mirror of
https://git.lyx.org/repos/lyx.git
synced 2024-11-14 06:57:01 +00:00
1156 lines
31 KiB
C++
1156 lines
31 KiB
C++
/**
|
|
* \file BiblioInfo.cpp
|
|
* This file is part of LyX, the document processor.
|
|
* Licence details can be found in the file COPYING.
|
|
*
|
|
* \author Angus Leeming
|
|
* \author Herbert Voß
|
|
* \author Richard Heck
|
|
* \author Julien Rioux
|
|
*
|
|
* Full author contact details are available in file CREDITS.
|
|
*/
|
|
|
|
#include <config.h>
|
|
|
|
#include "BiblioInfo.h"
|
|
#include "Buffer.h"
|
|
#include "BufferParams.h"
|
|
#include "buffer_funcs.h"
|
|
#include "Encoding.h"
|
|
#include "InsetIterator.h"
|
|
#include "Language.h"
|
|
#include "output_xhtml.h"
|
|
#include "Paragraph.h"
|
|
#include "TextClass.h"
|
|
#include "TocBackend.h"
|
|
|
|
#include "support/convert.h"
|
|
#include "support/debug.h"
|
|
#include "support/docstream.h"
|
|
#include "support/gettext.h"
|
|
#include "support/lassert.h"
|
|
#include "support/lstrings.h"
|
|
#include "support/regex.h"
|
|
#include "support/textutils.h"
|
|
|
|
#include <set>
|
|
|
|
using namespace std;
|
|
using namespace lyx::support;
|
|
|
|
|
|
namespace lyx {
|
|
|
|
namespace {
|
|
|
|
// gets the "family name" from an author-type string
|
|
docstring familyName(docstring const & name)
|
|
{
|
|
if (name.empty())
|
|
return docstring();
|
|
|
|
// first we look for a comma, and take the last name to be everything
|
|
// preceding the right-most one, so that we also get the "jr" part.
|
|
docstring::size_type idx = name.rfind(',');
|
|
if (idx != docstring::npos)
|
|
return ltrim(name.substr(0, idx));
|
|
|
|
// OK, so now we want to look for the last name. We're going to
|
|
// include the "von" part. This isn't perfect.
|
|
// Split on spaces, to get various tokens.
|
|
vector<docstring> pieces = getVectorFromString(name, from_ascii(" "));
|
|
// If we only get two, assume the last one is the last name
|
|
if (pieces.size() <= 2)
|
|
return pieces.back();
|
|
|
|
// Now we look for the first token that begins with a lower case letter.
|
|
vector<docstring>::const_iterator it = pieces.begin();
|
|
vector<docstring>::const_iterator en = pieces.end();
|
|
for (; it != en; ++it) {
|
|
if ((*it).empty())
|
|
continue;
|
|
char_type const c = (*it)[0];
|
|
if (isLower(c))
|
|
break;
|
|
}
|
|
|
|
if (it == en) // we never found a "von"
|
|
return pieces.back();
|
|
|
|
// reconstruct what we need to return
|
|
docstring retval;
|
|
bool first = true;
|
|
for (; it != en; ++it) {
|
|
if (!first)
|
|
retval += " ";
|
|
else
|
|
first = false;
|
|
retval += *it;
|
|
}
|
|
return retval;
|
|
}
|
|
|
|
|
|
// converts a string containing LaTeX commands into unicode
|
|
// for display.
|
|
docstring convertLaTeXCommands(docstring const & str)
|
|
{
|
|
docstring val = str;
|
|
docstring ret;
|
|
|
|
bool scanning_cmd = false;
|
|
bool scanning_math = false;
|
|
bool escaped = false; // used to catch \$, etc.
|
|
while (!val.empty()) {
|
|
char_type const ch = val[0];
|
|
|
|
// if we're scanning math, we output everything until we
|
|
// find an unescaped $, at which point we break out.
