always null-terminates and always returns the
-// number of characters that would be in an untruncated formatted
-// string, even when truncation occurs.
-int vsnprintf(char* buffer, size_t size, const char* format, va_list arguments)
- PRINTF_FORMAT(3, 0);
-
-// Some of these implementations need to be inlined.
-
-// We separate the declaration from the implementation of this inline
-// function just so the PRINTF_FORMAT works.
-inline int snprintf(char* buffer,
- size_t size,
- _Printf_format_string_ const char* format,
- ...) PRINTF_FORMAT(3, 4);
-inline int snprintf(char* buffer,
- size_t size,
- _Printf_format_string_ const char* format,
- ...) {
- va_list arguments;
- va_start(arguments, format);
- int result = vsnprintf(buffer, size, format, arguments);
- va_end(arguments);
- return result;
-}
-
-// BSD-style safe and consistent string copy functions.
-// Copies |src| to |dst|, where |dst_size| is the total allocated size of |dst|.
-// Copies at most |dst_size|-1 characters, and always NULL terminates |dst|, as
-// long as |dst_size| is not 0. Returns the length of |src| in characters.
-// If the return value is >= dst_size, then the output was truncated.
-// NOTE: All sizes are in number of characters, NOT in bytes.
-BASE_EXPORT size_t strlcpy(char* dst, const char* src, size_t dst_size);
-BASE_EXPORT size_t wcslcpy(wchar_t* dst, const wchar_t* src, size_t dst_size);
-
-// Scan a wprintf format string to determine whether it's portable across a
-// variety of systems. This function only checks that the conversion
-// specifiers used by the format string are supported and have the same meaning
-// on a variety of systems. It doesn't check for other errors that might occur
-// within a format string.
-//
-// Nonportable conversion specifiers for wprintf are:
-// - 's' and 'c' without an 'l' length modifier. %s and %c operate on char
-// data on all systems except Windows, which treat them as wchar_t data.
-// Use %ls and %lc for wchar_t data instead.
-// - 'S' and 'C', which operate on wchar_t data on all systems except Windows,
-// which treat them as char data. Use %ls and %lc for wchar_t data
-// instead.
-// - 'F', which is not identified by Windows wprintf documentation.
-// - 'D', 'O', and 'U', which are deprecated and not available on all systems.
-// Use %ld, %lo, and %lu instead.
-//
-// Note that there is no portable conversion specifier for char data when
-// working with wprintf.
-//
-// This function is intended to be called from base::vswprintf.
-BASE_EXPORT bool IsWprintfFormatPortable(const wchar_t* format);
-
-// ASCII-specific tolower. The standard library's tolower is locale sensitive,
-// so we don't want to use it here.
-inline char ToLowerASCII(char c) {
- return (c >= 'A' && c <= 'Z') ? (c + ('a' - 'A')) : c;
-}
-inline char16 ToLowerASCII(char16 c) {
- return (c >= 'A' && c <= 'Z') ? (c + ('a' - 'A')) : c;
-}
-
-// ASCII-specific toupper. The standard library's toupper is locale sensitive,
-// so we don't want to use it here.
-inline char ToUpperASCII(char c) {
- return (c >= 'a' && c <= 'z') ? (c + ('A' - 'a')) : c;
-}
-inline char16 ToUpperASCII(char16 c) {
- return (c >= 'a' && c <= 'z') ? (c + ('A' - 'a')) : c;
-}
-
-// Converts the given string to it's ASCII-lowercase equivalent.
-BASE_EXPORT std::string ToLowerASCII(StringPiece str);
-BASE_EXPORT string16 ToLowerASCII(StringPiece16 str);
-
-// Converts the given string to it's ASCII-uppercase equivalent.
-BASE_EXPORT std::string ToUpperASCII(StringPiece str);
-BASE_EXPORT string16 ToUpperASCII(StringPiece16 str);
-
-// Functor for case-insensitive ASCII comparisons for STL algorithms like
-// std::search.
