918 lines
22 KiB
C++
918 lines
22 KiB
C++
// NeL - MMORPG Framework <http://dev.ryzom.com/projects/nel/>
|
|
// Copyright (C) 2010 Winch Gate Property Limited
|
|
//
|
|
// This program is free software: you can redistribute it and/or modify
|
|
// it under the terms of the GNU Affero General Public License as
|
|
// published by the Free Software Foundation, either version 3 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 Affero General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU Affero General Public License
|
|
// along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
#include "stdmisc.h"
|
|
|
|
#include "nel/misc/diff_tool.h"
|
|
#include "nel/misc/path.h"
|
|
|
|
using namespace NLMISC;
|
|
using namespace std;
|
|
|
|
#ifdef DEBUG_NEW
|
|
#define new DEBUG_NEW
|
|
#endif
|
|
|
|
namespace STRING_MANAGER
|
|
{
|
|
|
|
uint64 makePhraseHash(const TPhrase &phrase)
|
|
{
|
|
ucstring text;
|
|
text = phrase.Parameters;
|
|
for (uint i=0; i<phrase.Clauses.size(); ++i)
|
|
{
|
|
text += phrase.Clauses[i].Conditions;
|
|
text += phrase.Clauses[i].Identifier;
|
|
text += phrase.Clauses[i].Text;
|
|
}
|
|
|
|
return CI18N::makeHash(text);
|
|
}
|
|
|
|
|
|
|
|
|
|
bool parseHashFromComment(const ucstring &comments, uint64 &hashValue)
|
|
{
|
|
string str = comments.toString();
|
|
|
|
string::size_type pos = str.find("HASH_VALUE ");
|
|
if (pos == string::npos)
|
|
return false;
|
|
|
|
string hashStr = str.substr(pos + 11, 16);
|
|
|
|
hashValue = CI18N::stringToHash(hashStr);
|
|
return true;
|
|
}
|
|
|
|
|
|
uint32 countLine(const ucstring &text, const ucstring::const_iterator upTo)
|
|
{
|
|
uint32 ret = 1;
|
|
ucstring::const_iterator first(text.begin());
|
|
|
|
for (; first != upTo; ++first)
|
|
{
|
|
if (*first == '\n')
|
|
ret++;
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
bool loadStringFile(const std::string filename, vector<TStringInfo> &stringInfos, bool forceRehash, ucchar openMark, ucchar closeMark, bool specialCase)
|
|
{
|
|
/* uint8 *buffer = 0;
|
|
uint size;
|
|
|
|
try
|
|
{
|
|
CIFile fp(filename);
|
|
size = fp.getFileSize();
|
|
buffer = new uint8[size];
|
|
fp.serialBuffer(buffer, size);
|
|
}
|
|
catch(const Exception &e)
|
|
{
|
|
nlinfo("Can't open file [%s] (%s)\n", filename.c_str(), e.what());
|
|
return true;
|
|
}
|
|
*/
|
|
/* FILE *fp = fopen(filename.c_str(), "rb");
|
|
|
|
if (fp == NULL)
|
|
{
|
|
nlinfo("Can't open file [%s]\n", filename.c_str());
|
|
if (buffer != 0)
|
|
delete [] buffer;
|
|
return true;
|
|
}
|
|
|
|
// move to end of file
|
|
fseek(fp, 0, SEEK_END);
|
|
|
|
fpos_t pos;
|
|
fgetpos(fp, &pos);
|
|
|
|
uint8 *buffer = new uint8[uint(pos)];
|
|
|
|
rewind(fp);
|
|
uint size = fread(buffer, 1, uint(pos), fp);
|
|
fclose (fp);
|
|
*/
|
|
ucstring text;
|
|
|
|
CI18N::readTextFile(filename, text, false, false, true, CI18N::LINE_FMT_CRLF);
|
|
// CI18N::readTextBuffer(buffer, size, text);
|
|
// delete [] buffer;
|
|
|
|
// ok, parse the file now.
