📄 modp_burl.h
字号:
/* -*- mode: c++; c-basic-offset: 4; indent-tabs-mode: nil; tab-width: 4 -*- *//* vi: set expandtab shiftwidth=4 tabstop=4: *//** * \file * <PRE> * High Performance URL Encoder/Decoder * * Copyright © 2006, 2007 Nick Galbreath -- nickg [at] modp [dot] com * All rights reserved. * * http://code.google.com/p/stringencoders/ * * Released under bsd license. See bfast64.c for details. * </PRE> */#ifndef COM_MODP_STRINGENCODERS_BURL#define COM_MODP_STRINGENCODERS_BURL#ifdef __cplusplus#define BEGIN_C extern "C" {#define END_C }#else#define BEGIN_C#define END_C#endifBEGIN_C/** * Url encode a string. This uses a very strict definition of url encoding. * The only characters NOT encoded are A-Z, a-z, 0-9, "-", "_", ".", along with * the space char getting mapped to "+". Everything else is escaped using * "%HEXHEX" format. This is identical to the implementation of php's urlencode * and nearly identical to Java's UrlEncoder class (they do not escape '*' for * some reason). * * \param[out] dest output string. Must * \param[in] str The input string * \param[in] len The length of the input string, excluding any * final null byte. */int modp_burl_encode(char* dest, const char* str, int len);/** * Url encode a string. This uses a minimal definition of url encoding. * This works similar to the previous function except '~', '!', '$', '\'', * '(', ')', '*', ',', ';', ':', '@', '/', '?' are NOT escaped. This will * allow decoding by standard url-decoders and make the encoded urls more * readable. * * \param[out] dest output string. Must * \param[in] str The input string * \param[in] len The length of the input string, excluding any * final null byte. */int modp_burl_min_encode(char* dest, const char* str, int len);/** \brief get size of output string w/o doing actual encoding * * \param[in] src input string, not null * \param[in] len length of input string * \return length of output string NOT including any final null byte */int modp_burl_min_encode_strlen(const char* src, const int len);/** * Provides the maximum size for output string given * and input size of A bytes. */#define modp_burl_encode_len(A) (3*A + 1)/** * Given the exact size of output string. * * Can be used to allocate the right amount of memory for * modp_burl_encode. Be sure to add 1 byte for final null. * * This is somewhat expensive since it examines every character * in the input string * * \param[in] str The input string * \param[in] len THe length of the input string, excluding any * final null byte (i.e. strlen(str)) * \return the size of the output string, excluding the final * null byte. */int modp_burl_encode_strlen(const char* str, const int len);/** * URL Decode a string * * \param[out] dest The output string. Must be at least (len + 1) * bytes allocated. This may be the same as the input buffer. * \param[in] str The input string that is URL encoded. * \param[in] len The length of the input string (excluding final * null byte) * \return the strlen of the output string. */int modp_burl_decode(char* dest, const char* str, int len);/** * Returns memory required to decoded a url-encoded * string of length A. * */#define modp_burl_decode_len(A) (A + 1)END_C#ifdef __cplusplus#include <string>namespace modp { inline std::string url_encode(const char* s, size_t len) { std::string x(modp_burl_encode_len(len), '\0'); int d = modp_burl_encode(const_cast<char*>(x.data()), s, len); x.erase(d, std::string::npos); return x; } inline std::string url_encode(const char* s) { return url_encode(s, strlen(s)); } inline std::string url_encode(const std::string& s) { return url_encode(s.data(), s.size()); } /** * Standard (maximal) url encoding. * * \param[in,out] s the string to be encoded * \return a reference to the input string */ inline std::string& url_encode(std::string& s) { std::string x(url_encode(s.data(), s.size())); s.swap(x); return s; } /** * Minimal Url Encoding * * \param[in,out] s the string to be encoded * \return a reference to the input string */ inline std::string& url_min_encode(std::string& s) { std::string x(modp_burl_encode_len(s.size()), '\0'); int d = modp_burl_min_encode(const_cast<char*>(x.data()), s.data(), s.size()); x.erase(d, std::string::npos); s.swap(x); return s; } inline std::string url_min_encode(const std::string& s) { std::string x(modp_burl_encode_len(s.size()), '\0'); int d = modp_burl_min_encode(const_cast<char*>(x.data()), s.data(), s.size()); x.erase(d, std::string::npos); return x; } /** * Url decode a string. * This function does not allocate memory. * * \param[in,out] s the string to be decoded * \return a reference to the input string. * There is no error case, bad characters are passed through */ inline std::string& url_decode(std::string& s) { int d = modp_burl_decode(const_cast<char*>(s.data()), s.data(), s.size()); s.erase(d, std::string::npos); return s; } inline std::string url_decode(const char* str) { std::string s(str); url_decode(s); return s; } inline std::string url_decode(const char* str, size_t len) { std::string s(str, len); url_decode(s); return s; } inline std::string url_decode(const std::string& s) { std::string x(s); url_decode(x); return x; }}#endif#endif
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -