rainmeter-studio/Plugins/PluginWebParser/WebParser.cpp

1816 lines
46 KiB
C++
Raw Normal View History

2009-02-14 10:11:28 +00:00
/*
Copyright (C) 2005 Kimmo Pekkola
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
2009-02-10 18:37:48 +00:00
*/
// Note: To compile this you need the PCRE library (http://www.pcre.org/).
// See: http://www.perldoc.com/perl5.8.0/pod/perlre.html
#include <windows.h>
#include <math.h>
#include <string>
#include <map>
#include <vector>
#include <Wininet.h>
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
#include <shlwapi.h>
#include <process.h>
2011-05-16 09:24:13 +00:00
#include "../../Library/pcre-8.10/config.h"
2011-02-03 18:09:24 +00:00
#include "../../Library/pcre-8.10/pcre.h"
#include "../../Library/Export.h" // Rainmeter's exported functions
2009-02-10 18:37:48 +00:00
2011-02-03 18:09:24 +00:00
#include "../../Library/DisableThreadLibraryCalls.h" // contains DllMain entry point
2009-02-10 18:37:48 +00:00
/* The exported functions */
extern "C"
{
__declspec( dllexport ) UINT Initialize(HMODULE instance, LPCTSTR iniFile, LPCTSTR section, UINT id);
__declspec( dllexport ) double Update2(UINT id);
__declspec( dllexport ) LPCTSTR GetString(UINT id, UINT flags);
__declspec( dllexport ) void Finalize(HMODULE instance, UINT id);
__declspec( dllexport ) UINT GetPluginVersion();
__declspec( dllexport ) LPCTSTR GetPluginAuthor();
}
struct UrlData
{
std::wstring url;
std::wstring regExp;
std::wstring resultString;
std::wstring errorString;
std::wstring proxy;
int codepage;
int stringIndex;
int stringIndex2;
int decodeCharacterReference;
2009-02-10 18:37:48 +00:00
UINT updateRate;
UINT updateCounter;
std::wstring section;
std::wstring finishAction;
bool download;
bool forceReload;
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
std::wstring downloadFile;
2009-02-10 18:37:48 +00:00
std::wstring downloadedFile;
std::wstring iniFile;
int debug;
std::wstring debugFileLocation;
2009-02-10 18:37:48 +00:00
HANDLE threadHandle;
HANDLE dlThreadHandle;
};
BYTE* DownloadUrl(std::wstring& url, DWORD* dwSize, bool forceReload);
2009-02-10 18:37:48 +00:00
void ShowError(int lineNumber, WCHAR* errorMsg = NULL);
unsigned __stdcall NetworkThreadProc(void* pParam);
unsigned __stdcall NetworkDownloadThreadProc(void* pParam);
2009-02-10 18:37:48 +00:00
void Log(const WCHAR* string);
void ParseData(UrlData* urlData, LPCSTR parseData);
2011-03-29 19:21:57 +00:00
CRITICAL_SECTION g_CriticalSection;
2009-02-10 18:37:48 +00:00
bool g_Initialized = false;
static std::map<UINT, UrlData*> g_UrlData;
static bool g_Debug = false;
static HINTERNET hRootHandle = NULL;
static std::map<std::wstring, WCHAR> g_CERs;
2009-02-10 18:37:48 +00:00
#define OVECCOUNT 300 // should be a multiple of 3
std::wstring ConvertToWide(LPCSTR str, int codepage)
2009-02-10 18:37:48 +00:00
{
std::wstring szWide;
2009-02-10 18:37:48 +00:00
if (str && *str)
2009-02-10 18:37:48 +00:00
{
int strLen = (int)strlen(str) + 1;
int bufLen = MultiByteToWideChar(codepage, 0, str, strLen, NULL, 0);
if (bufLen > 0)
{
WCHAR* wideSz = new WCHAR[bufLen];
wideSz[0] = 0;
MultiByteToWideChar(codepage, 0, str, strLen, wideSz, bufLen);
szWide = wideSz;
delete [] wideSz;
}
2009-02-10 18:37:48 +00:00
}
return szWide;
}
std::string ConvertToAscii(LPCWSTR str, int codepage)
{
std::string szAscii;
if (str && *str)
{
int strLen = (int)wcslen(str) + 1;
int bufLen = WideCharToMultiByte(codepage, 0, str, strLen, NULL, 0, NULL, NULL);
if (bufLen > 0)
{
char* tmpSz = new char[bufLen];
tmpSz[0] = 0;
WideCharToMultiByte(codepage, 0, str, strLen, tmpSz, bufLen, NULL, NULL);
szAscii = tmpSz;
delete [] tmpSz;
}
}
return szAscii;
}
std::string ConvertWideToAscii(LPCWSTR str)
{
return ConvertToAscii(str, CP_ACP);
}
std::wstring ConvertAsciiToWide(LPCSTR str)
{
return ConvertToWide(str, CP_ACP);
}
std::wstring ConvertUTF8ToWide(LPCSTR str)
{
return ConvertToWide(str, CP_UTF8);
}
std::string ConvertWideToUTF8(LPCWSTR str)
{
return ConvertToAscii(str, CP_UTF8);
}
std::string ConvertAsciiToUTF8(LPCSTR str, int codepage)
{
std::string szUTF8;
if (str && *str)
{
std::wstring wide = ConvertToWide(str, codepage);
if (!wide.empty())
{
szUTF8.swap(ConvertWideToUTF8(wide.c_str()));
}
}
return szUTF8;
}
std::wstring& DecodeReferences(std::wstring& str, int opt)
{
// (opt <= 0 || opt > 3) : Do nothing.
// (opt == 1) : Decode both numeric character references and character entity references.
// (opt == 2) : Decode only numeric character references.
// (opt == 3) : Decode only character entity references.
if (opt >= 1 && opt <= 3)
{
std::wstring::size_type start = 0;
while ((start = str.find(L'&', start)) != std::wstring::npos)
{
std::wstring::size_type end, pos;
if ((end = str.find(L';', start)) == std::wstring::npos) break;
pos = start + 1;
if (pos == end) // &; - skip
{
start = end + 1;
continue;
}
else if ((end - pos) > 10) // name (or number) is too long
{
++start;
continue;
}
if (str[pos] == L'#') // Numeric character reference
{
if (opt == 3 || // Decode only character entity references,
++pos == end) // &#; - skip
{
start = end + 1;
continue;
}
int base;
if (str[pos] == L'x' || str[pos] == L'X')
{
if (++pos == end) // &#x; or &#X; - skip
{
start = end + 1;
continue;
}
base = 16;
}
else
{
base = 10;
}
std::wstring num(str, pos, end - pos);
WCHAR* pch = NULL;
errno = 0;
long ch = wcstol(num.c_str(), &pch, base);
if (pch == NULL || *pch != L'\0' || errno == ERANGE || ch <= 0 || ch >= 0xFFFE) // invalid character
{
start = pos;
continue;
}
str.replace(start, end - start + 1, 1, (WCHAR)ch);
++start;
}
else // Character entity reference
{
if (opt == 2) // Decode only numeric character references - skip
{
start = end + 1;
continue;
}
std::wstring name(str, pos, end - pos);
std::map<std::wstring, WCHAR>::const_iterator iter = g_CERs.find(name);
if (iter != g_CERs.end())
{
str.replace(start, end - start + 1, 1, (*iter).second);
}
++start;
}
}
}
return str;
}
void FillCharacterEntityReferences()
{
// List from:
// http://www.w3.org/TR/html4/sgml/entities.html
// http://www.w3.org/TR/xhtml1/#C_16
// for markup-significant and internationalization characters
g_CERs[L"quot"] = (WCHAR)34;
g_CERs[L"amp"] = (WCHAR)38;
g_CERs[L"apos"] = (WCHAR)39;
g_CERs[L"lt"] = (WCHAR)60;
g_CERs[L"gt"] = (WCHAR)62;
g_CERs[L"OElig"] = (WCHAR)338;
g_CERs[L"oelig"] = (WCHAR)339;
g_CERs[L"Scaron"] = (WCHAR)352;
g_CERs[L"scaron"] = (WCHAR)353;
g_CERs[L"Yuml"] = (WCHAR)376;
g_CERs[L"circ"] = (WCHAR)710;
g_CERs[L"tilde"] = (WCHAR)732;
g_CERs[L"ensp"] = (WCHAR)8194;
g_CERs[L"emsp"] = (WCHAR)8195;
g_CERs[L"thinsp"] = (WCHAR)8201;
g_CERs[L"zwnj"] = (WCHAR)8204;
g_CERs[L"zwj"] = (WCHAR)8205;
g_CERs[L"lrm"] = (WCHAR)8206;
g_CERs[L"rlm"] = (WCHAR)8207;
g_CERs[L"ndash"] = (WCHAR)8211;
g_CERs[L"mdash"] = (WCHAR)8212;
g_CERs[L"lsquo"] = (WCHAR)8216;
g_CERs[L"rsquo"] = (WCHAR)8217;
g_CERs[L"sbquo"] = (WCHAR)8218;
g_CERs[L"ldquo"] = (WCHAR)8220;
g_CERs[L"rdquo"] = (WCHAR)8221;
g_CERs[L"bdquo"] = (WCHAR)8222;
g_CERs[L"dagger"] = (WCHAR)8224;
g_CERs[L"Dagger"] = (WCHAR)8225;
g_CERs[L"permil"] = (WCHAR)8240;
g_CERs[L"lsaquo"] = (WCHAR)8249;
g_CERs[L"rsaquo"] = (WCHAR)8250;
g_CERs[L"euro"] = (WCHAR)8364;
2011-03-29 19:21:57 +00:00
// for ISO 8859-1 characters
g_CERs[L"nbsp"] = (WCHAR)160;
g_CERs[L"iexcl"] = (WCHAR)161;
g_CERs[L"cent"] = (WCHAR)162;
g_CERs[L"pound"] = (WCHAR)163;
g_CERs[L"curren"] = (WCHAR)164;
g_CERs[L"yen"] = (WCHAR)165;
g_CERs[L"brvbar"] = (WCHAR)166;
g_CERs[L"sect"] = (WCHAR)167;
g_CERs[L"uml"] = (WCHAR)168;
g_CERs[L"copy"] = (WCHAR)169;
g_CERs[L"ordf"] = (WCHAR)170;
g_CERs[L"laquo"] = (WCHAR)171;
g_CERs[L"not"] = (WCHAR)172;
g_CERs[L"shy"] = (WCHAR)173;
g_CERs[L"reg"] = (WCHAR)174;
g_CERs[L"macr"] = (WCHAR)175;
g_CERs[L"deg"] = (WCHAR)176;
g_CERs[L"plusmn"] = (WCHAR)177;
g_CERs[L"sup2"] = (WCHAR)178;
g_CERs[L"sup3"] = (WCHAR)179;
g_CERs[L"acute"] = (WCHAR)180;
g_CERs[L"micro"] = (WCHAR)181;
g_CERs[L"para"] = (WCHAR)182;
g_CERs[L"middot"] = (WCHAR)183;
g_CERs[L"cedil"] = (WCHAR)184;
g_CERs[L"sup1"] = (WCHAR)185;
g_CERs[L"ordm"] = (WCHAR)186;
g_CERs[L"raquo"] = (WCHAR)187;
g_CERs[L"frac14"] = (WCHAR)188;
g_CERs[L"frac12"] = (WCHAR)189;
g_CERs[L"frac34"] = (WCHAR)190;
g_CERs[L"iquest"] = (WCHAR)191;
g_CERs[L"Agrave"] = (WCHAR)192;
g_CERs[L"Aacute"] = (WCHAR)193;
g_CERs[L"Acirc"] = (WCHAR)194;
g_CERs[L"Atilde"] = (WCHAR)195;
g_CERs[L"Auml"] = (WCHAR)196;
g_CERs[L"Aring"] = (WCHAR)197;
g_CERs[L"AElig"] = (WCHAR)198;
g_CERs[L"Ccedil"] = (WCHAR)199;
g_CERs[L"Egrave"] = (WCHAR)200;
g_CERs[L"Eacute"] = (WCHAR)201;
g_CERs[L"Ecirc"] = (WCHAR)202;
g_CERs[L"Euml"] = (WCHAR)203;
g_CERs[L"Igrave"] = (WCHAR)204;
g_CERs[L"Iacute"] = (WCHAR)205;
g_CERs[L"Icirc"] = (WCHAR)206;
g_CERs[L"Iuml"] = (WCHAR)207;
g_CERs[L"ETH"] = (WCHAR)208;
g_CERs[L"Ntilde"] = (WCHAR)209;
g_CERs[L"Ograve"] = (WCHAR)210;
g_CERs[L"Oacute"] = (WCHAR)211;
g_CERs[L"Ocirc"] = (WCHAR)212;
g_CERs[L"Otilde"] = (WCHAR)213;
g_CERs[L"Ouml"] = (WCHAR)214;
g_CERs[L"times"] = (WCHAR)215;
g_CERs[L"Oslash"] = (WCHAR)216;
g_CERs[L"Ugrave"] = (WCHAR)217;
g_CERs[L"Uacute"] = (WCHAR)218;
g_CERs[L"Ucirc"] = (WCHAR)219;
g_CERs[L"Uuml"] = (WCHAR)220;
g_CERs[L"Yacute"] = (WCHAR)221;
g_CERs[L"THORN"] = (WCHAR)222;
g_CERs[L"szlig"] = (WCHAR)223;
g_CERs[L"agrave"] = (WCHAR)224;
g_CERs[L"aacute"] = (WCHAR)225;
g_CERs[L"acirc"] = (WCHAR)226;
g_CERs[L"atilde"] = (WCHAR)227;
g_CERs[L"auml"] = (WCHAR)228;
g_CERs[L"aring"] = (WCHAR)229;
g_CERs[L"aelig"] = (WCHAR)230;
g_CERs[L"ccedil"] = (WCHAR)231;
g_CERs[L"egrave"] = (WCHAR)232;
g_CERs[L"eacute"] = (WCHAR)233;
g_CERs[L"ecirc"] = (WCHAR)234;
g_CERs[L"euml"] = (WCHAR)235;
g_CERs[L"igrave"] = (WCHAR)236;
g_CERs[L"iacute"] = (WCHAR)237;
g_CERs[L"icirc"] = (WCHAR)238;
g_CERs[L"iuml"] = (WCHAR)239;
g_CERs[L"eth"] = (WCHAR)240;
g_CERs[L"ntilde"] = (WCHAR)241;
g_CERs[L"ograve"] = (WCHAR)242;
g_CERs[L"oacute"] = (WCHAR)243;
g_CERs[L"ocirc"] = (WCHAR)244;
g_CERs[L"otilde"] = (WCHAR)245;
g_CERs[L"ouml"] = (WCHAR)246;
g_CERs[L"divide"] = (WCHAR)247;
g_CERs[L"oslash"] = (WCHAR)248;
g_CERs[L"ugrave"] = (WCHAR)249;
g_CERs[L"uacute"] = (WCHAR)250;
g_CERs[L"ucirc"] = (WCHAR)251;
g_CERs[L"uuml"] = (WCHAR)252;
g_CERs[L"yacute"] = (WCHAR)253;
g_CERs[L"thorn"] = (WCHAR)254;
g_CERs[L"yuml"] = (WCHAR)255;
2011-03-29 19:21:57 +00:00
// for symbols, mathematical symbols, and Greek letters
g_CERs[L"fnof"] = (WCHAR)402;
g_CERs[L"Alpha"] = (WCHAR)913;
g_CERs[L"Beta"] = (WCHAR)914;
g_CERs[L"Gamma"] = (WCHAR)915;
g_CERs[L"Delta"] = (WCHAR)916;
g_CERs[L"Epsilon"] = (WCHAR)917;
g_CERs[L"Zeta"] = (WCHAR)918;
g_CERs[L"Eta"] = (WCHAR)919;
g_CERs[L"Theta"] = (WCHAR)920;
g_CERs[L"Iota"] = (WCHAR)921;
g_CERs[L"Kappa"] = (WCHAR)922;
g_CERs[L"Lambda"] = (WCHAR)923;
g_CERs[L"Mu"] = (WCHAR)924;
g_CERs[L"Nu"] = (WCHAR)925;
g_CERs[L"Xi"] = (WCHAR)926;
g_CERs[L"Omicron"] = (WCHAR)927;
g_CERs[L"Pi"] = (WCHAR)928;
g_CERs[L"Rho"] = (WCHAR)929;
g_CERs[L"Sigma"] = (WCHAR)931;
g_CERs[L"Tau"] = (WCHAR)932;
g_CERs[L"Upsilon"] = (WCHAR)933;
g_CERs[L"Phi"] = (WCHAR)934;
g_CERs[L"Chi"] = (WCHAR)935;
g_CERs[L"Psi"] = (WCHAR)936;
g_CERs[L"Omega"] = (WCHAR)937;
g_CERs[L"alpha"] = (WCHAR)945;
g_CERs[L"beta"] = (WCHAR)946;
g_CERs[L"gamma"] = (WCHAR)947;
g_CERs[L"delta"] = (WCHAR)948;
g_CERs[L"epsilon"] = (WCHAR)949;
g_CERs[L"zeta"] = (WCHAR)950;
g_CERs[L"eta"] = (WCHAR)951;
g_CERs[L"theta"] = (WCHAR)952;
g_CERs[L"iota"] = (WCHAR)953;
g_CERs[L"kappa"] = (WCHAR)954;
g_CERs[L"lambda"] = (WCHAR)955;
g_CERs[L"mu"] = (WCHAR)956;
g_CERs[L"nu"] = (WCHAR)957;
g_CERs[L"xi"] = (WCHAR)958;
g_CERs[L"omicron"] = (WCHAR)959;
g_CERs[L"pi"] = (WCHAR)960;
g_CERs[L"rho"] = (WCHAR)961;
g_CERs[L"sigmaf"] = (WCHAR)962;
g_CERs[L"sigma"] = (WCHAR)963;
g_CERs[L"tau"] = (WCHAR)964;
g_CERs[L"upsilon"] = (WCHAR)965;
g_CERs[L"phi"] = (WCHAR)966;
g_CERs[L"chi"] = (WCHAR)967;
g_CERs[L"psi"] = (WCHAR)968;
g_CERs[L"omega"] = (WCHAR)969;
g_CERs[L"thetasym"] = (WCHAR)977;
g_CERs[L"upsih"] = (WCHAR)978;
g_CERs[L"piv"] = (WCHAR)982;
g_CERs[L"bull"] = (WCHAR)8226;
g_CERs[L"hellip"] = (WCHAR)8230;
g_CERs[L"prime"] = (WCHAR)8242;
g_CERs[L"Prime"] = (WCHAR)8243;
g_CERs[L"oline"] = (WCHAR)8254;
g_CERs[L"frasl"] = (WCHAR)8260;
g_CERs[L"weierp"] = (WCHAR)8472;
g_CERs[L"image"] = (WCHAR)8465;
g_CERs[L"real"] = (WCHAR)8476;
g_CERs[L"trade"] = (WCHAR)8482;
g_CERs[L"alefsym"] = (WCHAR)8501;
g_CERs[L"larr"] = (WCHAR)8592;
g_CERs[L"uarr"] = (WCHAR)8593;
g_CERs[L"rarr"] = (WCHAR)8594;
g_CERs[L"darr"] = (WCHAR)8595;
g_CERs[L"harr"] = (WCHAR)8596;
g_CERs[L"crarr"] = (WCHAR)8629;
g_CERs[L"lArr"] = (WCHAR)8656;
g_CERs[L"uArr"] = (WCHAR)8657;
g_CERs[L"rArr"] = (WCHAR)8658;
g_CERs[L"dArr"] = (WCHAR)8659;
g_CERs[L"hArr"] = (WCHAR)8660;
g_CERs[L"forall"] = (WCHAR)8704;
g_CERs[L"part"] = (WCHAR)8706;
g_CERs[L"exist"] = (WCHAR)8707;
g_CERs[L"empty"] = (WCHAR)8709;
g_CERs[L"nabla"] = (WCHAR)8711;
g_CERs[L"isin"] = (WCHAR)8712;
g_CERs[L"notin"] = (WCHAR)8713;
g_CERs[L"ni"] = (WCHAR)8715;
g_CERs[L"prod"] = (WCHAR)8719;
g_CERs[L"sum"] = (WCHAR)8721;
g_CERs[L"minus"] = (WCHAR)8722;
g_CERs[L"lowast"] = (WCHAR)8727;
g_CERs[L"radic"] = (WCHAR)8730;
g_CERs[L"prop"] = (WCHAR)8733;
g_CERs[L"infin"] = (WCHAR)8734;
g_CERs[L"ang"] = (WCHAR)8736;
g_CERs[L"and"] = (WCHAR)8743;
g_CERs[L"or"] = (WCHAR)8744;
g_CERs[L"cap"] = (WCHAR)8745;
g_CERs[L"cup"] = (WCHAR)8746;
g_CERs[L"int"] = (WCHAR)8747;
g_CERs[L"there4"] = (WCHAR)8756;
g_CERs[L"sim"] = (WCHAR)8764;
g_CERs[L"cong"] = (WCHAR)8773;
g_CERs[L"asymp"] = (WCHAR)8776;
g_CERs[L"ne"] = (WCHAR)8800;
g_CERs[L"equiv"] = (WCHAR)8801;
g_CERs[L"le"] = (WCHAR)8804;
g_CERs[L"ge"] = (WCHAR)8805;
g_CERs[L"sub"] = (WCHAR)8834;
g_CERs[L"sup"] = (WCHAR)8835;
g_CERs[L"nsub"] = (WCHAR)8836;
g_CERs[L"sube"] = (WCHAR)8838;
g_CERs[L"supe"] = (WCHAR)8839;
g_CERs[L"oplus"] = (WCHAR)8853;
g_CERs[L"otimes"] = (WCHAR)8855;
g_CERs[L"perp"] = (WCHAR)8869;
g_CERs[L"sdot"] = (WCHAR)8901;
g_CERs[L"lceil"] = (WCHAR)8968;
g_CERs[L"rceil"] = (WCHAR)8969;
g_CERs[L"lfloor"] = (WCHAR)8970;
g_CERs[L"rfloor"] = (WCHAR)8971;
g_CERs[L"lang"] = (WCHAR)9001;
g_CERs[L"rang"] = (WCHAR)9002;
g_CERs[L"loz"] = (WCHAR)9674;
g_CERs[L"spades"] = (WCHAR)9824;
g_CERs[L"clubs"] = (WCHAR)9827;
g_CERs[L"hearts"] = (WCHAR)9829;
g_CERs[L"diams"] = (WCHAR)9830;
// for DEBUG
//std::map<std::wstring, WCHAR>::const_iterator iter = g_CERs.begin();
//for ( ; iter != g_CERs.end(); ++iter)
//{
// WCHAR buffer[64];
// wsprintf(buffer, L"%s - %c", (*iter).first.c_str(), (*iter).second);
// Log(buffer);
//}
}
bool BelongToSameProcess(HWND wnd)
{
DWORD procId = 0;
GetWindowThreadProcessId(wnd, &procId);
return (procId == GetCurrentProcessId());
}
HWND FindMeterWindow(HWND parent)
{
HWND wnd = NULL;
while (wnd = FindWindowEx(parent, wnd, L"RainmeterMeterWindow", NULL))
{
if (BelongToSameProcess(wnd))
{
return wnd;
* Changed the way to get the information of the multiple display monitors. This change brings the order of monitors close to the order of "Display Properties" due to using EnumDisplayDevices and EnumDisplaySettings instead of EnumDisplayMonitors. (If EnumDisplayDevices failed, EnumDisplayMonitors is used as before.) ----- * Added the "Display Monitor" submenu in [Skins Menu]-[Position]. These menus convert the present position to the relative position from the specified monitor. (But the meter window doesn't move to the specified monitor area immediately. Only converts.) - "Use default: Primary monitor" removes the @-directive from WindowX/Y. - @0(@1, @2, ...) adds the specified monitor number to WindowX/Y. @0 means "The Virtual Screen". (http://msdn.microsoft.com/en-us/library/dd145136%28VS.85%29.aspx) - If "Auto-select based on window position" is checked, the WindowX and WindowY "@n" settings are made automatically based on the position of the meter's window. If a monitor is selected directly using "Display Monitor" in the Rainmeter / skin context menu, this menu is unchecked. This setting can be manually made in either the [Rainmeter] (all configs) or individual config sections of Rainmeter.ini. AutoSelectScreen If set to 1, the WindowX and WindowY "@n" settings are made automatically based on the position of the meter's window. If a monitor is selected directly using "Display Monitor" in the Rainmeter / skin context menu, this setting is reset to 0. ----- * Added the variables for multiple display monitors and the virtual screen. All X/Y positions are represented in the virtual screen coordinates. The following variables are for the virtual screen. #VSCREENAREAX# is the X-position of the left-side of the virtual screen. #VSCREENAREAY# is the Y-position of the top-side of the virtual screen. #VSCREENAREAWIDTH# is the width of the virtual screen. #VSCREENAREAHEIGHT# is the height of the virtual screen. The following variables are for the PRESENT monitor. Note that these variables automatically change by the WindowX and WindowY "@n" settings. If "@n" is not set, these variables return the value of the primary monitor. #WORKAREAX# is the X-position of the left-side of the work area. #WORKAREAY# is the Y-position of the top-side of the work area. #WORKAREAWIDTH# is the width of the work area. #WORKAREAHEIGHT# is the height of the work area. #SCREENAREAX# is the X-position of the left-side of the monitor screen. #SCREENAREAY# is the Y-position of the top-side of the monitor screen. #SCREENAREAWIDTH# is the width of the display resolution. #SCREENAREAHEIGHT# is the height of the display resolution. The following variables are for the PRIMARY monitor. #PWORKAREAX# is the X-position of the left-side of the work area. #PWORKAREAY# is the Y-position of the top-side of the work area. #PWORKAREAWIDTH# is the width of the work area. #PWORKAREAHEIGHT# is the height of the work area. #PSCREENAREAX# is the X-position of the left-side of the monitor screen. (maybe, always 0) #PSCREENAREAY# is the Y-position of the top-side of the monitor screen. (maybe, always 0) #PSCREENAREAWIDTH# is the width of the display resolution. #PSCREENAREAHEIGHT# is the height of the display resolution. The following variables are for the SPECIFIED monitor. (@n = @1, @2, ...) #WORKAREAX@n# is the X-position of the left-side of the work area. #WORKAREAY@n# is the Y-position of the top-side of the work area. #WORKAREAWIDTH@n# is the width of the work area. #WORKAREAHEIGHT@n# is the height of the work area. #SCREENAREAX@n# is the X-position of the left-side of the monitor screen. #SCREENAREAY@n# is the Y-position of the top-side of the monitor screen. #SCREENAREAWIDTH@n# is the width of the display resolution. #SCREENAREAHEIGHT@n# is the height of the display resolution. ----- * Other related changes: - Fixed the problem that the primary monitor isn't recognized correctly. - Fixed the problem that the information of the multiple display monitors is refreshed excessively. - For DynamicVariables, when display setting or workarea size has been changed, all variables are now updated to apply changed WORKAREA/SCREENAREA variables. - Fixed the problem that the "On Desktop" window isn't dragged correctly when the top-left corner of the virtual screen has negative coordinates. - Changed the way to stick the "On Desktop" window. ("SysListView32/FolderView" is used instead of "Progman/Program Manager".) ----- * Other changes: - When the meter window is draggable and isn't dragged, LeftMouseUpAction is now executed. - Added MouseDoubleClickAction (LeftMouseDoubleClickAction, RightMouseDoubleClickAction, MiddleMouseDoubleClickAction). If MouseDoubleClickAction is empty when mouse button is double-clicked, MouseDownAction is executed instead. - Fixed the problem that the Meter's hit-test code checks outside the area. - Changed the way to set the #CURRENTCONFIG#. (CMeterWindow::GetSkinName() is now used instead of parsing the path.)
2009-12-18 05:58:37 +00:00
}
}
* Changed the way to get the information of the multiple display monitors. This change brings the order of monitors close to the order of "Display Properties" due to using EnumDisplayDevices and EnumDisplaySettings instead of EnumDisplayMonitors. (If EnumDisplayDevices failed, EnumDisplayMonitors is used as before.) ----- * Added the "Display Monitor" submenu in [Skins Menu]-[Position]. These menus convert the present position to the relative position from the specified monitor. (But the meter window doesn't move to the specified monitor area immediately. Only converts.) - "Use default: Primary monitor" removes the @-directive from WindowX/Y. - @0(@1, @2, ...) adds the specified monitor number to WindowX/Y. @0 means "The Virtual Screen". (http://msdn.microsoft.com/en-us/library/dd145136%28VS.85%29.aspx) - If "Auto-select based on window position" is checked, the WindowX and WindowY "@n" settings are made automatically based on the position of the meter's window. If a monitor is selected directly using "Display Monitor" in the Rainmeter / skin context menu, this menu is unchecked. This setting can be manually made in either the [Rainmeter] (all configs) or individual config sections of Rainmeter.ini. AutoSelectScreen If set to 1, the WindowX and WindowY "@n" settings are made automatically based on the position of the meter's window. If a monitor is selected directly using "Display Monitor" in the Rainmeter / skin context menu, this setting is reset to 0. ----- * Added the variables for multiple display monitors and the virtual screen. All X/Y positions are represented in the virtual screen coordinates. The following variables are for the virtual screen. #VSCREENAREAX# is the X-position of the left-side of the virtual screen. #VSCREENAREAY# is the Y-position of the top-side of the virtual screen. #VSCREENAREAWIDTH# is the width of the virtual screen. #VSCREENAREAHEIGHT# is the height of the virtual screen. The following variables are for the PRESENT monitor. Note that these variables automatically change by the WindowX and WindowY "@n" settings. If "@n" is not set, these variables return the value of the primary monitor. #WORKAREAX# is the X-position of the left-side of the work area. #WORKAREAY# is the Y-position of the top-side of the work area. #WORKAREAWIDTH# is the width of the work area. #WORKAREAHEIGHT# is the height of the work area. #SCREENAREAX# is the X-position of the left-side of the monitor screen. #SCREENAREAY# is the Y-position of the top-side of the monitor screen. #SCREENAREAWIDTH# is the width of the display resolution. #SCREENAREAHEIGHT# is the height of the display resolution. The following variables are for the PRIMARY monitor. #PWORKAREAX# is the X-position of the left-side of the work area. #PWORKAREAY# is the Y-position of the top-side of the work area. #PWORKAREAWIDTH# is the width of the work area. #PWORKAREAHEIGHT# is the height of the work area. #PSCREENAREAX# is the X-position of the left-side of the monitor screen. (maybe, always 0) #PSCREENAREAY# is the Y-position of the top-side of the monitor screen. (maybe, always 0) #PSCREENAREAWIDTH# is the width of the display resolution. #PSCREENAREAHEIGHT# is the height of the display resolution. The following variables are for the SPECIFIED monitor. (@n = @1, @2, ...) #WORKAREAX@n# is the X-position of the left-side of the work area. #WORKAREAY@n# is the Y-position of the top-side of the work area. #WORKAREAWIDTH@n# is the width of the work area. #WORKAREAHEIGHT@n# is the height of the work area. #SCREENAREAX@n# is the X-position of the left-side of the monitor screen. #SCREENAREAY@n# is the Y-position of the top-side of the monitor screen. #SCREENAREAWIDTH@n# is the width of the display resolution. #SCREENAREAHEIGHT@n# is the height of the display resolution. ----- * Other related changes: - Fixed the problem that the primary monitor isn't recognized correctly. - Fixed the problem that the information of the multiple display monitors is refreshed excessively. - For DynamicVariables, when display setting or workarea size has been changed, all variables are now updated to apply changed WORKAREA/SCREENAREA variables. - Fixed the problem that the "On Desktop" window isn't dragged correctly when the top-left corner of the virtual screen has negative coordinates. - Changed the way to stick the "On Desktop" window. ("SysListView32/FolderView" is used instead of "Progman/Program Manager".) ----- * Other changes: - When the meter window is draggable and isn't dragged, LeftMouseUpAction is now executed. - Added MouseDoubleClickAction (LeftMouseDoubleClickAction, RightMouseDoubleClickAction, MiddleMouseDoubleClickAction). If MouseDoubleClickAction is empty when mouse button is double-clicked, MouseDownAction is executed instead. - Fixed the problem that the Meter's hit-test code checks outside the area. - Changed the way to set the #CURRENTCONFIG#. (CMeterWindow::GetSkinName() is now used instead of parsing the path.)
2009-12-18 05:58:37 +00:00
return NULL;
}
HWND FindMeterWindow(const std::wstring& iniFile)
{
std::wstring str = PluginBridge(L"getconfig", iniFile.c_str());
if (!str.empty())
{
str = PluginBridge(L"getwindow", str.c_str());
if (str != L"error")
* Changed the way to get the information of the multiple display monitors. This change brings the order of monitors close to the order of "Display Properties" due to using EnumDisplayDevices and EnumDisplaySettings instead of EnumDisplayMonitors. (If EnumDisplayDevices failed, EnumDisplayMonitors is used as before.) ----- * Added the "Display Monitor" submenu in [Skins Menu]-[Position]. These menus convert the present position to the relative position from the specified monitor. (But the meter window doesn't move to the specified monitor area immediately. Only converts.) - "Use default: Primary monitor" removes the @-directive from WindowX/Y. - @0(@1, @2, ...) adds the specified monitor number to WindowX/Y. @0 means "The Virtual Screen". (http://msdn.microsoft.com/en-us/library/dd145136%28VS.85%29.aspx) - If "Auto-select based on window position" is checked, the WindowX and WindowY "@n" settings are made automatically based on the position of the meter's window. If a monitor is selected directly using "Display Monitor" in the Rainmeter / skin context menu, this menu is unchecked. This setting can be manually made in either the [Rainmeter] (all configs) or individual config sections of Rainmeter.ini. AutoSelectScreen If set to 1, the WindowX and WindowY "@n" settings are made automatically based on the position of the meter's window. If a monitor is selected directly using "Display Monitor" in the Rainmeter / skin context menu, this setting is reset to 0. ----- * Added the variables for multiple display monitors and the virtual screen. All X/Y positions are represented in the virtual screen coordinates. The following variables are for the virtual screen. #VSCREENAREAX# is the X-position of the left-side of the virtual screen. #VSCREENAREAY# is the Y-position of the top-side of the virtual screen. #VSCREENAREAWIDTH# is the width of the virtual screen. #VSCREENAREAHEIGHT# is the height of the virtual screen. The following variables are for the PRESENT monitor. Note that these variables automatically change by the WindowX and WindowY "@n" settings. If "@n" is not set, these variables return the value of the primary monitor. #WORKAREAX# is the X-position of the left-side of the work area. #WORKAREAY# is the Y-position of the top-side of the work area. #WORKAREAWIDTH# is the width of the work area. #WORKAREAHEIGHT# is the height of the work area. #SCREENAREAX# is the X-position of the left-side of the monitor screen. #SCREENAREAY# is the Y-position of the top-side of the monitor screen. #SCREENAREAWIDTH# is the width of the display resolution. #SCREENAREAHEIGHT# is the height of the display resolution. The following variables are for the PRIMARY monitor. #PWORKAREAX# is the X-position of the left-side of the work area. #PWORKAREAY# is the Y-position of the top-side of the work area. #PWORKAREAWIDTH# is the width of the work area. #PWORKAREAHEIGHT# is the height of the work area. #PSCREENAREAX# is the X-position of the left-side of the monitor screen. (maybe, always 0) #PSCREENAREAY# is the Y-position of the top-side of the monitor screen. (maybe, always 0) #PSCREENAREAWIDTH# is the width of the display resolution. #PSCREENAREAHEIGHT# is the height of the display resolution. The following variables are for the SPECIFIED monitor. (@n = @1, @2, ...) #WORKAREAX@n# is the X-position of the left-side of the work area. #WORKAREAY@n# is the Y-position of the top-side of the work area. #WORKAREAWIDTH@n# is the width of the work area. #WORKAREAHEIGHT@n# is the height of the work area. #SCREENAREAX@n# is the X-position of the left-side of the monitor screen. #SCREENAREAY@n# is the Y-position of the top-side of the monitor screen. #SCREENAREAWIDTH@n# is the width of the display resolution. #SCREENAREAHEIGHT@n# is the height of the display resolution. ----- * Other related changes: - Fixed the problem that the primary monitor isn't recognized correctly. - Fixed the problem that the information of the multiple display monitors is refreshed excessively. - For DynamicVariables, when display setting or workarea size has been changed, all variables are now updated to apply changed WORKAREA/SCREENAREA variables. - Fixed the problem that the "On Desktop" window isn't dragged correctly when the top-left corner of the virtual screen has negative coordinates. - Changed the way to stick the "On Desktop" window. ("SysListView32/FolderView" is used instead of "Progman/Program Manager".) ----- * Other changes: - When the meter window is draggable and isn't dragged, LeftMouseUpAction is now executed. - Added MouseDoubleClickAction (LeftMouseDoubleClickAction, RightMouseDoubleClickAction, MiddleMouseDoubleClickAction). If MouseDoubleClickAction is empty when mouse button is double-clicked, MouseDownAction is executed instead. - Fixed the problem that the Meter's hit-test code checks outside the area. - Changed the way to set the #CURRENTCONFIG#. (CMeterWindow::GetSkinName() is now used instead of parsing the path.)
2009-12-18 05:58:37 +00:00
{
return (HWND)UlongToPtr(wcstoul(str.c_str(), NULL, 10));
}
}
return FindMeterWindow(NULL); // Use old way to find
}
2009-02-10 18:37:48 +00:00
/*
This function is called when the measure is initialized.
2011-03-29 19:21:57 +00:00
The function must return the maximum value that can be measured.
2009-02-10 18:37:48 +00:00
The return value can also be 0, which means that Rainmeter will
track the maximum value automatically. The parameters for this
function are:
instance The instance of this DLL
iniFile The name of the ini-file (usually Rainmeter.ini)
section The name of the section in the ini-file for this measure
id The identifier for the measure. This is used to identify the measures that use the same plugin.
*/
UINT Initialize(HMODULE instance, LPCTSTR iniFile, LPCTSTR section, UINT id)
{
2011-03-29 19:21:57 +00:00
LPCTSTR tmpSz;
2009-02-10 18:37:48 +00:00
if (!g_Initialized)
{
InitializeCriticalSection(&g_CriticalSection);
FillCharacterEntityReferences();
2009-02-10 18:37:48 +00:00
g_Initialized = true;
}
UrlData* data = new UrlData;
data->section = section;
data->decodeCharacterReference = 0;
data->updateRate = 600;
2009-02-10 18:37:48 +00:00
data->updateCounter = 0;
data->iniFile = iniFile;
/* Read our own settings from the ini-file */
data->url = ReadConfigString(section, L"Url", L"");
data->regExp = ReadConfigString(section, L"RegExp", L"");
data->finishAction = ReadConfigString(section, L"FinishAction", L"");
data->errorString = ReadConfigString(section, L"ErrorString", L"");
data->proxy = ReadConfigString(section, L"ProxyServer", L"");
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
data->downloadFile = ReadConfigString(section, L"DownloadFile", L"");
data->debugFileLocation = ReadConfigString(section, L"Debug2File", L"c:\\WebParserDump.txt");
2011-03-29 19:21:57 +00:00
if (data->debugFileLocation.find(L"\\") == std::wstring::npos)
{
2011-03-29 19:21:57 +00:00
std::wstring str = data->iniFile.substr(0,data->iniFile.find_last_of(L"\\")+1);
str += data->debugFileLocation;
Log(str.c_str());
data->debugFileLocation = str;
}
2009-02-10 18:37:48 +00:00
tmpSz = ReadConfigString(section, L"StringIndex", L"0");
if (tmpSz)
{
data->stringIndex = _wtoi(tmpSz);
}
tmpSz = ReadConfigString(section, L"StringIndex2", L"0");
if (tmpSz)
{
data->stringIndex2 = _wtoi(tmpSz);
}
tmpSz = ReadConfigString(section, L"DecodeCharacterReference", L"0");
2009-02-10 18:37:48 +00:00
if (tmpSz)
{
data->decodeCharacterReference = _wtoi(tmpSz);
}
tmpSz = ReadConfigString(section, L"UpdateRate", L"600");
if (tmpSz && *tmpSz)
2009-02-10 18:37:48 +00:00
{
data->updateRate = _wtoi(tmpSz);
}
tmpSz = ReadConfigString(section, L"Download", L"0");
if (tmpSz)
{
data->download = 1 == _wtoi(tmpSz);
}
tmpSz = ReadConfigString(section, L"ForceReload", L"0");
if (tmpSz)
{
data->forceReload = 1 == _wtoi(tmpSz);
}
2009-02-10 18:37:48 +00:00
tmpSz = ReadConfigString(section, L"Debug", L"0");
if (tmpSz)
{
data->debug = _wtoi(tmpSz);
}
tmpSz = ReadConfigString(section, L"CodePage", L"0");
if (tmpSz)
{
data->codepage = _wtoi(tmpSz);
}
if (hRootHandle == NULL)
{
if (data->proxy.empty())
{
hRootHandle = InternetOpen(L"Rainmeter WebParser plugin",
INTERNET_OPEN_TYPE_PRECONFIG,
NULL,
NULL,
0);
}
else
{
hRootHandle = InternetOpen(L"Rainmeter WebParser plugin",
INTERNET_OPEN_TYPE_PROXY,
data->proxy.c_str(),
NULL,
0);
}
if (hRootHandle == NULL)
{
ShowError(__LINE__);
delete data;
return 0;
2009-02-10 18:37:48 +00:00
}
}
data->threadHandle = 0;
data->dlThreadHandle = 0;
// During initialization there is no threads yet so no need to do any locking
g_UrlData[id] = data;
return 0;
}
/*
This function is called when new value should be measured.
The function returns the new value.
*/
double Update2(UINT id)
{
double value = 0;
UrlData* urlData = NULL;
// Find the data for this instance (the data structure is not changed by anyone so this should be safe)
std::map<UINT, UrlData*>::iterator urlIter = g_UrlData.find(id);
2011-03-29 19:21:57 +00:00
if (urlIter != g_UrlData.end())
2009-02-10 18:37:48 +00:00
{
urlData = (*urlIter).second;
}
2011-03-29 19:21:57 +00:00
if (urlData)
2009-02-10 18:37:48 +00:00
{
if (urlData->download && urlData->regExp.empty() && urlData->url.find(L'[') == std::wstring::npos)
{
// If RegExp is empty download the file that is pointed by the Url
if (urlData->dlThreadHandle == 0)
{
if (urlData->updateCounter == 0)
{
// Launch a new thread to fetch the web data
unsigned int id;
HANDLE threadHandle = (HANDLE)_beginthreadex(NULL, 0, NetworkDownloadThreadProc, urlData, 0, &id);
if (threadHandle)
{
urlData->dlThreadHandle = threadHandle;
}
else // error
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Failed to begin download thread.";
Log(log.c_str());
}
2009-02-10 18:37:48 +00:00
}
urlData->updateCounter++;
if (urlData->updateCounter >= urlData->updateRate)
{
urlData->updateCounter = 0;
}
}
// Else download the file pointed by the result string (this is done later)
}
else
{
// Make sure that the thread is not writing to the result at the same time
EnterCriticalSection(&g_CriticalSection);
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
if (!urlData->resultString.empty())
{
value = wcstod(urlData->resultString.c_str(), NULL);
}
LeaveCriticalSection(&g_CriticalSection);
if (urlData->url.size() > 0 && urlData->url.find(L'[') == std::wstring::npos)
{
// This is not a reference; need to update.
if (urlData->threadHandle == 0 && urlData->dlThreadHandle == 0)
2009-02-10 18:37:48 +00:00
{
if (urlData->updateCounter == 0)
{
// Launch a new thread to fetch the web data
unsigned int id;
HANDLE threadHandle = (HANDLE)_beginthreadex(NULL, 0, NetworkThreadProc, urlData, 0, &id);
if (threadHandle)
{
urlData->threadHandle = threadHandle;
}
else // error
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Failed to begin thread.";
Log(log.c_str());
}
2009-02-10 18:37:48 +00:00
}
urlData->updateCounter++;
if (urlData->updateCounter >= urlData->updateRate)
{
urlData->updateCounter = 0;
}
}
}
}
}
return value;
}
/*
Thread that fetches the data from the net and parses the page.
*/
unsigned __stdcall NetworkThreadProc(void* pParam)
2009-02-10 18:37:48 +00:00
{
UrlData* urlData = (UrlData*)pParam;
DWORD dwSize = 0;
BYTE* data = DownloadUrl(urlData->url, &dwSize, urlData->forceReload);
2009-02-10 18:37:48 +00:00
if (data)
{
if (urlData->debug == 2)
{
// Dump to a file
FILE* file = _wfopen(urlData->debugFileLocation.c_str(), L"wb");
if (file)
{
fwrite(data, sizeof(BYTE), dwSize, file);
fclose(file);
}
else
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Failed to dump debug data: ";
log += urlData->debugFileLocation;
Log(log.c_str());
}
2009-02-10 18:37:48 +00:00
}
ParseData(urlData, (LPCSTR)data);
delete [] data;
}
EnterCriticalSection(&g_CriticalSection);
CloseHandle(urlData->threadHandle);
urlData->threadHandle = 0;
LeaveCriticalSection(&g_CriticalSection);
2011-03-29 19:21:57 +00:00
return 0; // thread completed successfully
2009-02-10 18:37:48 +00:00
}
void ParseData(UrlData* urlData, LPCSTR parseData)
{
size_t dwSize = 0;
// Parse the value from the data
pcre* re;
const char* error;
int erroffset;
int ovector[OVECCOUNT];
int rc;
int flags = PCRE_UTF8;
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
if (urlData->codepage == 0)
{
flags = 0;
}
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Compile the regular expression in the first argument
re = pcre_compile(
ConvertWideToUTF8(urlData->regExp.c_str()).c_str(), // the pattern
2009-02-10 18:37:48 +00:00
flags, // default options
&error, // for error message
&erroffset, // for error offset
NULL); // use default character tables
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
if (re != NULL)
{
// Compilation succeeded: match the subject in the second argument
std::string utf8Data;
if (urlData->codepage != CP_UTF8 && urlData->codepage != 0) // 0 = CP_ACP
{
// Must convert the data to utf8
utf8Data = ConvertAsciiToUTF8(parseData, urlData->codepage);
2009-02-10 18:37:48 +00:00
parseData = utf8Data.c_str();
}
dwSize = strlen(parseData);
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
rc = pcre_exec(
re, // the compiled pattern
NULL, // no extra data - we didn't study the pattern
parseData, // the subject string
dwSize, // the length of the subject
0, // start at offset 0 in the subject
0, // default options
ovector, // output vector for substring information
OVECCOUNT); // number of elements in the output vector
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
if (rc >= 0)
{
if (rc == 0)
{
// The output vector wasn't big enough
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Too many substrings!";
Log(log.c_str());
2009-02-10 18:37:48 +00:00
}
else
{
if (urlData->stringIndex < rc)
{
if (urlData->debug != 0)
{
for (int i = 0; i < rc; ++i)
2009-02-10 18:37:48 +00:00
{
const char* substring_start = parseData + ovector[2 * i];
2009-02-10 18:37:48 +00:00
int substring_length = ovector[2 * i + 1] - ovector[2 * i];
substring_length = min(substring_length, 256);
std::string tmpStr(substring_start, substring_length);
WCHAR buffer[32];
wsprintf(buffer, L"%2d", i);
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] (Index ";
log += buffer;
log += L") ";
log += ConvertUTF8ToWide(tmpStr.c_str());
Log(log.c_str());
2009-02-10 18:37:48 +00:00
}
}
const char* substring_start = parseData + ovector[2 * urlData->stringIndex];
int substring_length = ovector[2 * urlData->stringIndex + 1] - ovector[2 * urlData->stringIndex];
EnterCriticalSection(&g_CriticalSection);
std::string szResult(substring_start, substring_length);
urlData->resultString = DecodeReferences(ConvertUTF8ToWide(szResult.c_str()), urlData->decodeCharacterReference);
2009-02-10 18:37:48 +00:00
LeaveCriticalSection(&g_CriticalSection);
}
else
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Not enough substrings!";
Log(log.c_str());
2010-08-10 12:21:46 +00:00
2011-03-29 19:21:57 +00:00
// Clear the old result
2010-08-10 12:21:46 +00:00
EnterCriticalSection(&g_CriticalSection);
urlData->resultString.clear();
if (urlData->download)
{
if (urlData->downloadFile.empty()) // cache mode
{
if (!urlData->downloadedFile.empty())
{
// Delete old downloaded file
DeleteFile(urlData->downloadedFile.c_str());
}
}
urlData->downloadedFile.clear();
}
2010-08-10 12:21:46 +00:00
LeaveCriticalSection(&g_CriticalSection);
2009-02-10 18:37:48 +00:00
}
// Update the references
std::map<UINT, UrlData*>::iterator i = g_UrlData.begin();
std::wstring compareStr = L"[";
compareStr += urlData->section;
compareStr += L"]";
for ( ; i != g_UrlData.end(); ++i)
2009-02-10 18:37:48 +00:00
{
if ((((*i).second)->url.find(compareStr) != std::wstring::npos) && (urlData->iniFile == ((*i).second)->iniFile))
{
if (((*i).second)->stringIndex < rc)
{
const char* substring_start = parseData + ovector[2 * ((*i).second)->stringIndex];
int substring_length = ovector[2 * ((*i).second)->stringIndex + 1] - ovector[2 * ((*i).second)->stringIndex];
std::string szResult(substring_start, substring_length);
2009-02-10 18:37:48 +00:00
2011-03-29 19:21:57 +00:00
if (!((*i).second)->regExp.empty())
2009-02-10 18:37:48 +00:00
{
// Change the index and parse the substring
int index = (*i).second->stringIndex;
(*i).second->stringIndex = (*i).second->stringIndex2;
ParseData((*i).second, szResult.c_str());
(*i).second->stringIndex = index;
}
else
{
2011-03-29 19:21:57 +00:00
// Set the result
2009-02-10 18:37:48 +00:00
EnterCriticalSection(&g_CriticalSection);
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Substitude the [measure] with szResult
std::wstring wzResult = ConvertUTF8ToWide(szResult.c_str());
2009-02-10 18:37:48 +00:00
std::wstring wzUrl = ((*i).second)->url;
wzUrl.replace(wzUrl.find(compareStr), compareStr.size(), wzResult);
((*i).second)->resultString = DecodeReferences(wzUrl, (*i).second->decodeCharacterReference);
2009-02-10 18:37:48 +00:00
// Start download threads for the references
if (((*i).second)->download)
{
// Start the download thread
unsigned int id;
HANDLE threadHandle = (HANDLE)_beginthreadex(NULL, 0, NetworkDownloadThreadProc, ((*i).second), 0, &id);
if (threadHandle)
{
((*i).second)->dlThreadHandle = threadHandle;
}
else // error
{
std::wstring log = L"WebParser: [";
log += (*i).second->section;
log += L"] Failed to begin download thread.";
Log(log.c_str());
}
2009-02-10 18:37:48 +00:00
}
LeaveCriticalSection(&g_CriticalSection);
}
}
else
{
std::wstring log = L"WebParser: [";
log += (*i).second->section;
log += L"] Not enough substrings!";
Log(log.c_str());
2011-03-29 19:21:57 +00:00
// Clear the old result
EnterCriticalSection(&g_CriticalSection);
((*i).second)->resultString.clear();
if ((*i).second->download)
{
if ((*i).second->downloadFile.empty()) // cache mode
{
if (!(*i).second->downloadedFile.empty())
{
// Delete old downloaded file
DeleteFile((*i).second->downloadedFile.c_str());
}
}
(*i).second->downloadedFile.clear();
}
LeaveCriticalSection(&g_CriticalSection);
2009-02-10 18:37:48 +00:00
}
}
}
}
}
else
{
// Matching failed: handle error cases
WCHAR buffer[32];
wsprintf(buffer, L"%d", rc);
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Matching error! (";
log += buffer;
log += L")\n";
Log(log.c_str());
2009-02-10 18:37:48 +00:00
EnterCriticalSection(&g_CriticalSection);
urlData->resultString = urlData->errorString;
// Update the references
std::map<UINT, UrlData*>::iterator i = g_UrlData.begin();
std::wstring compareStr = L"[";
compareStr += urlData->section;
compareStr += L"]";
for ( ; i != g_UrlData.end(); ++i)
2009-02-10 18:37:48 +00:00
{
if ((((*i).second)->url.find(compareStr) != std::wstring::npos) && (urlData->iniFile == ((*i).second)->iniFile))
{
((*i).second)->resultString = ((*i).second)->errorString;
}
}
LeaveCriticalSection(&g_CriticalSection);
}
2011-03-29 19:21:57 +00:00
// Release memory used for the compiled pattern
pcre_free(re);
2009-02-10 18:37:48 +00:00
}
else
{
// Compilation failed: print the error message and exit
WCHAR buffer[32];
wsprintf(buffer, L"%d", erroffset);
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] PCRE compilation failed at offset ";
log += buffer;
log += L": ";
log += ConvertAsciiToWide(error);
log += L"\n";
Log(log.c_str());
2009-02-10 18:37:48 +00:00
}
if (urlData->download)
{
// Start the download thread
unsigned int id;
HANDLE threadHandle = (HANDLE)_beginthreadex(NULL, 0, NetworkDownloadThreadProc, urlData, 0, &id);
if (threadHandle)
{
urlData->dlThreadHandle = threadHandle;
}
else // error
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Failed to begin download thread.";
Log(log.c_str());
}
2009-02-10 18:37:48 +00:00
}
else
{
2011-03-29 19:21:57 +00:00
if (!urlData->finishAction.empty())
2009-02-10 18:37:48 +00:00
{
HWND wnd = FindMeterWindow(urlData->iniFile);
2009-02-10 18:37:48 +00:00
if (wnd != NULL)
{
COPYDATASTRUCT copyData;
copyData.dwData = 1;
copyData.cbData = (DWORD)(urlData->finishAction.size() + 1) * sizeof(WCHAR);
copyData.lpData = (void*)urlData->finishAction.c_str();
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
// Send the bang to the Rainmeter window
2009-02-10 18:37:48 +00:00
SendMessage(wnd, WM_COPYDATA, (WPARAM)NULL, (LPARAM)&copyData);
}
}
}
}
/*
Thread that downloads the file from the new.
*/
unsigned __stdcall NetworkDownloadThreadProc(void* pParam)
2009-02-10 18:37:48 +00:00
{
UrlData* urlData = (UrlData*)pParam;
const bool download = !urlData->downloadFile.empty();
bool ready = false;
2009-02-10 18:37:48 +00:00
std::wstring url;
if (urlData->regExp.empty() && urlData->resultString.empty())
{
if (!urlData->url.empty() && urlData->url[0] != L'[')
{
url = urlData->url;
}
2009-02-10 18:37:48 +00:00
}
else
{
EnterCriticalSection(&g_CriticalSection);
url = urlData->resultString;
LeaveCriticalSection(&g_CriticalSection);
2011-03-29 19:21:57 +00:00
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
std::wstring::size_type pos = url.find(L':');
if (pos == std::wstring::npos && !url.empty()) // No protocol
2009-02-10 18:37:48 +00:00
{
// Add the base url to the string
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (url[0] == L'/')
2009-02-10 18:37:48 +00:00
{
// Absolute path
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
pos = urlData->url.find(L'/', 7); // Assume "http://" (=7)
if (pos != std::wstring::npos)
2009-02-10 18:37:48 +00:00
{
std::wstring path(urlData->url.substr(0, pos));
url = path + url;
}
}
else
{
// Relative path
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
pos = urlData->url.rfind(L'/');
if (pos != std::wstring::npos)
2009-02-10 18:37:48 +00:00
{
std::wstring path(urlData->url.substr(0, pos + 1));
url = path + url;
}
}
}
}
if (!url.empty())
{
// Create the filename
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
WCHAR buffer[MAX_PATH] = {0};
std::wstring fullpath, directory;
if (download) // download mode
{
PathCanonicalize(buffer, urlData->downloadFile.c_str());
std::wstring path = buffer;
std::wstring::size_type pos = path.find_first_not_of(L'\\');
if (pos != std::wstring::npos)
{
path = path.substr(pos);
}
PathCanonicalize(buffer, urlData->iniFile.substr(0, urlData->iniFile.find_last_of(L'\\') + 1).c_str()); // "#CURRENTPATH#"
wcscat(buffer, L"DownloadFile\\"); // "#CURRENTPATH#DownloadFile\"
CreateDirectory(buffer, NULL); // Make sure that the folder exists
wcscat(buffer, path.c_str());
if (buffer[wcslen(buffer)-1] != L'\\') // path is a file
{
fullpath = buffer;
PathRemoveFileSpec(buffer);
}
PathAddBackslash(buffer);
}
else // cache mode
{
GetTempPath(MAX_PATH, buffer);
wcscat(buffer, L"Rainmeter-Cache\\"); // "%TEMP%\Rainmeter-Cache\"
}
2009-02-10 18:37:48 +00:00
CreateDirectory(buffer, NULL); // Make sure that the folder exists
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
directory = buffer;
if (fullpath.empty())
{
fullpath = directory;
std::wstring::size_type pos2 = url.find_first_of(L"?#");
std::wstring::size_type pos1 = url.find_last_of(L'/', pos2);
pos1 = (pos1 != std::wstring::npos) ? pos1 + 1 : 0;
std::wstring name;
if (pos2 != std::wstring::npos)
{
name = url.substr(pos1, pos2 - pos1);
}
else
{
name = url.substr(pos1);
}
if (!name.empty())
{
// Replace reserved characters to "_"
pos1 = 0;
while ((pos1 = name.find_first_of(L"\\/:*?\"<>|", pos1)) != std::wstring::npos)
{
name[pos1] = L'_';
}
fullpath += name;
}
else
{
fullpath += L"index";
}
}
ready = true;
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (download) // download mode
{
std::wstring log;
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (!PathFileExists(directory.c_str()) || !PathIsDirectory(directory.c_str()))
{
ready = false;
log = L"WebParser: [";
log += urlData->section;
log += L"] Directory not exists: ";
log += directory;
log += L"\n";
Log(log.c_str());
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
}
else if (PathIsDirectory(fullpath.c_str()))
{
ready = false;
log = L"WebParser: [";
log += urlData->section;
log += L"] Path is a directory, not a file: ";
log += fullpath;
log += L"\n";
Log(log.c_str());
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
}
else if (PathFileExists(fullpath.c_str()))
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
{
DWORD attr = GetFileAttributes(fullpath.c_str());
if (attr != INVALID_FILE_ATTRIBUTES && (attr & FILE_ATTRIBUTE_READONLY))
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
{
ready = false;
log = L"WebParser: [";
log += urlData->section;
log += L"] File is READ-ONLY: ";
log += fullpath;
log += L"\n";
Log(log.c_str());
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
}
}
}
else // cache mode
{
EnterCriticalSection(&g_CriticalSection);
if (PathFileExists(fullpath.c_str()))
{
std::wstring::size_type pos = fullpath.find_last_of(L'.');
std::wstring path, ext;
if (pos != std::wstring::npos)
{
path = fullpath.substr(0, pos);
ext = fullpath.substr(pos);
}
else
{
path = fullpath;
}
// Assign a serial number
int i = 1;
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
do
{
wsprintf(buffer, L"_%i", i++);
fullpath = path;
fullpath += buffer;
if (!ext.empty())
{
fullpath += ext;
}
} while (PathFileExists(fullpath.c_str()));
}
2009-02-10 18:37:48 +00:00
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
// Create empty file
HANDLE hFile = CreateFile(fullpath.c_str(), GENERIC_WRITE, 0, NULL, CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
if (hFile != INVALID_HANDLE_VALUE) CloseHandle(hFile);
LeaveCriticalSection(&g_CriticalSection);
}
if (ready)
2009-02-10 18:37:48 +00:00
{
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
// Delete IE cache before download if "SyncMode5" is not 3 (every visit to the page)
{
// Check "Temporary Internet Files" sync mode (SyncMode5)
// Values:
// Every visit to the page 3
// Every time you start Internet Explorer 2
// Automatically (default) 4
// Never 0
// http://support.microsoft.com/kb/263070/en
HKEY hKey;
LONG ret;
DWORD mode;
ret = RegOpenKeyEx(HKEY_CURRENT_USER, L"SOFTWARE\\Microsoft\\Windows\\CurrentVersion\\Internet Settings", 0, KEY_QUERY_VALUE, &hKey);
if (ret == ERROR_SUCCESS)
{
DWORD size = sizeof(mode);
ret = RegQueryValueEx(hKey, L"SyncMode5", NULL, NULL, (LPBYTE)&mode, &size);
RegCloseKey(hKey);
}
if (ret != ERROR_SUCCESS || mode != 3)
{
std::wstring::size_type pos = url.find_first_of(L'#');
if (pos != std::wstring::npos)
{
DeleteUrlCacheEntry(url.substr(0, pos).c_str());
}
else
{
DeleteUrlCacheEntry(url.c_str());
}
}
}
2009-02-10 18:37:48 +00:00
// Write some log info
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Downloading url ";
log += url;
log += L" to ";
log += fullpath;
log += L"\n";
Log(log.c_str());
2009-02-10 18:37:48 +00:00
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
HRESULT resultCoInitialize = CoInitialize(NULL); // requires before calling URLDownloadToFile function
2009-02-10 18:37:48 +00:00
// Download the file
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
HRESULT result = URLDownloadToFile(NULL, url.c_str(), fullpath.c_str(), NULL, NULL);
if (result == S_OK)
2009-02-10 18:37:48 +00:00
{
EnterCriticalSection(&g_CriticalSection);
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (!download) // cache mode
{
if (!urlData->downloadedFile.empty())
{
// Delete old downloaded file
DeleteFile(urlData->downloadedFile.c_str());
}
}
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
// Convert LFN to 8.3 filename if the path contains blank character
if (fullpath.find_first_of(L' ') != std::wstring::npos)
{
DWORD size = GetShortPathName(fullpath.c_str(), buffer, MAX_PATH);
if (size > 0 && size <= MAX_PATH)
{
fullpath = buffer;
}
}
urlData->downloadedFile = fullpath;
2009-02-10 18:37:48 +00:00
LeaveCriticalSection(&g_CriticalSection);
2011-03-29 19:21:57 +00:00
if (!urlData->finishAction.empty())
2009-02-10 18:37:48 +00:00
{
HWND wnd = FindMeterWindow(urlData->iniFile);
2009-02-10 18:37:48 +00:00
if (wnd != NULL)
{
COPYDATASTRUCT copyData;
copyData.dwData = 1;
copyData.cbData = (DWORD)(urlData->finishAction.size() + 1) * sizeof(WCHAR);
copyData.lpData = (void*)urlData->finishAction.c_str();
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
// Send the bang to the Rainmeter window
2009-02-10 18:37:48 +00:00
SendMessage(wnd, WM_COPYDATA, (WPARAM)NULL, (LPARAM)&copyData);
}
}
}
else
{
ready = false;
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (!download) // cache mode
{
// Delete empty file
DeleteFile(fullpath.c_str());
}
wsprintf(buffer, L"result=0x%08X, COM=0x%08X", result, resultCoInitialize);
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Download failed (";
log += buffer;
log += L"): ";
log += url;
Log(log.c_str());
2009-02-10 18:37:48 +00:00
}
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (SUCCEEDED(resultCoInitialize))
{
CoUninitialize();
}
2009-02-10 18:37:48 +00:00
}
else
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] Download failed: ";
log += url;
Log(log.c_str());
2009-02-10 18:37:48 +00:00
}
}
else
{
std::wstring log = L"WebParser: [";
log += urlData->section;
log += L"] The url is empty.\n";
Log(log.c_str());
}
if (!ready) // download failed
{
EnterCriticalSection(&g_CriticalSection);
if (!download) // cache mode
{
if (!urlData->downloadedFile.empty())
{
// Delete old downloaded file
DeleteFile(urlData->downloadedFile.c_str());
}
}
// Clear old downloaded filename
urlData->downloadedFile.clear();
LeaveCriticalSection(&g_CriticalSection);
2009-02-10 18:37:48 +00:00
}
EnterCriticalSection(&g_CriticalSection);
CloseHandle(urlData->dlThreadHandle);
urlData->dlThreadHandle = 0;
LeaveCriticalSection(&g_CriticalSection);
2011-03-29 19:21:57 +00:00
return 0; // thread completed successfully
2009-02-10 18:37:48 +00:00
}
/*
This function is called when the value should be
returned as a string.
*/
2011-03-29 19:21:57 +00:00
LPCTSTR GetString(UINT id, UINT flags)
2009-02-10 18:37:48 +00:00
{
static std::wstring resultString;
std::map<UINT, UrlData*>::iterator urlIter = g_UrlData.find(id);
2011-03-29 19:21:57 +00:00
if (urlIter != g_UrlData.end())
2009-02-10 18:37:48 +00:00
{
EnterCriticalSection(&g_CriticalSection);
if (((*urlIter).second)->download)
{
resultString = ((*urlIter).second)->downloadedFile;
}
else
{
resultString = ((*urlIter).second)->resultString;
}
LeaveCriticalSection(&g_CriticalSection);
return resultString.c_str();
}
return NULL;
}
/*
If the measure needs to free resources before quitting.
The plugin can export Finalize function, which is called
when Rainmeter quits (or refreshes).
*/
void Finalize(HMODULE instance, UINT id)
{
std::map<UINT, UrlData*>::iterator urlIter = g_UrlData.find(id);
2011-03-29 19:21:57 +00:00
if (urlIter != g_UrlData.end())
2009-02-10 18:37:48 +00:00
{
if (((*urlIter).second)->threadHandle)
{
// Thread is killed inside critical section so that itself is not inside one when it is terminated
EnterCriticalSection(&g_CriticalSection);
TerminateThread(((*urlIter).second)->threadHandle, 0);
(*urlIter).second->threadHandle = NULL;
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
LeaveCriticalSection(&g_CriticalSection);
}
if (((*urlIter).second)->dlThreadHandle)
{
// Thread is killed inside critical section so that itself is not inside one when it is terminated
EnterCriticalSection(&g_CriticalSection);
TerminateThread(((*urlIter).second)->dlThreadHandle, 0);
(*urlIter).second->dlThreadHandle = NULL;
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
LeaveCriticalSection(&g_CriticalSection);
}
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (((*urlIter).second)->downloadFile.empty()) // cache mode
2009-02-10 18:37:48 +00:00
{
[WebParser] Solved problems: * Multiple files of the same file name cannot be downloaded. (Issue 132) * The file including the character that cannot be used for the filename cannot be downloaded. (Issue 13) * As for the same file downloaded from the same URL, cache is not cleared and the file is not updated until the skin is restarted. (Issue 87) * Some kinds of files cannot be downloaded in Windows in which IE of an old version is installed. (because API used in WebParser depends on IE) -------------------- Added new "Download" option: The user can now download the file to the "DownloadFile" folder in a current skin folder as a file name defined in "DownloadFile=...". If you define it as "DownloadFile=image.jpg", a "DownloadFile" folder will be created in the current skin path then the downloaded file is saved as "image.jpg" Since this file is not a temporary file, it is not deleted even if it close a skin. And, when the refresh of skin or update of Webparser, the file will be overwrited by the latest file. The user can also make the subfolder in the "DownloadFile" folder. If you define as "DownloadFile=photo\image.jpg", the subfolder "photo" is created and a file is saved in it. As for specification of path, it is not possible to specify by the absolute path and also specification to the outside of a "DownloadFile" folder is not possible. It means that save of a file is limited to under the "DownloadFile" directory, and this does not permit save to other directories. This limitation is the measures that considered security. Valid specification examples: * DownloadFile=filename.jpg * DownloadFile=subfolder\filename.jpg these specification is invalid: * DownloadFile=C:\foo\filename.jpg * DownloadFile=#CURRENTPATH#filename.jpg * DownloadFile=..\filename.jpg
2010-01-12 21:45:47 +00:00
if (!((*urlIter).second)->downloadedFile.empty())
{
// Delete the file
DeleteFile(((*urlIter).second)->downloadedFile.c_str());
}
2009-02-10 18:37:48 +00:00
}
delete (*urlIter).second;
g_UrlData.erase(urlIter);
}
if (g_UrlData.empty())
{
// Last one, close all handles
if (hRootHandle)
{
InternetCloseHandle(hRootHandle);
hRootHandle = NULL;
}
g_CERs.clear();
2009-02-10 18:37:48 +00:00
// Last instance deletes the critical section
DeleteCriticalSection(&g_CriticalSection);
g_Initialized = false;
}
}
/*
Downloads the given url and returns the webpage as dynamically allocated string.
You need to delete the returned string after use!
*/
BYTE* DownloadUrl(std::wstring& url, DWORD* dwDataSize, bool forceReload)
2009-02-10 18:37:48 +00:00
{
HINTERNET hUrlDump;
DWORD dwSize;
BYTE* lpData;
BYTE* lpOutPut;
BYTE* lpHolding = NULL;
int nCounter = 1;
int nBufferSize;
const int CHUNK_SIZE = 8192;
std::wstring err = L"WebParser: Fetching URL: ";
err += url;
Log(err.c_str());
2011-03-29 19:21:57 +00:00
DWORD flags = INTERNET_FLAG_RESYNCHRONIZE;
if (forceReload)
{
flags = INTERNET_FLAG_RELOAD;
}
hUrlDump = InternetOpenUrl(hRootHandle, url.c_str(), NULL, NULL, flags, 0);
2009-02-10 18:37:48 +00:00
if (hUrlDump == NULL)
{
2010-09-17 08:47:22 +00:00
if (_wcsnicmp(url.c_str(), L"file://", 7) == 0) // file scheme
{
std::string urlACP = ConvertWideToAscii(url.c_str());
hUrlDump = InternetOpenUrlA(hRootHandle, urlACP.c_str(), NULL, NULL, flags, 0);
}
if (hUrlDump == NULL)
{
ShowError(__LINE__);
return NULL;
}
2009-02-10 18:37:48 +00:00
}
*dwDataSize = 0;
// Allocate the buffer.
lpData = new BYTE[CHUNK_SIZE];
2009-02-10 18:37:48 +00:00
do
{
// Read the data.
if (!InternetReadFile(hUrlDump, (LPVOID)lpData, CHUNK_SIZE, &dwSize))
{
ShowError(__LINE__);
break;
}
else
{
// Check if all of the data has been read. This should
// never get called on the first time through the loop.
if (dwSize == 0)
{
break;
}
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Determine the buffer size to hold the new data and the data
// already written (if any).
nBufferSize = *dwDataSize + dwSize;
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Allocate the output buffer.
lpOutPut = new BYTE[nBufferSize + 2];
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Make sure the buffer is not the initial buffer.
if (lpHolding != NULL)
{
// Copy the data in the holding buffer.
memcpy(lpOutPut, lpHolding, *dwDataSize);
// Delete the old buffer
delete [] lpHolding;
lpHolding = lpOutPut;
lpOutPut = lpOutPut + *dwDataSize;
}
else
{
lpHolding = lpOutPut;
}
// Copy the data buffer.
memcpy(lpOutPut, lpData, dwSize);
*dwDataSize += dwSize;
// End with double null
lpOutPut[dwSize] = 0;
lpOutPut[dwSize + 1] = 0;
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Increment the number of buffers read.
++nCounter;
2009-02-10 18:37:48 +00:00
// Clear the buffer
memset(lpData, 0, CHUNK_SIZE);
2009-02-10 18:37:48 +00:00
}
} while (TRUE);
2011-03-29 19:21:57 +00:00
2009-02-10 18:37:48 +00:00
// Close the HINTERNET handle.
InternetCloseHandle(hUrlDump);
err = L"WebParser: Finished URL: ";
err += url;
Log(err.c_str());
// Delete the existing buffers.
delete [] lpData;
2009-02-10 18:37:48 +00:00
// Return.
return lpHolding;
}
/*
Writes the last error to log.
*/
void ShowError(int lineNumber, WCHAR* errorMsg)
{
DWORD dwErr = GetLastError();
2009-02-10 18:37:48 +00:00
WCHAR buffer[16];
wsprintf(buffer, L"%i", lineNumber);
std::wstring err = L"WebParser (";
err += buffer;
err += L") ";
if (errorMsg == NULL)
{
2011-03-29 19:21:57 +00:00
if (dwErr == ERROR_INTERNET_EXTENDED_ERROR)
2009-02-10 18:37:48 +00:00
{
WCHAR szBuffer[1024];
DWORD dwError, dwLen = 1024;
2009-02-10 18:37:48 +00:00
if (InternetGetLastResponseInfo(&dwError, szBuffer, &dwLen))
{
err += szBuffer;
wsprintf(buffer, L"%i", dwError);
2009-02-10 18:37:48 +00:00
}
else
{
err += L"Unknown error";
wsprintf(buffer, L"%i", dwErr);
}
err += L" (ErrorCode=";
err += buffer;
err += L")";
2009-02-10 18:37:48 +00:00
}
else
{
LPVOID lpMsgBuf = NULL;
2009-02-10 18:37:48 +00:00
2011-03-29 19:21:57 +00:00
FormatMessage(
FORMAT_MESSAGE_ALLOCATE_BUFFER |
FORMAT_MESSAGE_FROM_HMODULE |
FORMAT_MESSAGE_FROM_SYSTEM |
FORMAT_MESSAGE_IGNORE_INSERTS |
FORMAT_MESSAGE_MAX_WIDTH_MASK,
GetModuleHandle(L"wininet"),
2009-02-10 18:37:48 +00:00
dwErr,
MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), // Default language
(LPTSTR) &lpMsgBuf,
0,
2011-03-29 19:21:57 +00:00
NULL
2009-02-10 18:37:48 +00:00
);
2011-03-29 19:21:57 +00:00
if (lpMsgBuf == NULL)
2009-02-10 18:37:48 +00:00
{
err += L"Unknown error";
2009-02-10 18:37:48 +00:00
}
else
{
err += (LPTSTR)lpMsgBuf;
LocalFree(lpMsgBuf);
}
wsprintf(buffer, L"%i", dwErr);
err += L" (ErrorCode=";
err += buffer;
err += L")";
2009-02-10 18:37:48 +00:00
}
}
else
{
err += errorMsg;
}
Log(err.c_str());
}
/*
Writes the log to a file (logging is thread safe (I think...)).
*/
void Log(const WCHAR* string)
{
// Todo: put logging into critical section
LSLog(LOG_DEBUG, L"Rainmeter", string);
}
UINT GetPluginVersion()
{
return 1013;
2009-02-10 18:37:48 +00:00
}
LPCTSTR GetPluginAuthor()
{
return L"Rainy (rainy@iki.fi)";
}