|
|
ucstring::const_iterator first(text.begin()), last(text.end());
|
|
std::string lastLabel("nothing");
|
|
|
|
while (first != last)
|
|
{
|
|
TStringInfo si;
|
|
CI18N::skipWhiteSpace(first, last, &si.Comments);
|
|
|
|
if (first == last)
|
|
{
|
|
// check if there is only swap command remaining in comment
|
|
if (si.Comments.find(ucstring("// DIFF SWAP ")) != ucstring::npos)
|
|
{
|
|
stringInfos.push_back(si);
|
|
}
|
|
break;
|
|
}
|
|
|
|
// try to read a #fileline preprocessor command
|
|
if (CI18N::matchToken("#fileline", first, last))
|
|
{
|
|
// for now, just skip
|
|
uint32 lineCounter =0; // we count line another way
|
|
CI18N::skipLine(first, last, lineCounter);
|
|
|
|
// begin parse of next line
|
|
continue;
|
|
}
|
|
|
|
if (!CI18N::parseLabel(first, last, si.Identifier))
|
|
{
|
|
uint32 line = countLine(text, first);
|
|
nlwarning("DT: Fatal : In '%s', line %u: Invalid label after '%s'\n",
|
|
filename.c_str(),
|
|
line,
|
|
lastLabel.c_str());
|
|
return false;
|
|
}
|
|
lastLabel = si.Identifier;
|
|
|
|
CI18N::skipWhiteSpace(first, last, &si.Comments);
|
|
|
|
if (!CI18N::parseMarkedString(openMark, closeMark, first, last, si.Text))
|
|
{
|
|
uint32 line = countLine(text, first);
|
|
nlwarning("DT: Fatal : In '%s', line %u: Invalid text value for label %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
lastLabel.c_str());
|
|
return false;
|
|
}
|
|
|
|
if (specialCase)
|
|
{
|
|
CI18N::skipWhiteSpace(first, last, &si.Comments);
|
|
|
|
if (!CI18N::parseMarkedString(openMark, closeMark, first, last, si.Text2))
|
|
{
|
|
uint32 line = countLine(text, first);
|
|
nlwarning("DT: Fatal: In '%s' line %u: Invalid text2 value label %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
lastLabel.c_str());
|
|
return false;
|
|
}
|
|
|
|
}
|
|
|
|
if (forceRehash || !parseHashFromComment(si.Comments, si.HashValue))
|
|
{
|
|
// compute the hash value from text.
|
|
si.HashValue = CI18N::makeHash(si.Text);
|
|
// nldebug("Generating hash for %s as %s", si.Identifier.c_str(), CI18N::hashToString(si.HashValue).c_str());
|
|
}
|
|
else
|
|
{
|
|
// nldebug("Comment = [%s]", si.Comments.toString().c_str());
|
|
// nldebug("Retrieving hash for %s as %s", si.Identifier.c_str(), CI18N::hashToString(si.HashValue).c_str());
|
|
}
|
|
stringInfos.push_back(si);
|
|
}
|
|
|
|
|
|
// check identifier uniqueness
|
|
{
|
|
bool error = false;
|
|
set<string> unik;
|
|
set<string>::iterator it;
|
|
for (uint i=0; i<stringInfos.size(); ++i)
|
|
{
|
|
it = unik.find(stringInfos[i].Identifier);
|
|
if (it != unik.end())
|
|
{
|
|
nlwarning("DT: loadStringFile : identifier '%s' exist twice", stringInfos[i].Identifier.c_str() );
|
|
error = true;
|
|
}
|
|
else
|
|
unik.insert(stringInfos[i].Identifier);
|
|
|
|
}
|
|
if (error)
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
|
|
ucstring prepareStringFile(const vector<TStringInfo> &strings, bool removeDiffComments, bool noDiffInfo)
|
|
{
|
|
ucstring diff;
|
|
|
|
vector<TStringInfo>::const_iterator first(strings.begin()), last(strings.end());
|
|
for (; first != last; ++first)
|
|
{
|
|
ucstring str;
|
|