-//
-// Note that a full Unicode version of this functor is not possible to write
-// because case mappings might change the number of characters, depend on
-// context (combining accents), and require handling UTF-16. If you need
-// proper Unicode support, use base::i18n::ToLower/FoldCase and then just
-// use a normal operator== on the result.
-template<typename Char> struct CaseInsensitiveCompareASCII {
- public:
- bool operator()(Char x, Char y) const {
- return ToLowerASCII(x) == ToLowerASCII(y);
- }
-};
-
-// Like strcasecmp for case-insensitive ASCII characters only. Returns:
-// -1 (a < b)
-// 0 (a == b)
-// 1 (a > b)
-// (unlike strcasecmp which can return values greater or less than 1/-1). For
-// full Unicode support, use base::i18n::ToLower or base::i18h::FoldCase
-// and then just call the normal string operators on the result.
-BASE_EXPORT int CompareCaseInsensitiveASCII(StringPiece a, StringPiece b);
-BASE_EXPORT int CompareCaseInsensitiveASCII(StringPiece16 a, StringPiece16 b);
-
-// Equality for ASCII case-insensitive comparisons. For full Unicode support,
-// use base::i18n::ToLower or base::i18h::FoldCase and then compare with either
-// == or !=.
-BASE_EXPORT bool EqualsCaseInsensitiveASCII(StringPiece a, StringPiece b);
-BASE_EXPORT bool EqualsCaseInsensitiveASCII(StringPiece16 a, StringPiece16 b);
-
-// These threadsafe functions return references to globally unique empty
-// strings.
-//
-// It is likely faster to construct a new empty string object (just a few
-// instructions to set the length to 0) than to get the empty string singleton
-// returned by these functions (which requires threadsafe singleton access).
-//
-// Therefore, DO NOT USE THESE AS A GENERAL-PURPOSE SUBSTITUTE FOR DEFAULT
-// CONSTRUCTORS. There is only one case where you should use these: functions
-// which need to return a string by reference (e.g. as a class member
-// accessor), and don't have an empty string to use (e.g. in an error case).
-// These should not be used as initializers, function arguments, or return
-// values for functions which return by value or outparam.
-BASE_EXPORT const std::string& EmptyString();
-BASE_EXPORT const string16& EmptyString16();
-
-// Contains the set of characters representing whitespace in the corresponding
-// encoding. Null-terminated. The ASCII versions are the whitespaces as defined
-// by HTML5, and don't include control characters.
-BASE_EXPORT extern const wchar_t kWhitespaceWide[]; // Includes Unicode.
-BASE_EXPORT extern const char16 kWhitespaceUTF16[]; // Includes Unicode.
-BASE_EXPORT extern const char kWhitespaceASCII[];
-BASE_EXPORT extern const char16 kWhitespaceASCIIAs16[]; // No unicode.
-
-// Null-terminated string representing the UTF-8 byte order mark.
-BASE_EXPORT extern const char kUtf8ByteOrderMark[];
-
-// Removes characters in |remove_chars| from anywhere in |input|. Returns true
-// if any characters were removed. |remove_chars| must be null-terminated.
-// NOTE: Safe to use the same variable for both |input| and |output|.
-BASE_EXPORT bool RemoveChars(const string16& input,
- const StringPiece16& remove_chars,
- string16* output);
-BASE_EXPORT bool RemoveChars(const std::string& input,
- const StringPiece& remove_chars,
- std::string* output);
-
-// Replaces characters in |replace_chars| from anywhere in |input| with
-// |replace_with|. Each character in |replace_chars| will be replaced with
-// the |replace_with| string. Returns true if any characters were replaced.
-// |replace_chars| must be null-terminated.
-// NOTE: Safe to use the same variable for both |input| and |output|.