|
|
if (scanning_math) {
|
|
if (escaped)
|
|
escaped = false;
|
|
else if (ch == '\\')
|
|
escaped = true;
|
|
else if (ch == '$')
|
|
scanning_math = false;
|
|
ret += ch;
|
|
val = val.substr(1);
|
|
continue;
|
|
}
|
|
|
|
// if we're scanning a command name, then we just
|
|
// discard characters until we hit something that
|
|
// isn't alpha.
|
|
if (scanning_cmd) {
|
|
if (isAlphaASCII(ch)) {
|
|
val = val.substr(1);
|
|
escaped = false;
|
|
continue;
|
|
}
|
|
// so we're done with this command.
|
|
// now we fall through and check this character.
|
|
scanning_cmd = false;
|
|
}
|
|
|
|
// was the last character a \? If so, then this is something like:
|
|
// \\ or \$, so we'll just output it. That's probably not always right...
|
|
if (escaped) {
|
|
// exception: output \, as THIN SPACE
|
|
if (ch == ',')
|
|
ret.push_back(0x2009);
|
|
else
|
|
ret += ch;
|
|
val = val.substr(1);
|
|
escaped = false;
|
|
continue;
|
|
}
|
|
|
|
if (ch == '$') {
|
|
ret += ch;
|
|
val = val.substr(1);
|
|
scanning_math = true;
|
|
continue;
|
|
}
|
|
|
|
// we just ignore braces
|
|
if (ch == '{' || ch == '}') {
|
|
val = val.substr(1);
|
|
continue;
|
|
}
|
|
|
|
// we're going to check things that look like commands, so if
|
|
// this doesn't, just output it.
|
|
if (ch != '\\') {
|
|
ret += ch;
|
|
val = val.substr(1);
|
|
continue;
|
|
}
|
|
|
|
// ok, could be a command of some sort
|
|
// let's see if it corresponds to some unicode
|
|
// unicodesymbols has things in the form: \"{u},
|
|
// whereas we may see things like: \"u. So we'll
|
|
// look for that and change it, if necessary.
|
|
// FIXME: This is a sort of mini-tex2lyx.
|
|
// Use the real tex2lyx instead!
|
|
static lyx::regex const reg("^\\\\\\W\\w");
|
|
if (lyx::regex_search(to_utf8(val), reg)) {
|
|
val.insert(3, from_ascii("}"));
|
|
val.insert(2, from_ascii("{"));
|
|
}
|
|
bool termination;
|
|
docstring rem;
|
|
docstring const cnvtd = Encodings::fromLaTeXCommand(val,
|
|
Encodings::TEXT_CMD, termination, rem);
|
|
if (!cnvtd.empty()) {
|
|
// it did, so we'll take that bit and proceed with what's left
|
|
ret += cnvtd;
|
|
val = rem;
|
|
continue;
|
|
}
|
|
// it's a command of some sort
|
|
scanning_cmd = true;
|
|
escaped = true;
|
|
val = val.substr(1);
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
|
|
// Escape '<' and '>' and remove richtext markers (e.g. {!this is richtext!}) from a string.