const TStringInfo &si = *first;
|
|
string comment = si.Comments.toString();
|
|
vector<string> lines;
|
|
explode(comment, string("\n"), lines, true);
|
|
|
|
uint i;
|
|
for (i=0; i<lines.size(); ++i)
|
|
{
|
|
if (removeDiffComments)
|
|
{
|
|
if (lines[i].find("// DIFF ") != string::npos)
|
|
{
|
|
lines.erase(lines.begin()+i);
|
|
--i;
|
|
continue;
|
|
}
|
|
}
|
|
if (lines[i].find("// INDEX ") != string::npos)
|
|
{
|
|
lines.erase(lines.begin()+i);
|
|
--i;
|
|
}
|
|
else if (lines[i].find("// HASH_VALUE ") != string::npos)
|
|
{
|
|
lines.erase(lines.begin()+i);
|
|
--i;
|
|
}
|
|
}
|
|
|
|
comment.erase();
|
|
for (i=0; i<lines.size(); ++i)
|
|
{
|
|
comment += lines[i] + "\n";
|
|
}
|
|
si.Comments = ucstring(comment);
|
|
|
|
str = si.Comments;
|
|
if (!si.Identifier.empty() || !si.Text.empty())
|
|
{
|
|
// add hash value comment if needed
|
|
// if (si.Comments.find(ucstring("// HASH_VALUE ")) == ucstring::npos)
|
|
if (!noDiffInfo)
|
|
{
|
|
str += ucstring("// HASH_VALUE ") + CI18N::hashToString(si.HashValue)+ nl;
|
|
str += ucstring("// INDEX ") + NLMISC::toString("%u", first-strings.begin())+ nl;
|
|
}
|
|
str += si.Identifier + '\t';
|
|
|
|
ucstring text = CI18N::makeMarkedString('[', ']', si.Text);;
|
|
ucstring text2;
|
|
// add new line and tab after each \n tag
|
|
ucstring::size_type pos;
|
|
const ucstring nlTag("\\n");
|
|
while ((pos = text.find(nlTag)) != ucstring::npos)
|
|
{
|
|
text2 += text.substr(0, pos+2) + nl + "\t";
|
|
text = text.substr(pos+2);
|
|
}
|
|
text2 += text;//.substr(0, pos+2);
|
|
str += text2 + nl + nl;
|
|
// str += CI18N::makeMarkedString('[', ']', si.Text) + nl + nl;
|
|
}
|
|
|
|
// nldebug("Adding string [%s]", str.toString().c_str());
|
|
diff += str;
|
|
}
|
|
|
|
return diff;
|
|
}
|
|
|
|
|
|
bool readPhraseFile(const std::string &filename, vector<TPhrase> &phrases, bool forceRehash)
|
|
{
|
|
ucstring doc;
|
|
|
|
CI18N::readTextFile(filename, doc, false, false, true, CI18N::LINE_FMT_CRLF);
|
|
|
|
return readPhraseFileFromString(doc, filename, phrases, forceRehash);
|
|
}
|
|
|
|
bool readPhraseFileFromString(ucstring const& doc, const std::string &filename, vector<TPhrase> &phrases, bool forceRehash)
|
|
{
|
|
std::string lastRead("nothing");
|
|
|
|
ucstring::const_iterator first(doc.begin()), last(doc.end());
|
|
while (first != last)
|
|
{
|
|
TPhrase phrase;
|
|
// parse the phrase
|
|
CI18N::skipWhiteSpace(first, last, &phrase.Comments);
|
|
|
|
if (first == last)
|
|
{
|
|
if (!phrase.Comments.empty())
|
|
{
|
|
// push the resulting comment
|
|
phrases.push_back(phrase);
|
|
}
|
|
break;
|
|
}
|
|
|
|
// try to read a #fileline preprocessor command
|
|
if (CI18N::matchToken("#fileline", first, last))
|
|
{
|
|
// for now, just skip
|
|
uint32 lineCounter =0; // we count line another way
|
|
CI18N::skipLine(first, last, lineCounter);
|
|
|
|
// begin parse of next line
|
|
continue;
|
|
}
|
|
|
|
if (!CI18N::parseLabel(first, last, phrase.Identifier))
|
|
{
|
|
uint32 line = countLine(doc, first);
|
|
nlwarning("DT: In '%s' line %u: Error parsing phrase identifier after %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
