\author André Pönitz (2003)
*/
-#include <config.h>
+// {[(
+
+#include "tex2lyx.h"
+#include "context.h"
+
+#include "debug.h"
+#include "lyx_main.h"
+#include "lyxtextclass.h"
+#include "support/path_defines.h"
+#include "support/os.h"
-#include <algorithm>
#include <cctype>
#include <fstream>
#include <iostream>
-#include <sstream>
-#include <stack>
#include <string>
+#include <sstream>
#include <vector>
-using std::atoi;
using std::cout;
using std::cerr;
using std::endl;
-using std::fill;
using std::getline;
-using std::ios;
-using std::ifstream;
using std::istream;
+using std::ifstream;
using std::istringstream;
using std::ostream;
using std::ostringstream;
-using std::stack;
+using std::stringstream;
using std::string;
using std::vector;
+// Hacks to allow the thing to link in the lyxlayout stuff
+Debug::type const Debug::ANY = Debug::type(0);
+DebugStream lyxerr;
-namespace {
-
-char const OPEN = '<';
-char const CLOSE = '>';
-
-const char * known_languages[] = { "austrian", "babel", "bahasa",
-"basque", "breton", "bulgarian", "catalan", "croatian", "czech", "danish",
-"dutch", "english", "esperanto", "estonian", "finnish", "francais",
-"frenchb", "galician", "germanb", "greek", "hebcal", "hebfont", "hebrew",
-"hebrew_newcode", "hebrew_oldcode", "hebrew_p", "hyphen", "icelandic",
-"irish", "italian", "latin", "lgrcmr", "lgrcmro", "lgrcmss", "lgrcmtt",
-"lgrenc", "lgrlcmss", "lgrlcmtt", "lheclas", "lhecmr", "lhecmss",
-"lhecmtt", "lhecrml", "lheenc", "lhefr", "lheredis", "lheshold",
-"lheshscr", "lheshstk", "lsorbian", "magyar", "naustrian", "ngermanb",
-"ngerman", "norsk", "polish", "portuges", "rlbabel", "romanian",
-"russianb", "samin", "scottish", "serbian", "slovak", "slovene", "spanish",
-"swedish", "turkish", "ukraineb", "usorbian", "welsh", 0};
-
-const char * known_fontsizes[] = { "10pt", "11pt", "12pt", 0 };
-
-
-// some ugly stuff
-string h_preamble;
-string h_textclass = "FIXME";
-string h_options = "FIXME";
-string h_language = "FIXME";
-string h_inputencoding = "FIXME";
-string h_fontscheme = "FIXME";
-string h_graphics = "default";
-string h_paperfontsize = "FIXME";
-string h_spacing = "single";
-string h_papersize = "FIXME";
-string h_paperpackage = "FIXME";
-string h_use_geometry = "0";
-string h_use_amsmath = "0";
-string h_use_natbib = "0";
-string h_use_numerical_citations = "0";
-string h_paperorientation = "portrait";
-string h_secnumdepth = "3";
-string h_tocdepth = "3";
-string h_paragraph_separation = "indent";
-string h_defskip = "medskip";
-string h_quotes_language = "2";
-string h_quotes_times = "1";
-string h_papercolumns = "1";
-string h_papersides = "1";
-string h_paperpagestyle = "default";
-string h_tracking_changes = "0";
-
-// indicates whether we are in the preamble
-bool in_preamble = true;
-
-// current stack of nested environments
-stack<string> active_environments;
+void LyX::emergencyCleanup() {}
+void handle_comment(Parser & p)
+{
+ string s;
+ while (p.good()) {
+ Token const & t = p.get_token();
+ if (t.cat() == catNewline)
+ break;
+ s += t.asString();
+ }
+ //cerr << "comment: " << s << "\n";
+ p.skip_spaces();
+}
-string const trim(string const & a, char const * p = " ")
+string const trim(string const & a, char const * p)
{
// lyx::Assert(p);
void split(string const & s, vector<string> & result, char delim)
{