|
|
docstring processRichtext(docstring const & str, bool richtext)
|
|
{
|
|
docstring val = str;
|
|
docstring ret;
|
|
|
|
bool scanning_rich = false;
|
|
while (!val.empty()) {
|
|
char_type const ch = val[0];
|
|
if (ch == '{' && val.size() > 1 && val[1] == '!') {
|
|
// beginning of rich text
|
|
scanning_rich = true;
|
|
val = val.substr(2);
|
|
continue;
|
|
}
|
|
if (scanning_rich && ch == '!' && val.size() > 1 && val[1] == '}') {
|
|
// end of rich text
|
|
scanning_rich = false;
|
|
val = val.substr(2);
|
|
continue;
|
|
}
|
|
if (richtext) {
|
|
if (scanning_rich)
|
|
ret += ch;
|
|
else {
|
|
// we need to escape '<' and '>'
|
|
if (ch == '<')
|
|
ret += "<";
|
|
else if (ch == '>')
|
|
ret += ">";
|
|
else
|
|
ret += ch;
|
|
}
|
|
} else if (!scanning_rich /* && !richtext */)
|
|
ret += ch;
|
|
// else the character is discarded, which will happen only if
|
|
// richtext == false and we are scanning rich text
|
|
val = val.substr(1);
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
} // anon namespace
|
|
|
|
|
|
//////////////////////////////////////////////////////////////////////
|
|
//
|
|
// BibTeXInfo
|
|
//
|
|
//////////////////////////////////////////////////////////////////////
|
|
|
|
BibTeXInfo::BibTeXInfo(docstring const & key, docstring const & type)
|
|
: is_bibtex_(true), bib_key_(key), entry_type_(type), info_(),
|
|
modifier_(0)
|
|
{}
|
|
|
|
|
|
docstring const BibTeXInfo::getAbbreviatedAuthor(bool jurabib_style) const
|
|
{
|
|
if (!is_bibtex_) {
|
|
docstring const opt = label();
|
|
if (opt.empty())
|
|
return docstring();
|
|
|
|
docstring authors;
|
|
docstring const remainder = trim(split(opt, authors, '('));
|
|
if (remainder.empty())
|
|
// in this case, we didn't find a "(",
|
|
// so we don't have author (year)
|
|
return docstring();
|
|
return authors;
|
|
}
|
|
|
|
docstring author = operator[]("author");
|
|
if (author.empty()) {
|
|
author = operator[]("editor");
|
|
if (author.empty())
|
|
return author;
|
|
}
|
|
|
|
// FIXME Move this to a separate routine that can
|
|
// be called from elsewhere.
|
|
//
|
|
// OK, we've got some names. Let's format them.
|
|
// Try to split the author list on " and "
|
|
vector<docstring> const authors =
|
|
getVectorFromString(author, from_ascii(" and "));
|
|
|
|
if (jurabib_style && (authors.size() == 2 || authors.size() == 3)) {
|
|
docstring shortauthor = familyName(authors[0])
|
|
+ "/" + familyName(authors[1]);
|
|
if (authors.size() == 3)
|
|
shortauthor += "/" + familyName(authors[2]);
|
|
return convertLaTeXCommands(shortauthor);
|
|
}
|
|
|
|
docstring retval = familyName(authors[0]);
|
|
|
|
if (authors.size() == 2 && authors[1] != "others")
|
|
retval = bformat(from_ascii("%1$s and %2$s"),
|
|
familyName(authors[0]), familyName(authors[1]));
|
|
|
|
if (authors.size() >= 2)
|
|
retval = bformat(from_ascii("%1$s et al."),
|
|
familyName(authors[0]));
|
|
|
|
return convertLaTeXCommands(retval);
|
|
}
|
|
|
|
|
|
docstring const BibTeXInfo::getAbbreviatedAuthor(Buffer const & buf, bool jurabib_style) const
|
|
{
|
|
docstring const author = getAbbreviatedAuthor(jurabib_style);
|
|
if (!is_bibtex_)
|
|
return author;
|
|
vector<docstring> const authors = getVectorFromString(author, from_ascii(" and "));
|
|
if (authors.size() == 2)
|
|
return bformat(buf.B_("%1$s and %2$s"), authors[0], authors[1]);
|
|
docstring::size_type const idx = author.rfind(from_ascii(" et al."));
|
|
if (idx != docstring::npos)
|
|
return bformat(buf.B_("%1$s et al."), author.substr(0, idx));
|
|
return author;
|
|
}
|
|
|
|
|
|
docstring const BibTeXInfo::getYear() const
|
|
{
|
|
if (is_bibtex_)
|
|
return operator[]("year");
|
|
|
|
docstring const opt = label();
|
|
if (opt.empty())
|
|
return docstring();
|
|
|
|
docstring authors;
|
|
docstring tmp = split(opt, authors, '(');
|
|
if (tmp.empty())
|
|
// we don't have author (year)
|
|
return docstring();
|
|
docstring year;
|
|
tmp = split(tmp, year, ')');
|
|
return year;
|
|
}
|
|
|
|
|
|
docstring const BibTeXInfo::getXRef() const
|
|
{
|
|
if (!is_bibtex_)
|
|
return docstring();
|
|
return operator[]("crossref");
|
|
}
|
|
|
|
|
|
namespace {
|
|
|
|
docstring parseOptions(docstring const & format, string & optkey,
|
|
docstring & ifpart, docstring & elsepart);
|
|
|
|
// Calls parseOptions to deal with an embedded option, such as:
|
|
// {%number%[[, no.~%number%]]}
|
|
// which must appear at the start of format. ifelsepart gets the
|
|
// whole of the option, and we return what's left after the option.
|
|
// we return format if there is an error.
|
|
docstring parseEmbeddedOption(docstring const & format, docstring & ifelsepart)
|
|
{
|
|
LASSERT(format[0] == '{' && format[1] == '%', return format);
|
|
string optkey;
|
|
docstring ifpart;
|
|
docstring elsepart;
|
|
docstring const rest = parseOptions(format, optkey, ifpart, elsepart);
|
|
if (format == rest) { // parse error
|
|
LYXERR0("ERROR! Couldn't parse `" << format <<"'.");
|
|
return format;
|
|
}
|
|
LASSERT(rest.size() <= format.size(),
|
|
{ ifelsepart = docstring(); return format; });
|
|
ifelsepart = format.substr(0, format.size() - rest.size());
|
|
return rest;
|
|
}
|
|
|
|
|
|
// Gets a "clause" from a format string, where the clause is
|
|
// delimited by '[[' and ']]'. Returns what is left after the
|
|
// clause is removed, and returns format if there is an error.
|
|
docstring getClause(docstring const & format, docstring & clause)
|
|
{
|
|
docstring fmt = format;
|
|
// remove '[['
|
|
fmt = fmt.substr(2);
|
|
// we'll remove characters from the front of fmt as we
|
|
// deal with them
|
|
while (!fmt.empty()) {
|
|
if (fmt[0] == ']' && fmt.size() > 1 && fmt[1] == ']') {
|
|
// that's the end
|
|
fmt = fmt.substr(2);
|
|
break;
|
|
}
|
|
// check for an embedded option
|
|
if (fmt[0] == '{' && fmt.size() > 1 && fmt[1] == '%') {
|
|
docstring part;
|
|
docstring const rest = parseEmbeddedOption(fmt, part);
|
|
if (fmt == rest) {
|
|
LYXERR0("ERROR! Couldn't parse embedded option in `" << format <<"'.");
|
|
return format;
|
|
}
|
|
clause += part;
|
|
fmt = rest;
|
|
} else { // it's just a normal character
|
|
clause += fmt[0];
|
|
fmt = fmt.substr(1);
|
|
}
|
|
}
|
|
return fmt;
|
|
}
|
|
|
|
|
|
// parse an options string, which must appear at the start of the
|
|
// format parameter. puts the parsed bits in optkey, ifpart, and
|
|
// elsepart and returns what's left after the option is removed.
|
|
// if there's an error, it returns format itself.
|
|
docstring parseOptions(docstring const & format, string & optkey,
|
|
docstring & ifpart, docstring & elsepart)
|
|
{
|
|
LASSERT(format[0] == '{' && format[1] == '%', return format);
|
|
// strip '{%'
|
|
docstring fmt = format.substr(2);
|
|
size_t pos = fmt.find('%'); // end of key
|
|
if (pos == string::npos) {
|
|
LYXERR0("Error parsing `" << format <<"'. Can't find end of key.");
|
|
return format;
|
|
}
|
|
optkey = to_utf8(fmt.substr(0, pos));
|
|
fmt = fmt.substr(pos + 1);
|
|
// [[format]] should be next
|
|
if (fmt[0] != '[' || fmt[1] != '[') {
|
|
LYXERR0("Error parsing `" << format <<"'. Can't find '[[' after key.");
|
|
return format;
|
|
}
|
|
|
|
docstring curfmt = fmt;
|
|
fmt = getClause(curfmt, ifpart);
|
|
if (fmt == curfmt) {
|
|
LYXERR0("Error parsing `" << format <<"'. Couldn't get if clause.");
|
|
return format;
|
|
}
|
|
|
|
if (fmt[0] == '}') // we're done, no else clause
|
|
return fmt.substr(1);
|
|
|
|
// else part should follow
|
|
if (fmt[0] != '[' || fmt[1] != '[') {
|
|
LYXERR0("Error parsing `" << format <<"'. Can't find else clause.");
|
|
return format;
|
|
}
|
|
|
|
curfmt = fmt;
|
|
fmt = getClause(curfmt, elsepart);
|
|
// we should be done
|
|
if (fmt == curfmt || fmt[0] != '}') {
|
|
LYXERR0("Error parsing `" << format <<"'. Can't find end of option.");
|
|
return format;
|
|
}
|
|
return fmt.substr(1);
|
|
}
|
|
|
|
|
|
} // anon namespace
|
|
|
|
/* FIXME
|
|
Bug #9131 revealed an oddity in how we are generating citation information
|
|
when more than one key is given. We end up building a longer and longer format
|
|
string as we go, which we then have to re-parse, over and over and over again,
|
|
rather than generating the information for the individual keys and then putting
|
|
all of that together. We do that to deal with the way separators work, from what
|
|
I can tell, but it still feels like a hack. Fixing this would require quite a
|
|
bit of work, however.
|
|
*/
|
|
docstring BibTeXInfo::expandFormat(docstring const & format,
|
|
BibTeXInfo const * const xref, int & counter, Buffer const & buf,
|
|
docstring before, docstring after, docstring dialog, bool next) const
|
|
{
|
|
// incorrect use of macros could put us in an infinite loop
|
|
static int const max_passes = 5000;
|
|
// the use of overly large keys can lead to performance problems, due
|
|
// to eventual attempts to convert LaTeX macros to unicode. See bug
|
|
// #8944. This is perhaps not the best solution, but it will have to
|
|
// do for now.
|
|
static size_t const max_keysize = 128;
|
|
odocstringstream ret; // return value
|
|
string key;
|
|
bool scanning_key = false;
|
|
bool scanning_rich = false;
|
|
|
|
CiteEngineType const engine_type = buf.params().citeEngineType();
|
|
docstring fmt = format;
|
|
// we'll remove characters from the front of fmt as we
|
|
// deal with them
|
|
while (!fmt.empty()) {
|
|
if (counter > max_passes) {
|
|
LYXERR0("Recursion limit reached while parsing `"
|
|
<< format << "'.");
|
|
return _("ERROR!");
|
|
}
|
|
|
|
char_type thischar = fmt[0];
|
|
if (thischar == '%') {
|
|
// beginning or end of key
|
|
if (scanning_key) {
|
|
// end of key
|
|
scanning_key = false;
|
|
// so we replace the key with its value, which may be empty
|
|
if (key[0] == '!') {
|
|
// macro
|
|
string const val =
|
|
buf.params().documentClass().getCiteMacro(engine_type, key);
|
|
fmt = from_utf8(val) + fmt.substr(1);
|
|
counter += 1;
|
|
continue;
|
|
} else if (key[0] == '_') {
|
|
// a translatable bit
|
|
string const val =
|
|
buf.params().documentClass().getCiteMacro(engine_type, key);
|
|
docstring const trans =
|
|
translateIfPossible(from_utf8(val), buf.params().language->code());
|
|
ret << trans;
|
|
} else {
|
|
docstring const val =
|
|
getValueForKey(key, buf, before, after, dialog, xref, max_keysize);
|
|
if (!scanning_rich)
|
|
ret << from_ascii("{!<span class=\"bib-" + key + "\">!}");
|
|
ret << val;
|
|
if (!scanning_rich)
|
|
ret << from_ascii("{!</span>!}");
|
|
}
|
|
} else {
|
|
// beginning of key
|
|
key.clear();
|
|
scanning_key = true;
|
|
}
|
|
}
|
|
else if (thischar == '{') {
|
|
// beginning of option?
