mirror of
https://github.com/VCMP-SqMod/SqMod.git
synced 2024-11-14 03:37:16 +01:00
165 lines
5.1 KiB
C++
165 lines
5.1 KiB
C++
#include "cpr/util.h"
|
|
|
|
#include <algorithm>
|
|
#include <cassert>
|
|
#include <cctype>
|
|
#include <cstdint>
|
|
#include <fstream>
|
|
#include <iomanip>
|
|
#include <sstream>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
namespace cpr {
|
|
namespace util {
|
|
|
|
Cookies parseCookies(curl_slist* raw_cookies) {
|
|
Cookies cookies;
|
|
for (curl_slist* nc = raw_cookies; nc; nc = nc->next) {
|
|
std::vector<std::string> tokens = cpr::util::split(nc->data, '\t');
|
|
std::string value = tokens.back();
|
|
tokens.pop_back();
|
|
cookies[tokens.back()] = value;
|
|
}
|
|
return cookies;
|
|
}
|
|
|
|
Header parseHeader(const std::string& headers, std::string* status_line, std::string* reason) {
|
|
Header header;
|
|
std::vector<std::string> lines;
|
|
std::istringstream stream(headers);
|
|
{
|
|
std::string line;
|
|
while (std::getline(stream, line, '\n')) {
|
|
lines.push_back(line);
|
|
}
|
|
}
|
|
|
|
for (std::string& line : lines) {
|
|
// NOLINTNEXTLINE (cppcoreguidelines-avoid-magic-numbers)
|
|
if (line.substr(0, 5) == "HTTP/") {
|
|
// set the status_line if it was given
|
|
if ((status_line != nullptr) || (reason != nullptr)) {
|
|
line.resize(std::min<size_t>(line.size(), line.find_last_not_of("\t\n\r ") + 1));
|
|
if (status_line != nullptr) {
|
|
*status_line = line;
|
|
}
|
|
|
|
// set the reason if it was given
|
|
if (reason != nullptr) {
|
|
size_t pos1 = line.find_first_of("\t ");
|
|
size_t pos2 = std::string::npos;
|
|
if (pos1 != std::string::npos) {
|
|
pos2 = line.find_first_of("\t ", pos1 + 1);
|
|
}
|
|
if (pos2 != std::string::npos) {
|
|
line.erase(0, pos2 + 1);
|
|
*reason = line;
|
|
}
|
|
}
|
|
}
|
|
header.clear();
|
|
}
|
|
|
|
if (line.length() > 0) {
|
|
size_t found = line.find(':');
|
|
if (found != std::string::npos) {
|
|
std::string value = line.substr(found + 1);
|
|
value.erase(0, value.find_first_not_of("\t "));
|
|
value.resize(std::min<size_t>(value.size(), value.find_last_not_of("\t\n\r ") + 1));
|
|
header[line.substr(0, found)] = value;
|
|
}
|
|
}
|
|
}
|
|
|
|
return header;
|
|
}
|
|
|
|
std::vector<std::string> split(const std::string& to_split, char delimiter) {
|
|
std::vector<std::string> tokens;
|
|
|
|
std::stringstream stream(to_split);
|
|
std::string item;
|
|
while (std::getline(stream, item, delimiter)) {
|
|
tokens.push_back(item);
|
|
}
|
|
|
|
return tokens;
|
|
}
|
|
|
|
size_t readUserFunction(char* ptr, size_t size, size_t nitems, const ReadCallback* read) {
|
|
size *= nitems;
|
|
return read->callback(ptr, size) ? size : CURL_READFUNC_ABORT;
|
|
}
|
|
|
|
size_t headerUserFunction(char* ptr, size_t size, size_t nmemb, const HeaderCallback* header) {
|
|
size *= nmemb;
|
|
return header->callback({ptr, size}) ? size : 0;
|
|
}
|
|
|
|
size_t writeFunction(char* ptr, size_t size, size_t nmemb, std::string* data) {
|
|
size *= nmemb;
|
|
data->append(ptr, size);
|
|
return size;
|
|
}
|
|
|
|
size_t writeFileFunction(char* ptr, size_t size, size_t nmemb, std::ofstream* file) {
|
|
size *= nmemb;
|
|
file->write(ptr, size);
|
|
return size;
|
|
}
|
|
|
|
size_t writeUserFunction(char* ptr, size_t size, size_t nmemb, const WriteCallback* write) {
|
|
size *= nmemb;
|
|
return write->callback({ptr, size}) ? size : 0;
|
|
}
|
|
|
|
#if LIBCURL_VERSION_NUM < 0x072000
|
|
int progressUserFunction(const ProgressCallback* progress, double dltotal, double dlnow,
|
|
double ultotal, double ulnow) {
|
|
#else
|
|
int progressUserFunction(const ProgressCallback* progress, curl_off_t dltotal, curl_off_t dlnow,
|
|
curl_off_t ultotal, curl_off_t ulnow) {
|
|
#endif
|
|
return progress->callback(dltotal, dlnow, ultotal, ulnow) ? 0 : 1;
|
|
}
|
|
|
|
int debugUserFunction(CURL* /*handle*/, curl_infotype type, char* data, size_t size,
|
|
const DebugCallback* debug) {
|
|
debug->callback(DebugCallback::InfoType(type), std::string(data, size));
|
|
return 0;
|
|
}
|
|
|
|
/**
|
|
* Creates a temporary CurlHolder object and uses it to escape the given string.
|
|
* If you plan to use this methode on a regular basis think about creating a CurlHolder
|
|
* object and calling urlEncode(std::string) on it.
|
|
*
|
|
* Example:
|
|
* CurlHolder holder;
|
|
* std::string input = "Hello World!";
|
|
* std::string result = holder.urlEncode(input);
|
|
**/
|
|
std::string urlEncode(const std::string& s) {
|
|
CurlHolder holder; // Create a temporary new holder for URL encoding
|
|
return holder.urlEncode(s);
|
|
}
|
|
|
|
/**
|
|
* Creates a temporary CurlHolder object and uses it to unescape the given string.
|
|
* If you plan to use this methode on a regular basis think about creating a CurlHolder
|
|
* object and calling urlDecode(std::string) on it.
|
|
*
|
|
* Example:
|
|
* CurlHolder holder;
|
|
* std::string input = "Hello%20World%21";
|
|
* std::string result = holder.urlDecode(input);
|
|
**/
|
|
std::string urlDecode(const std::string& s) {
|
|
CurlHolder holder; // Create a temporary new holder for URL decoding
|
|
return holder.urlDecode(s);
|
|
}
|
|
|
|
} // namespace util
|
|
} // namespace cpr
|