- istringstream is(s);
+ //cerr << "split 1: '" << s << "'\n";
+ istringstream is(s);
string t;
while (getline(is, t, delim))
result.push_back(t);
+ //cerr << "split 2\n";
}
-string join(vector<string> const & input, char delim)
+string join(vector<string> const & input, char const * delim)
{
ostringstream os;
- for (size_t i = 0; i != input.size(); ++i) {
+ for (size_t i = 0; i < input.size(); ++i) {
if (i)
- os << delim;
- os << input[i];
+ os << delim;
+ os << input[i];
}
return os.str();
}
-void handle_opt(vector<string> & opts, char const ** what, string & target)
-{
- if (opts.empty())
- return;
-
- for ( ; what; ++what) {
- vector<string>::iterator it = find(opts.begin(), opts.end(), *what);
- if (it != opts.end()) {
- //cerr << "### found option '" << *what << "'\n";
- target = *what;
- opts.erase(it);
- return;
- }
- }
-}
-
-
-void handle_ert(ostream & os, string const & s)
-{
- os << "\n\\begin_inset ERT\nstatus Collapsed\n\n\\layout Standard\n\n";
- os << s;
- os << "\n\\end_inset\n";
-}
-
-
-void handle_package(string const & name, string const & options)
-{
- if (name == "a4wide") {
- h_papersize = "a4";
- h_paperpackage = "widemarginsa4";
- } else if (name == "ae")
- h_fontscheme = "ae";
- else if (name == "aecompl")
- h_fontscheme = "ae";
- else if (name == "amsmath")
- h_use_amsmath = "1";
- else if (name == "amssymb")
- h_use_amsmath = "1";
- else if (name == "babel")
- ; // ignore this
- else if (name == "fontenc")
- ; // ignore this
- else if (name == "inputenc")
- h_inputencoding = options;
- else if (name == "makeidx")
- ; // ignore this
- else if (name == "verbatim")
- ; // ignore this
- else {
- if (options.size())
- h_preamble += "\\usepackage[" + options + "]{" + name + "}\n";
- else
- h_preamble += "\\usepackage{" + name + "}\n";
- }
-}
-
-
-string wrap(string const & cmd, string const & str)
-{
- return OPEN + cmd + ' ' + str + CLOSE;
-}
-
-
-string wrap(string const & cmd, string const & str, string const & str2)
-{
- return OPEN + cmd + ' ' + str + ' ' + str2 + CLOSE;
-}
-
-
-enum mode_type {UNDECIDED_MODE, TEXT_MODE, MATH_MODE};
-
-mode_type asMode(mode_type oldmode, string const & str)
-{
- if (str == "mathmode")
- return MATH_MODE;
- if (str == "textmode" || str == "forcetext")
- return TEXT_MODE;
- return oldmode;
-}
-
-
-// These are TeX's catcodes
-enum CatCode {
- catEscape, // 0 backslash
- catBegin, // 1 {
- catEnd, // 2 }
- cat, // 3 $
- catAlign, // 4 &
- catNewline, // 5 ^^M
- catParameter, // 6 #
- catSuper, // 7 ^
- catSub, // 8 _
- catIgnore, // 9
- catSpace, // 10 space
- catLetter, // 11 a-zA-Z
- catOther, // 12 none of the above
- catActive, // 13 ~
- catComment, // 14 %
- catInvalid // 15 <delete>
-};
-
-CatCode theCatcode[256];
-
-
-inline CatCode catcode(unsigned char c)
-{
- return theCatcode[c];
-}
-
-
-enum {
- FLAG_BRACE_LAST = 1 << 1, // last closing brace ends the parsing
- FLAG_RIGHT = 1 << 2, // next \\right ends the parsing process
- FLAG_END = 1 << 3, // next \\end ends the parsing process
- FLAG_BRACK_LAST = 1 << 4, // next closing bracket ends the parsing
- FLAG_TEXTMODE = 1 << 5, // we are in a box
- FLAG_ITEM = 1 << 6, // read a (possibly braced token)
- FLAG_LEAVE = 1 << 7, // leave the loop at the end
- FLAG_SIMPLE = 1 << 8, // next $ leaves the loop
- FLAG_EQUATION = 1 << 9, // next \] leaves the loop
- FLAG_SIMPLE2 = 1 << 10, // next \) leaves the loop
- FLAG_OPTION = 1 << 11, // read [...] style option
- FLAG_BRACED = 1 << 12 // read {...} style argument
-};
-
-
-void catInit()
-{
- fill(theCatcode, theCatcode + 256, catOther);
- fill(theCatcode + 'a', theCatcode + 'z' + 1, catLetter);
- fill(theCatcode + 'A', theCatcode + 'Z' + 1, catLetter);
-
- theCatcode['\\'] = catEscape;
- theCatcode['{'] = catBegin;
- theCatcode['}'] = catEnd;
- theCatcode['$'] = cat;
- theCatcode['&'] = catAlign;
- theCatcode['\n'] = catNewline;
- theCatcode['#'] = catParameter;
- theCatcode['^'] = catSuper;
- theCatcode['_'] = catSub;
- theCatcode['\7f'] = catIgnore;
- theCatcode[' '] = catSpace;
- theCatcode['\t'] = catSpace;
- theCatcode['\r'] = catNewline;
- theCatcode['~'] = catActive;
- theCatcode['%'] = catComment;
-}
-
-
-
-//
-// Helper class for parsing
-//
-
-class Token {
-public:
- ///
- Token() : cs_(), char_(0), cat_(catIgnore) {}
- ///
- Token(char c, CatCode cat) : cs_(), char_(c), cat_(cat) {}
- ///
- Token(string const & cs) : cs_(cs), char_(0), cat_(catIgnore) {}
-
- ///
- string const & cs() const { return cs_; }
- ///
- CatCode cat() const { return cat_; }
- ///
- char character() const { return char_; }
- ///
- string asString() const { return cs_.size() ? cs_ : string(1, char_); }
-
-private:
- ///
- string cs_;
- ///
- char char_;
- ///
- CatCode cat_;
-};
-
-ostream & operator<<(ostream & os, Token const & t)
-{
- if (t.cs().size())
- os << '\\' << t.cs();
- else
- os << '[' << t.character() << ',' << t.cat() << ']';
- return os;
-}
-
-
-class Parser {
-
-public:
- ///
- Parser(istream & is);
-
- ///
- string parse();
- ///
- string parse(unsigned flags, mode_type mode);
- ///
- int lineno() const { return lineno_; }
- ///
- void putback();
- /// dump contents to screen
- void dump() const;
-
-private:
- ///
- string getArg(char left, char right);
- ///
- char getChar();
- ///
- void error(string const & msg);
- ///
- void tokenize(istream & is);
- ///
- void tokenize(string const & s);
- ///
- void skipSpaceTokens(istream & is, char c);
- ///
- void push_back(Token const & t);
- ///
- void pop_back();
- ///
- Token const & prevToken() const;
- ///
- Token const & nextToken() const;
- ///
- Token const & getToken();
- /// skips spaces if any
- void skipSpaces();
- ///
- void lex(string const & s);
- ///
- bool good() const;
- ///
- string parse_verbatim_item();
- ///
- string parse_verbatim_option();
-
- ///
- int lineno_;
- ///
- vector<Token> tokens_;
- ///
- unsigned pos_;
-};
-
-
-Parser::Parser(istream & is)
- : lineno_(0), pos_(0)
-{
- tokenize(is);
-}
-
-
-void Parser::push_back(Token const & t)
-{
- tokens_.push_back(t);
-}
-
-
-void Parser::pop_back()
-{
- tokens_.pop_back();
-}
-
-
-Token const & Parser::prevToken() const
-{
- static const Token dummy;
- return pos_ > 0 ? tokens_[pos_ - 1] : dummy;
-}
-
-
-Token const & Parser::nextToken() const
-{
- static const Token dummy;
- return good() ? tokens_[pos_] : dummy;
-}
-
-
-Token const & Parser::getToken()
-{
- static const Token dummy;
- //cerr << "looking at token " << tokens_[pos_] << " pos: " << pos_ << '\n';
- return good() ? tokens_[pos_++] : dummy;
-}
-
-
-void Parser::skipSpaces()
-{
- while (nextToken().cat() == catSpace || nextToken().cat() == catNewline)
- getToken();
-}
-
-
-void Parser::putback()
-{
- --pos_;
-}
-
-
-bool Parser::good() const
-{
- return pos_ < tokens_.size();
-}
-
-
-char Parser::getChar()
-{
- if (!good())
- error("The input stream is not well...");
- return tokens_[pos_++].character();
-}
-
-
-string Parser::getArg(char left, char right)
-{
- skipSpaces();
-
- string result;
- char c = getChar();
-
- if (c != left)
- putback();
- else
- while ((c = getChar()) != right && good())
- result += c;
-
- return result;
-}
-
-
-void Parser::skipSpaceTokens(istream & is, char c)
-{
- // skip trailing spaces
- while (catcode(c) == catSpace || catcode(c) == catNewline)
- if (!is.get(c))
- break;
- //cerr << "putting back: " << c << "\n";
- is.putback(c);
-}
-
-
-void Parser::tokenize(istream & is)
-{
- // eat everything up to the next \end_inset or end of stream
- // and store it in s for further tokenization
- string s;
- char c;
- while (is.get(c)) {
- s += c;
- if (s.size() >= 10 && s.substr(s.size() - 10) == "\\end_inset") {
- s = s.substr(0, s.size() - 10);
- break;
- }
- }
- // Remove the space after \end_inset
- if (is.get(c) && c != ' ')
- is.unget();
-
- // tokenize buffer
- tokenize(s);
-}
-
-
-void Parser::tokenize(string const & buffer)
-{
- static bool init_done = false;
-
- if (!init_done) {
- catInit();
- init_done = true;
- }
-
- istringstream is(buffer.c_str(), ios::in | ios::binary);
-
- char c;
- while (is.get(c)) {
- //cerr << "reading c: " << c << "\n";
-
- switch (catcode(c)) {
- case catNewline: {
- ++lineno_;
- is.get(c);
- if (catcode(c) == catNewline)
- push_back(Token("par"));
- else {
- push_back(Token('\n', catNewline));
- is.putback(c);
- }
- break;
- }
-
-/*
- case catComment: {
- while (is.get(c) && catcode(c) != catNewline)
- ;
- ++lineno_;
- break;
- }
-*/
-
- case catEscape: {
- is.get(c);
- if (!is) {
- error("unexpected end of input");
- } else {
- string s(1, c);
- if (catcode(c) == catLetter) {
- // collect letters
- while (is.get(c) && catcode(c) == catLetter)
- s += c;
- skipSpaceTokens(is, c);
- }
- push_back(Token(s));
- }
- break;
- }
-
- case catSuper:
- case catSub: {
- push_back(Token(c, catcode(c)));
- is.get(c);
- skipSpaceTokens(is, c);
- break;
- }
-
- case catIgnore: {
- cerr << "ignoring a char: " << int(c) << "\n";
- break;
- }
-
- default:
- push_back(Token(c, catcode(c)));
- }
- }
-
-#ifdef FILEDEBUG
- dump();
-#endif
-}
-
-
-void Parser::dump() const
-{
- cerr << "\nTokens: ";
- for (unsigned i = 0; i < tokens_.size(); ++i) {
- if (i == pos_)
- cerr << " <#> ";
- cerr << tokens_[i];
- }
- cerr << " pos: " << pos_ << "\n";
-}
-
-
-void Parser::error(string const & msg)
-{
- cerr << "Line ~" << lineno_ << ": parse error: " << msg << endl;
- dump();
- //exit(1);
-}
-
-
-string Parser::parse()
+char const ** is_known(string const & str, char const ** what)
{
- skipSpaces();
- return parse(0, UNDECIDED_MODE);
+ for ( ; *what; ++what)
+ if (str == *what)
+ return what;
+ return 0;
}
-string Parser::parse_verbatim_option()
-{
- string res;
- if (nextToken().character() == '[') {
- Token t = getToken();
- for (Token t = getToken(); t.character() != ']' && good(); t = getToken()) {
- if (t.cat() == catBegin) {
- putback();
- res += '{' + parse_verbatim_item() + '}';
- } else
- res += t.asString();
- }
- }
- return res;
-}
-
-string Parser::parse_verbatim_item()
-{
- string res;
- if (nextToken().cat() == catBegin) {
- Token t = getToken();
- for (Token t = getToken(); t.cat() != catEnd && good(); t = getToken()) {
- if (t.cat() == catBegin) {
- putback();
- res += '{' + parse_verbatim_item() + '}';
- }
- else
- res += t.asString();
- }
- }
- return res;
-}
+// current stack of nested environments
+vector<string> active_environments;
-string Parser::parse(unsigned flags, mode_type mode)
+string active_environment()
{
- //int limits = 0;
-
- ostringstream result;
- while (good()) {
- Token const & t = getToken();
-
-#ifdef FILEDEBUG
- cerr << "t: " << t << " flags: " << flags << "\n";
- cell->dump();
- cerr << "\n";
-#endif
-
- if (flags & FLAG_ITEM) {
- if (t.cat() == catSpace)
- continue;
-
- flags &= ~FLAG_ITEM;
- if (t.cat() == catBegin) {
- // skip the brace and collect everything to the next matching
- // closing brace
- flags |= FLAG_BRACE_LAST;
- continue;
- }
-
- // handle only this single token, leave the loop if done
- flags |= FLAG_LEAVE;
- }
-
-
- if (flags & FLAG_BRACED) {
- if (t.cat() == catSpace)
- continue;
-
- if (t.cat() != catBegin) {
- error("opening brace expected");
- return result.str();
- }
-
- // skip the brace and collect everything to the next matching
- // closing brace
- flags = FLAG_BRACE_LAST;
- }
-
-
- if (flags & FLAG_OPTION) {
- if (t.cat() == catOther && t.character() == '[') {
- result << parse(FLAG_BRACK_LAST, mode);
- } else {
- // no option found, put back token and we are done
- putback();
- }
- return result.str();
- }
-
- //
- // cat codes
- //
- if (t.cat() == cat) {
- if (mode != MATH_MODE) {
- // we are inside some text mode thingy, so opening new math is allowed
- Token const & n = getToken();
- if (n.cat() == cat) {
- // TeX's $$...$$ syntax for displayed math
- result << wrap("equation", parse(FLAG_SIMPLE, MATH_MODE));
- getToken(); // skip the second '$' token
- } else {
- // simple $...$ stuff
- putback();
- result << wrap("simple", parse(FLAG_SIMPLE, MATH_MODE));
- }
- }
-
- else if (flags & FLAG_SIMPLE) {
- // this is the end of the formula
- return result.str();
- }
-
- else {
- error("something strange in the parser\n");
- break;
- }
- }
-
- else if (t.cat() == catLetter)
- result << t.character();
-
- else if (t.cat() == catSpace && mode != MATH_MODE) {
- //if (result.empty() || result[result.size() - 1] != ' ')
- result << t.character();
- }
-
- else if (t.cat() == catNewline && mode != MATH_MODE)
- result << t.character();
-
- else if (t.cat() == catParameter) {
- Token const & n = getToken();
- result << wrap("macroarg", string(1, n.character()));
- }
-
- else if (t.cat() == catActive)
- result << wrap("active", string(1, t.character()));
-
- else if (t.cat() == catBegin)
- result << wrap("braced", parse(FLAG_BRACE_LAST, mode));
-
- else if (t.cat() == catEnd) {
- if (flags & FLAG_BRACE_LAST)
- return result.str();
- error("found '}' unexpectedly");
- //lyx::Assert(0);
- //add(cell, '}', LM_TC_TEX);
- }
-
-/*
- else if (t.cat() == catAlign) {
- ++cellcol;
- //cerr << " column now " << cellcol << " max: " << grid.ncols() << "\n";
- if (cellcol == grid.ncols()) {
- //cerr << "adding column " << cellcol << "\n";
- grid.addCol(cellcol - 1);
- }
- cell = &grid.cell(grid.index(cellrow, cellcol));
- }
-*/
-
- else if (t.character() == ']' && (flags & FLAG_BRACK_LAST)) {
- //cerr << "finished reading option\n";
- return result.str();
- }
-
- else if (t.cat() == catOther)
- result << string(1, t.character());
-
- else if (t.cat() == catComment) {
- string s;
- while (good()) {
- Token const & t = getToken();
- if (t.cat() == catNewline)
- break;
- s += t.asString();
- }
- //result << wrap("comment", s);
- skipSpaces();
- }
-
- //
- // control sequences
- //
-
- else if (t.cs() == "lyxlock") {
- // ignored;
- }
-
- else if (t.cs() == "newcommand" || t.cs() == "providecommand") {
- string const name = parse_verbatim_item();
- string const opts = getArg('[', ']');
- string const body = parse_verbatim_item();
- // only non-lyxspecific stuff
- if (name != "noun" && name != "tabularnewline") {
- h_preamble += "\\" + t.cs() + "{" + name + "}";
- if (opts.size())
- h_preamble += "[" + opts + "]";
- h_preamble += "{" + body + "}\n";
- }
- }
-
- else if (t.cs() == "(")
- result << wrap("simple", parse(FLAG_SIMPLE2, MATH_MODE));
-
- else if (t.cs() == "[")
- result << wrap("equation", parse(FLAG_EQUATION, MATH_MODE));
-
- else if (t.cs() == "protect")
- // ignore \\protect, will hopefully be re-added during output
- ;
-
- else if (t.cs() == "end") {
- if (flags & FLAG_END) {
- // eat environment name
- string const name = getArg('{', '}');
- if (name != active_environments.top())
- error("\\end{" + name + "} does not match \\begin{"
- + active_environments.top() + "}");
- active_environments.pop();
- return result.str();
- }
- error("found 'end' unexpectedly");
- }
-
- else if (t.cs() == ")") {
- if (flags & FLAG_SIMPLE2)
- return result.str();
- error("found '\\)' unexpectedly");
- }
-
- else if (t.cs() == "]") {
- if (flags & FLAG_EQUATION)
- return result.str();
- error("found '\\]' unexpectedly");
- }
-
-/*
- else if (t.cs() == "\\") {
- grid.vcrskip(LyXLength(getArg('[', ']')), cellrow);
- ++cellrow;
- cellcol = 0;
- if (cellrow == grid.nrows())
- grid.addRow(cellrow - 1);
- if (grid.asHullstring())
- grid.asHullstring()->numbered(cellrow, numbered);
- cell = &grid.cell(grid.index(cellrow, cellcol));
- }
-*/
- else if (t.cs() == "documentclass") {
- vector<string> opts;
- split(getArg('[', ']'), opts, ',');
- handle_opt(opts, known_languages, h_language);
- handle_opt(opts, known_fontsizes, h_paperfontsize);
- h_options = join(opts, ',');
- h_textclass = getArg('{', '}');
- }
-
- else if (t.cs() == "usepackage") {
- string const options = getArg('[', ']');
- string const name = getArg('{', '}');
- if (options.empty() && name.find(',')) {
- vector<string> vecnames;
- split(name, vecnames, ',');
- vector<string>::const_iterator it = vecnames.begin();
- vector<string>::const_iterator end = vecnames.end();
- for (; it != end; ++it) {
- handle_package(trim(*it), string());
- }
- } else {
- handle_package(name, options);
- }
- }
-
- else if (t.cs() == "newenvironment") {
- string const name = getArg('{', '}');
- skipSpaces();
- string const begin = parse_verbatim_item();
- skipSpaces();
- string const end = parse_verbatim_item();
- // ignore out mess
- if (name != "lyxcode")
- result << wrap("newenvironment", begin + end);
- }
-
- else if (t.cs() == "def") {
- string const name = getToken().cs();
- string res;
- while (nextToken().cat() != catBegin)
- res += getToken().asString();
- handle_ert(result, "\\def" + res + '{' + parse_verbatim_item() + '}');
- }
-
- else if (t.cs() == "setcounter") {
- string const name = getArg('{', '}');
- string const content = getArg('{', '}');
- if (name == "secnumdepth")
- h_secnumdepth = content;
- else if (name == "tocdepth")
- h_tocdepth = content;
- else
- h_preamble += "\\setcounter{" + name + "}{" + content + "}\n";
- }
-
- else if (t.cs() == "setlength") {
- string const name = getToken().cs();
- string const content = getArg('{', '}');
- if (name == "parskip")
- h_paragraph_separation = "skip";
- else if (name == "parindent")
- h_paragraph_separation = "skip";
- else
- h_preamble += "\\setcounter{" + name + "}{" + content + "}\n";
- }
-
- else if (t.cs() == "par") {
- if (!active_environments.empty())
- result << "\n\\layout " << active_environments.top() << "\n\n";
- }
-
- else if (t.cs() == "title")
- result << "\\layout Title\n\n" + parse_verbatim_item();
-
- else if (t.cs() == "author")
- result << "\\layout Author\n\n" + parse_verbatim_item();
-
- else if (t.cs() == "abstract")
- result << "\\layout Abstract\n\n" + parse_verbatim_item();
-
- else if (t.cs() == "begin") {
- string const name = getArg('{', '}');
- active_environments.push(name);
- result << parse(FLAG_END, mode);
- }
-
- if (flags & FLAG_LEAVE) {
- flags &= ~FLAG_LEAVE;
- break;
- }
- }
-
- return result.str();
+ return active_environments.empty() ? string() : active_environments.back();
}
-} // anonymous namespace
-
-
int main(int argc, char * argv[])
{
if (argc <= 1) {
return 2;
}
- string t;
+ lyx::support::os::init(&argc, &argv);
+ lyx::support::setLyxPaths();
+
ifstream is(argv[1]);
Parser p(is);
//p.dump();
- string s = p.parse();
- cout << "# tex2lyx 0.0.2 created this file\n"
- << "\\lyxformat 222\n"
- << "\\textclass " << h_textclass << "\n"
- << "\\begin_preamble\n" << h_preamble << "\\end_preamble\n"
- << "\\options " << h_options << "\n"
- << "\\language " << h_language << "\n"
- << "\\inputencoding " << h_inputencoding << "\n"
- << "\\fontscheme " << h_fontscheme << "\n"
- << "\\graphics " << h_graphics << "\n"
- << "\\paperfontsize " << h_paperfontsize << "\n"
- << "\\spacing " << h_spacing << "\n"
- << "\\papersize " << h_papersize << "\n"
- << "\\paperpackage " << h_paperpackage << "\n"
- << "\\use_geometry " << h_use_geometry << "\n"
- << "\\use_amsmath " << h_use_amsmath << "\n"
- << "\\use_natbib " << h_use_natbib << "\n"
- << "\\use_numerical_citations " << h_use_numerical_citations << "\n"
- << "\\paperorientation " << h_paperorientation << "\n"
- << "\\secnumdepth " << h_secnumdepth << "\n"
- << "\\tocdepth " << h_tocdepth << "\n"
- << "\\paragraph_separation " << h_paragraph_separation << "\n"
- << "\\defskip " << h_defskip << "\n"
- << "\\quotes_language " << h_quotes_language << "\n"
- << "\\quotes_times " << h_quotes_times << "\n"
- << "\\papercolumns " << h_papercolumns << "\n"
- << "\\papersides " << h_papersides << "\n"
- << "\\paperpagestyle " << h_paperpagestyle << "\n"
- << "\\tracking_changes " << h_tracking_changes << "\n"
- << s << "\n"
- << "\\the_end";
-
- return 0;
-}
+
+ stringstream ss;
+ LyXTextClass textclass = parse_preamble(p, ss);
+ active_environments.push_back("document");
+ Context context(true, textclass);
+ parse_text(p, ss, FLAG_END, true, context);
+ context.check_end_layout(ss);
+ ss << "\n\\end_document\n";
+
+ ss.seekg(0);
+ cout << ss.str();
+ return 0;
+}
+
+// }])