-BASE_EXPORT bool ReplaceChars(const string16& input,
- const StringPiece16& replace_chars,
- const string16& replace_with,
- string16* output);
-BASE_EXPORT bool ReplaceChars(const std::string& input,
- const StringPiece& replace_chars,
- const std::string& replace_with,
- std::string* output);
-
-enum TrimPositions {
- TRIM_NONE = 0,
- TRIM_LEADING = 1 << 0,
- TRIM_TRAILING = 1 << 1,
- TRIM_ALL = TRIM_LEADING | TRIM_TRAILING,
-};
-
-// Removes characters in |trim_chars| from the beginning and end of |input|.
-// The 8-bit version only works on 8-bit characters, not UTF-8.
-//
-// It is safe to use the same variable for both |input| and |output| (this is
-// the normal usage to trim in-place).
-BASE_EXPORT bool TrimString(const string16& input,
- StringPiece16 trim_chars,
- string16* output);
-BASE_EXPORT bool TrimString(const std::string& input,
- StringPiece trim_chars,
- std::string* output);
-
-// StringPiece versions of the above. The returned pieces refer to the original
-// buffer.
-BASE_EXPORT StringPiece16 TrimString(StringPiece16 input,
- const StringPiece16& trim_chars,
- TrimPositions positions);
-BASE_EXPORT StringPiece TrimString(StringPiece input,
- const StringPiece& trim_chars,
- TrimPositions positions);
-
-// Truncates a string to the nearest UTF-8 character that will leave
-// the string less than or equal to the specified byte size.
-BASE_EXPORT void TruncateUTF8ToByteSize(const std::string& input,
- const size_t byte_size,
- std::string* output);
-
-// Trims any whitespace from either end of the input string.
-//
-// The StringPiece versions return a substring referencing the input buffer.
-// The ASCII versions look only for ASCII whitespace.
-//
-// The std::string versions return where whitespace was found.
-// NOTE: Safe to use the same variable for both input and output.
-BASE_EXPORT TrimPositions TrimWhitespace(const string16& input,
- TrimPositions positions,
- string16* output);
-BASE_EXPORT StringPiece16 TrimWhitespace(StringPiece16 input,
- TrimPositions positions);
-BASE_EXPORT TrimPositions TrimWhitespaceASCII(const std::string& input,
- TrimPositions positions,
- std::string* output);
-BASE_EXPORT StringPiece TrimWhitespaceASCII(StringPiece input,
- TrimPositions positions);
-
-// Searches for CR or LF characters. Removes all contiguous whitespace
-// strings that contain them. This is useful when trying to deal with text
-// copied from terminals.
-// Returns |text|, with the following three transformations:
-// (1) Leading and trailing whitespace is trimmed.
-// (2) If |trim_sequences_with_line_breaks| is true, any other whitespace
-// sequences containing a CR or LF are trimmed.
-// (3) All other whitespace sequences are converted to single spaces.
-BASE_EXPORT string16 CollapseWhitespace(
- const string16& text,
- bool trim_sequences_with_line_breaks);
-BASE_EXPORT std::string CollapseWhitespaceASCII(
- const std::string& text,
- bool trim_sequences_with_line_breaks);
-
-// Returns true if |input| is empty or contains only characters found in
-// |characters|.
-BASE_EXPORT bool ContainsOnlyChars(const StringPiece& input,
- const StringPiece& characters);
-BASE_EXPORT bool ContainsOnlyChars(const StringPiece16& input,
- const StringPiece16& characters);
-
-// Returns true if the specified string matches the criteria. How can a wide
-// string be 8-bit or UTF8? It contains only characters that are < 256 (in the
-// first case) or characters that use only 8-bits and whose 8-bit
-// representation looks like a UTF-8 string (the second case).
-//
-// Note that IsStringUTF8 checks not only if the input is structurally
-// valid but also if it doesn't contain any non-character codepoint
-// (e.g. U+FFFE). It's done on purpose because all the existing callers want
-// to have the maximum 'discriminating' power from other encodings. If
-// there's a use case for just checking the structural validity, we have to
-// add a new function for that.
-//
-// IsStringASCII assumes the input is likely all ASCII, and does not leave early
-// if it is not the case.
-BASE_EXPORT bool IsStringUTF8(const StringPiece& str);
-BASE_EXPORT bool IsStringASCII(const StringPiece& str);
-BASE_EXPORT bool IsStringASCII(const StringPiece16& str);
-// A convenience adaptor for WebStrings, as they don't convert into
-// StringPieces directly.
-BASE_EXPORT bool IsStringASCII(const string16& str);
-#if defined(WCHAR_T_IS_UTF32)
-BASE_EXPORT bool IsStringASCII(const std::wstring& str);
-#endif
-
-// Compare the lower-case form of the given string against the given
-// previously-lower-cased ASCII string (typically a constant).
-BASE_EXPORT bool LowerCaseEqualsASCII(StringPiece str,
- StringPiece lowecase_ascii);
-BASE_EXPORT bool LowerCaseEqualsASCII(StringPiece16 str,
- StringPiece lowecase_ascii);
-
-// Performs a case-sensitive string compare of the given 16-bit string against
-// the given 8-bit ASCII string (typically a constant). The behavior is
-// undefined if the |ascii| string is not ASCII.
-BASE_EXPORT bool EqualsASCII(StringPiece16 str, StringPiece ascii);
-
-// Indicates case sensitivity of comparisons. Only ASCII case insensitivity
-// is supported. Full Unicode case-insensitive conversions would need to go in
-// base/i18n so it can use ICU.
-//
-// If you need to do Unicode-aware case-insensitive StartsWith/EndsWith, it's
-// best to call base::i18n::ToLower() or base::i18n::FoldCase() (see
-// base/i18n/case_conversion.h for usage advice) on the arguments, and then use
-// the results to a case-sensitive comparison.
-enum class CompareCase {
- SENSITIVE,
- INSENSITIVE_ASCII,
-};
-
-BASE_EXPORT bool StartsWith(StringPiece str,
- StringPiece search_for,
- CompareCase case_sensitivity);
-BASE_EXPORT bool StartsWith(StringPiece16 str,
- StringPiece16 search_for,
- CompareCase case_sensitivity);
-BASE_EXPORT bool EndsWith(StringPiece str,
- StringPiece search_for,
- CompareCase case_sensitivity);
-BASE_EXPORT bool EndsWith(StringPiece16 str,
- StringPiece16 search_for,
- CompareCase case_sensitivity);
-
-// Determines the type of ASCII character, independent of locale (the C
-// library versions will change based on locale).
-template <typename Char>
-inline bool IsAsciiWhitespace(Char c) {
- return c == ' ' || c == '\r' || c == '\n' || c == '\t';
-}
-template <typename Char>
-inline bool IsAsciiAlpha(Char c) {
- return ((c >= 'A') && (c <= 'Z')) || ((c >= 'a') && (c <= 'z'));
-}
-template <typename Char>
-inline bool IsAsciiDigit(Char c) {
- return c >= '0' && c <= '9';
-}
-
-template <typename Char>
-inline bool IsHexDigit(Char c) {
- return (c >= '0' && c <= '9') ||
- (c >= 'A' && c <= 'F') ||
- (c >= 'a' && c <= 'f');
-}
-
-// Returns the integer corresponding to the given hex character. For example:
-// '4' -> 4
-// 'a' -> 10
-// 'B' -> 11
-// Assumes the input is a valid hex character. DCHECKs in debug builds if not.
-BASE_EXPORT char HexDigitToInt(wchar_t c);
-
-// Returns true if it's a Unicode whitespace character.
-BASE_EXPORT bool IsUnicodeWhitespace(wchar_t c);
-
-// Return a byte string in human-readable format with a unit suffix. Not
-// appropriate for use in any UI; use of FormatBytes and friends in ui/base is
-// highly recommended instead. TODO(avi): Figure out how to get callers to use
-// FormatBytes instead; remove this.
-BASE_EXPORT string16 FormatBytesUnlocalized(int64_t bytes);
-
-// Starting at |start_offset| (usually 0), replace the first instance of
-// |find_this| with |replace_with|.