lastRead.c_str());
|
|
return false;
|
|
}
|
|
// nldebug("DT: parsing phrase '%s'", phrase.Identifier.c_str());
|
|
lastRead = phrase.Identifier;
|
|
CI18N::skipWhiteSpace(first, last, &phrase.Comments);
|
|
if (!CI18N::parseMarkedString('(', ')', first, last, phrase.Parameters))
|
|
{
|
|
uint32 line = countLine(doc, first);
|
|
nlwarning("DT: in '%s', line %u: Error parsing parameter list for phrase %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
phrase.Identifier.c_str());
|
|
return false;
|
|
}
|
|
CI18N::skipWhiteSpace(first, last, &phrase.Comments);
|
|
if (first == last || *first != '{')
|
|
{
|
|
uint32 line = countLine(doc, first);
|
|
nlwarning("DT: In '%s', line %u: Error parsing block opening '{' in phase %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
phrase.Identifier.c_str());
|
|
return false;
|
|
}
|
|
++first;
|
|
|
|
ucstring temp;
|
|
|
|
while (first != last && *first != '}')
|
|
{
|
|
TClause clause;
|
|
// append the comment preread at previous pass
|
|
clause.Comments = temp;
|
|
temp.erase();
|
|
// parse the clauses
|
|
CI18N::skipWhiteSpace(first, last, &clause.Comments);
|
|
if (first == last)
|
|
{
|
|
nlwarning("DT: Found end of file in non closed block for phrase %s\n", phrase.Identifier.c_str());
|
|
return false;
|
|
}
|
|
|
|
if (*first == '}')
|
|
break;
|
|
|
|
// skip the conditional expression
|
|
ucstring cond;
|
|
while (first != last && *first == '(')
|
|
{
|
|
if (!CI18N::parseMarkedString('(', ')', first, last, cond))
|
|
{
|
|
uint32 line = countLine(doc, first);
|
|
nlwarning("DT: In '%s' line %u: Error parsing conditional expression in phrase %s, clause %u\n",
|
|
filename.c_str(),
|
|
line,
|
|
phrase.Identifier.c_str(),
|
|
phrase.Clauses.size()+1);
|
|
return false;
|
|
}
|
|
clause.Conditions += "(" + cond + ") ";
|
|
CI18N::skipWhiteSpace(first, last, &clause.Comments);
|
|
}
|
|
if (first == last)
|
|
{
|
|
nlwarning("DT: in '%s': Found end of file in non closed block for phrase %s\n",
|
|
filename.c_str(),
|
|
phrase.Identifier.c_str());
|
|
return false;
|
|
}
|
|
// read the idnetifier (if any)
|
|
CI18N::parseLabel(first, last, clause.Identifier);
|
|
CI18N::skipWhiteSpace(first, last, &temp);
|
|
// read the text
|
|
if (CI18N::parseMarkedString('[', ']', first, last, clause.Text))
|
|
{
|
|
// the last read comment is for this clause.
|
|
clause.Comments += temp;
|
|
temp.erase();
|
|
}
|
|
else
|
|
{
|
|
uint32 line = countLine(doc, first);
|
|
nlwarning("DT: in '%s' line %u: Error reading text for clause %u (%s) in phrase %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
phrase.Clauses.size()+1,
|
|
clause.Identifier.c_str(),
|
|
phrase.Identifier.c_str());
|
|
return false;
|
|
|
|
}
|
|
|
|
phrase.Clauses.push_back(clause);
|
|
}
|
|
CI18N::skipWhiteSpace(first, last);
|
|
if (first == last || *first != '}')
|
|
{
|
|
uint32 line = countLine(doc, first);
|
|
nlwarning("DT: in '%s' line %u: Missing block closing tag '}' in phrase %s\n",
|
|
filename.c_str(),
|
|
line,
|
|
phrase.Identifier.c_str());
|
|
return false;
|
|
}
|
|
++first;
|
|
|
|
// handle hash value.
|
|
if (forceRehash || !parseHashFromComment(phrase.Comments, phrase.HashValue))
|
|
{
|
|
// the hash is not in the comment, compute it.
|
|
phrase.HashValue = makePhraseHash(phrase);
|
|
if (forceRehash)
|
|
{
|
|
// the has is perhaps in the comment
|
|
ucstring::size_type pos = phrase.Comments.find(ucstring("// HASH_VALUE"));
|
|
if (pos != ucstring::npos)
|
|
{
|
|
phrase.Comments = phrase.Comments.substr(0, pos);
|
|
}
|
|
}
|
|
}
|
|
|
|
// nldebug("DT : storing phrase '%s'", phrase.Identifier.c_str());
|
|
phrases.push_back(phrase);
|
|
}
|
|
|
|
// check identifier uniqueness
|
|
{
|
|
bool error = false;
|
|
set<string> unik;
|
|
set<string>::iterator it;
|
|
for (uint i=0; i<phrases.size(); ++i)
|
|
{
|
|
it = unik.find(phrases[i].Identifier);
|
|
if (it != unik.end())
|
|
{
|
|
nlwarning("DT: readPhraseFile : identifier '%s' exist twice", phrases[i].Identifier.c_str() );
|
|
error = true;
|
|
}
|
|
else
|
|
unik.insert(phrases[i].Identifier);
|
|
}
|
|
if (error)
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
ucstring tabLines(uint nbTab, const ucstring &str)
|
|
{
|
|
ucstring ret;
|
|
ucstring tabs;
|
|
|
|
for (uint i =0; i<nbTab; ++i)
|
|
tabs.push_back('\t');
|
|
|
|
ret = tabs;
|
|
ucstring::const_iterator first(str.begin()), last(str.end());
|
|
for (; first != last; ++first)
|
|
{
|
|
ret += *first;
|
|
if (*first == '\n')
|
|
ret += tabs;
|
|
}
|
|
|
|
while (ret[ret.size()-1] == '\t')
|
|
ret = ret.substr(0, ret.size()-1);
|
|
|
|
return ret;
|
|
}
|
|
|
|
ucstring preparePhraseFile(const vector<TPhrase> &phrases, bool removeDiffComments)
|
|
{
|
|
ucstring ret;
|
|
vector<TPhrase>::const_iterator first(phrases.begin()), last(phrases.end());
|
|
for (; first != last; ++first)
|
|
{
|
|
const TPhrase &p = *first;
|
|
|
|
if (removeDiffComments)
|
|
{
|
|
string comment = p.Comments.toString();
|
|
vector<string> lines;
|
|
explode(comment, string("\n"), lines, true);
|
|
|
|
uint i;
|
|
for (i=0; i<lines.size(); ++i)
|
|
{
|
|
if (lines[i].find("// DIFF ") != string::npos)
|
|
{
|
|
lines.erase(lines.begin()+i);
|
|
--i;
|
|
}
|
|
}
|
|
|
|
comment.erase();
|
|
for (i=0; i<lines.size(); ++i)
|
|
{
|
|
comment += lines[i] + "\n";
|
|
}
|
|
p.Comments = ucstring(comment);
|
|
}
|
|
ret += p.Comments;
|
|
|
|
if (!p.Identifier.empty() || !p.Clauses.empty())
|
|
{
|
|
if (p.Comments.find(ucstring("// HASH_VALUE ")) == ucstring::npos)
|
|
{
|
|
// add the hash value.
|
|
ret += ucstring("// HASH_VALUE ")+CI18N::hashToString(p.HashValue) + nl;
|
|
}
|
|
ret += p.Identifier + " ("+p.Parameters + ")" + nl;
|
|
ret += '{';
|
|
ret += nl;
|
|
for (uint i=0; i<p.Clauses.size(); ++i)
|
|
{
|
|
const TClause &c = p.Clauses[i];
|
|
if (!c.Comments.empty())
|
|
{
|
|
ucstring comment = tabLines(1, c.Comments);
|
|
ret += comment; // + '\r'+'\n';
|
|
}
|
|
if (!c.Conditions.empty())
|
|
{
|
|
ucstring cond = tabLines(1, c.Conditions);
|
|
ret += cond + nl;
|
|
}
|
|
ret += '\t';
|
|
// ucstring text = CI18N::makeMarkedString('[', ']', c.Text);
|
|
|
|
ucstring text = CI18N::makeMarkedString('[', ']', c.Text);;
|
|
ucstring text2;
|
|
// add new line and tab after each \n tag
|
|
ucstring::size_type pos;
|
|
const ucstring nlTag("\\n");
|
|
while ((pos = text.find(nlTag)) != ucstring::npos)
|
|
{
|
|
text2 += text.substr(0, pos+2) + nl;
|
|
text = text.substr(pos+2);
|
|
}
|
|
text2 += text;//.substr(0, pos+2);
|
|
|
|
text.swap(text2);
|
|
|
|
text = tabLines(3, text);
|
|
// remove begin tabs
|
|
text = text.substr(3);
|
|
ret += '\t' + c.Identifier + '\t' + text + nl + nl;
|
|
}
|
|
ret += '}';
|
|
}
|
|
ret += nl + nl;
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
bool loadExcelSheet(const string filename, TWorksheet &worksheet, bool checkUnique)
|
|
{
|
|
// Yoyo: must test with CIFile because can be packed into a .bnp on client...
|
|
CIFile fp;
|
|
if(!fp.open(filename))
|
|
{
|
|
nldebug("DT: Can't open file [%s]\n", filename.c_str());
|
|
return true;
|
|
}
|
|
fp.close();
|
|
|
|
ucstring str;
|
|
CI18N::readTextFile(filename, str, false, false, false, CI18N::LINE_FMT_CRLF);
|
|
|
|
if (!readExcelSheet(str, worksheet, checkUnique))
|
|
return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
bool readExcelSheet(const ucstring &str, TWorksheet &worksheet, bool checkUnique)
|
|
{
|
|
if(str.empty())
|
|
return true;
|
|
|
|
// copy the str to a big ucchar array => Avoid allocation / free
|
|
vector<ucchar> strArray;
|
|
// append a '\0'
|
|
strArray.resize(str.size()+1);
|
|
strArray[strArray.size()-1]= 0;
|
|
memcpy(&strArray[0], &str[0], str.size()*sizeof(ucchar));
|
|
|
|
|
|
// **** Build array of lines. just point to strArray, and fill 0 where appropriated
|
|
vector<ucchar*> lines;
|
|
lines.reserve(500);
|
|
ucstring::size_type pos = 0;
|
|
ucstring::size_type lastPos = 0;
|
|
while ((pos = str.find(nl, lastPos)) != ucstring::npos)
|
|
{
|
|
if (pos>lastPos)
|
|
{
|
|
strArray[pos]= 0;
|
|
// nldebug("Found line : [%s]", ucstring(&strArray[lastPos]).toString().c_str());
|
|
lines.push_back(&strArray[lastPos]);
|
|
}
|
|
lastPos = pos + 2;
|
|
}
|
|
|
|
// Must add last line if no \r\n ending
|
|
if (lastPos < str.size())
|
|
{
|
|
pos= str.size();
|
|
strArray[pos]= 0;
|
|
// nldebug("Found line : [%s]", ucstring(&strArray[lastPos]).toString().c_str());
|
|
lines.push_back(&strArray[lastPos]);
|
|
}
|
|
|
|
// nldebug("Found %u lines", lines.size());
|
|
|
|
// **** Do 2 pass.1st count the cell number, then fill. => avoid reallocation
|
|
uint newColCount= 0;
|
|
uint i;
|
|
for (i=0; i<lines.size(); ++i)
|
|
{
|
|
uint numCells;
|
|
numCells= 0;
|
|
|
|
ucchar *first= lines[i];
|
|
for (; *first != 0; ++first)
|
|
{
|
|
if (*first == '\t')
|
|
{
|
|
numCells++;
|
|
}
|
|
else if (*first == '"' && first==lines[i])
|
|
{
|
|
// read a quoted field.
|
|
first++;
|
|
while (*first != 0 && *first != '"' && *(first+1) != 0 && *(first+1) != '"')
|
|
{
|
|
first++;
|
|
if (*first != 0 && *first == '"')
|
|
{
|
|
// skip this
|
|
first++;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
// last cell
|
|
numCells++;
|
|
|
|
// take max cell of all lines
|
|
if (newColCount != max(newColCount, numCells))
|
|
{
|
|
newColCount = max(newColCount, numCells);
|
|
nldebug("At line %u, numCol changed to %u",
|
|
i, newColCount);
|
|
}
|
|
}
|
|
|
|
|
|
// **** alloc / enlarge worksheet
|
|
// enlarge Worksheet column size, as needed
|
|
while (worksheet.ColCount < newColCount)
|
|
worksheet.insertColumn(worksheet.ColCount);
|
|
|
|
// enlarge Worksheet row size, as needed
|
|
uint startLine= worksheet.size();
|
|
worksheet.resize(startLine + (uint)lines.size());
|
|
|
|
|
|
// **** fill worksheet
|
|
ucstring cell;
|
|
for (i=0; i<lines.size(); ++i)
|
|
{
|
|
uint numCells;
|
|
numCells= 0;
|
|
cell.erase();
|
|
|
|
ucchar *first= lines[i];
|
|
for (; *first != 0; ++first)
|
|
{
|
|
if (*first == '\t')
|
|
{
|
|
// nldebug("Found cell [%s]", cell.toString().c_str());
|
|
worksheet.setData(startLine + i, numCells, cell);
|
|
numCells++;
|
|
cell.erase();
|
|
}
|
|
else if (*first == '"' && first==lines[i])
|
|
{
|
|
// read a quoted field.
|
|
first++;
|
|
while (*first != 0 && *first != '"' && *(first+1) != 0 && *(first+1) != '"')
|
|
{
|
|
cell += *first;
|
|
first++;
|
|
if (*first != 0 && *first == '"')
|
|
{
|
|
// skip this
|
|
first++;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
cell += *first;
|
|
}
|
|
}
|
|
// nldebug("Found cell [%s]", cell.toString().c_str());
|
|
/// append last cell
|
|
worksheet.setData(startLine + i, numCells, cell);
|
|
numCells++;
|
|
nlassertex(numCells<=newColCount, ("readExcelSheet: bad row format: at line %u, the row has %u cell, max is %u", i, numCells, newColCount));
|
|
// nldebug("Found %u cells in line %u", numCells, i);
|
|
}
|
|
|
|
|
|
// **** identifier uniqueness checking.
|
|
if (checkUnique)
|
|
{
|
|
if (worksheet.size() > 0)
|
|
{
|
|
// look for the first non '* tagged' or 'DIFF_CMD' column
|
|
uint nameCol = 0;
|
|
while (nameCol < worksheet.ColCount && (*worksheet.getData(0, nameCol).begin() == uint16('*') || worksheet.getData(0, nameCol) == ucstring("DIFF_CMD")))
|
|
++nameCol;
|
|
|
|
if (nameCol < worksheet.ColCount )
|
|
{
|
|
// ok we can check unikness
|
|
bool error = false;
|
|
set<ucstring> unik;
|
|
set<ucstring>::iterator it;
|
|
for (uint j=0; j<worksheet.size(); ++j)
|
|
{
|
|
it = unik.find(worksheet.getData(j, nameCol));
|
|
if (it != unik.end())
|
|
{
|
|
nlwarning("DT: readExcelSheet : identifier '%s' exist twice", worksheet.getData(j, nameCol).toString().c_str() );
|
|
error = true;
|
|
}
|
|
else
|
|
unik.insert(worksheet.getData(j, nameCol));
|
|
}
|
|
if (error)
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
void makeHashCode(TWorksheet &sheet, bool forceRehash)
|
|
{
|
|
if (!sheet.Data.empty())
|
|
{
|
|
TWorksheet::TRow::iterator it = find(sheet.Data[0].begin(), sheet.Data[0].end(), ucstring("*HASH_VALUE"));
|
|
if (forceRehash || it == sheet.Data[0].end())
|
|
{
|
|
// we need to generate HASH_VALUE column !
|
|
if (it == sheet.Data[0].end())
|
|
{
|
|
sheet.insertColumn(0);
|
|
sheet.Data[0][0] = ucstring("*HASH_VALUE");
|
|
}
|
|
|
|
// Check columns
|
|
vector<bool> columnOk;
|
|
columnOk.resize(sheet.ColCount, false);
|
|
for (uint k=1; k<sheet.ColCount; ++k)
|
|
{
|
|
if (sheet.Data[0][k].find(ucstring("*")) != 0 && sheet.Data[0][k].find(ucstring("DIFF ")) != 0)
|
|
{
|
|
columnOk[k]= true;
|
|
}
|
|
}
|
|
|
|
// make hash for each line
|
|
ucstring str;
|
|
for (uint j=1; j<sheet.Data.size(); ++j)
|
|
{
|
|
str.erase();
|
|
for (uint k=1; k<sheet.ColCount; ++k)
|
|
{
|
|
if (columnOk[k])
|
|
{
|
|
str += sheet.Data[j][k];
|
|
}
|
|
}
|
|
uint64 hash = CI18N::makeHash(str);
|
|
CI18N::hashToUCString(hash, sheet.Data[j][0]);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
uint index = (uint)(it - sheet.Data[0].begin());
|
|
for (uint j=1; j<sheet.Data.size(); ++j)
|
|
{
|
|
ucstring &field = sheet.Data[j][index];
|
|
|
|
if (!field.empty() && field[0] == '_')
|
|
field = field.substr(1);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
ucstring prepareExcelSheet(const TWorksheet &worksheet)
|
|
{
|
|
if(worksheet.Data.empty())
|
|
return ucstring();
|
|
|
|
// **** First pass: count approx the size
|
|
uint approxSize= 0;
|
|
for (uint i=0; i<worksheet.Data.size(); ++i)
|
|
{
|
|
for (uint j=0; j<worksheet.Data[i].size(); ++j)
|
|
{
|
|
approxSize+= (uint)worksheet.Data[i][j].size() + 1;
|
|
}
|
|
approxSize++;
|
|
}
|
|
|
|
// Hash value for each column?
|
|
vector<bool> hashValue;
|
|
hashValue.resize(worksheet.Data[0].size());
|
|
for (uint j=0; j<worksheet.Data[0].size(); ++j)
|
|
{
|
|
hashValue[j]= worksheet.Data[0][j] == ucstring("*HASH_VALUE");
|
|
}
|
|
|
|
// **** Second pass: fill
|
|
ucstring text;
|
|
text.reserve(approxSize*2);
|
|
for (uint i=0; i<worksheet.Data.size(); ++i)
|
|
{
|
|
for (uint j=0; j<worksheet.Data[i].size(); ++j)
|
|
{
|
|
if (i > 0 && hashValue[j] && (!worksheet.Data[i][j].empty() && worksheet.Data[i][j][0] != '_'))
|
|
text += "_";
|
|
text += worksheet.Data[i][j];
|
|
if (j != worksheet.Data[i].size()-1)
|
|
text += '\t';
|
|
}
|
|
text += nl;
|
|
}
|
|
|
|
return text;
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
} // namespace STRING_MANAGER
|
|
|