-BASE_EXPORT void ReplaceFirstSubstringAfterOffset(
- base::string16* str,
- size_t start_offset,
- StringPiece16 find_this,
- StringPiece16 replace_with);
-BASE_EXPORT void ReplaceFirstSubstringAfterOffset(
- std::string* str,
- size_t start_offset,
- StringPiece find_this,
- StringPiece replace_with);
-
-// Starting at |start_offset| (usually 0), look through |str| and replace all
-// instances of |find_this| with |replace_with|.
-//
-// This does entire substrings; use std::replace in <algorithm> for single
-// characters, for example:
-// std::replace(str.begin(), str.end(), 'a', 'b');
-BASE_EXPORT void ReplaceSubstringsAfterOffset(
- string16* str,
- size_t start_offset,
- StringPiece16 find_this,
- StringPiece16 replace_with);
-BASE_EXPORT void ReplaceSubstringsAfterOffset(
- std::string* str,
- size_t start_offset,
- StringPiece find_this,
- StringPiece replace_with);
-
-// Reserves enough memory in |str| to accommodate |length_with_null| characters,
-// sets the size of |str| to |length_with_null - 1| characters, and returns a
-// pointer to the underlying contiguous array of characters. This is typically
-// used when calling a function that writes results into a character array, but
-// the caller wants the data to be managed by a string-like object. It is
-// convenient in that is can be used inline in the call, and fast in that it
-// avoids copying the results of the call from a char* into a string.
-//
-// |length_with_null| must be at least 2, since otherwise the underlying string
-// would have size 0, and trying to access &((*str)[0]) in that case can result
-// in a number of problems.
-//
-// Internally, this takes linear time because the resize() call 0-fills the
-// underlying array for potentially all
-// (|length_with_null - 1| * sizeof(string_type::value_type)) bytes. Ideally we
-// could avoid this aspect of the resize() call, as we expect the caller to
-// immediately write over this memory, but there is no other way to set the size
-// of the string, and not doing that will mean people who access |str| rather
-// than str.c_str() will get back a string of whatever size |str| had on entry
-// to this function (probably 0).
-BASE_EXPORT char* WriteInto(std::string* str, size_t length_with_null);
-BASE_EXPORT char16* WriteInto(string16* str, size_t length_with_null);
-#ifndef OS_WIN
-BASE_EXPORT wchar_t* WriteInto(std::wstring* str, size_t length_with_null);
-#endif
-
-// Does the opposite of SplitString().
-BASE_EXPORT std::string JoinString(const std::vector<std::string>& parts,
- StringPiece separator);
-BASE_EXPORT string16 JoinString(const std::vector<string16>& parts,
- StringPiece16 separator);
-
-// Replace $1-$2-$3..$9 in the format string with |a|-|b|-|c|..|i| respectively.
-// Additionally, any number of consecutive '$' characters is replaced by that
-// number less one. Eg $$->$, $$$->$$, etc. The offsets parameter here can be
-// NULL. This only allows you to use up to nine replacements.
-BASE_EXPORT string16 ReplaceStringPlaceholders(
- const string16& format_string,
- const std::vector<string16>& subst,
- std::vector<size_t>* offsets);
-
-BASE_EXPORT std::string ReplaceStringPlaceholders(
- const StringPiece& format_string,
- const std::vector<std::string>& subst,
- std::vector<size_t>* offsets);
-
-// Single-string shortcut for ReplaceStringHolders. |offset| may be NULL.
-BASE_EXPORT string16 ReplaceStringPlaceholders(const string16& format_string,
- const string16& a,
- size_t* offset);
-
-} // namespace base
-
-#if defined(OS_WIN)
-#include "base/strings/string_util_win.h"
-#elif defined(OS_POSIX)
-#include "base/strings/string_util_posix.h"
-#else
-#error Define string operations appropriately for your platform
-#endif
-
-#endif // BASE_STRINGS_STRING_UTIL_H_