|
|
if (scanning_key) {
|
|
LYXERR0("ERROR: Found `{' when scanning key in `" << format << "'.");
|
|
return _("ERROR!");
|
|
}
|
|
if (fmt.size() > 1) {
|
|
if (fmt[1] == '%') {
|
|
// it is the beginning of an optional format
|
|
string optkey;
|
|
docstring ifpart;
|
|
docstring elsepart;
|
|
docstring const newfmt =
|
|
parseOptions(fmt, optkey, ifpart, elsepart);
|
|
if (newfmt == fmt) // parse error
|
|
return _("ERROR!");
|
|
fmt = newfmt;
|
|
docstring const val =
|
|
getValueForKey(optkey, buf, before, after, dialog, xref);
|
|
if (optkey == "next" && next)
|
|
ret << ifpart; // without expansion
|
|
else if (!val.empty()) {
|
|
int newcounter = 0;
|
|
ret << expandFormat(ifpart, xref, newcounter, buf,
|
|
before, after, dialog, next);
|
|
} else if (!elsepart.empty()) {
|
|
int newcounter = 0;
|
|
ret << expandFormat(elsepart, xref, newcounter, buf,
|
|
before, after, dialog, next);
|
|
}
|
|
// fmt will have been shortened for us already
|
|
continue;
|
|
}
|
|
if (fmt[1] == '!') {
|
|
// beginning of rich text
|
|
scanning_rich = true;
|
|
fmt = fmt.substr(2);
|
|
ret << from_ascii("{!");
|
|
continue;
|
|
}
|
|
}
|
|
// we are here if '{' was not followed by % or !.
|
|
// So it's just a character.
|
|
ret << thischar;
|
|
}
|
|
else if (scanning_rich && thischar == '!'
|
|
&& fmt.size() > 1 && fmt[1] == '}') {
|
|
// end of rich text
|
|
scanning_rich = false;
|
|
fmt = fmt.substr(2);
|
|
ret << from_ascii("!}");
|
|
continue;
|
|
}
|
|
else if (scanning_key)
|
|
key += char(thischar);
|
|
else {
|
|
try {
|
|
ret.put(thischar);
|
|
} catch (EncodingException & /* e */) {
|
|
LYXERR0("Uncodable character '" << docstring(1, thischar) << " in citation label!");
|
|
}
|
|
}
|
|
fmt = fmt.substr(1);
|
|
} // for loop
|
|
if (scanning_key) {
|
|
LYXERR0("Never found end of key in `" << format << "'!");
|
|
return _("ERROR!");
|
|
}
|
|
if (scanning_rich) {
|
|
LYXERR0("Never found end of rich text in `" << format << "'!");
|
|
return _("ERROR!");
|
|
}
|
|
return ret.str();
|
|
}
|
|
|
|
|
|
docstring const & BibTeXInfo::getInfo(BibTeXInfo const * const xref,
|
|
Buffer const & buf, bool richtext) const
|
|
{
|
|
if (!richtext && !info_.empty())
|
|
return info_;
|
|
if (richtext && !info_richtext_.empty())
|
|
return info_richtext_;
|
|
|
|
if (!is_bibtex_) {
|
|
BibTeXInfo::const_iterator it = find(from_ascii("ref"));
|
|
info_ = it->second;
|
|
return info_;
|
|
}
|
|
|
|
CiteEngineType const engine_type = buf.params().citeEngineType();
|
|
DocumentClass const & dc = buf.params().documentClass();
|
|
docstring const & format =
|
|
from_utf8(dc.getCiteFormat(engine_type, to_utf8(entry_type_)));
|
|
int counter = 0;
|
|
info_ = expandFormat(format, xref, counter, buf,
|
|
docstring(), docstring(), docstring(), false);
|
|
|
|
if (info_.empty()) {
|
|
// this probably shouldn't happen
|
|
return info_;
|
|
}
|
|
|
|
if (richtext) {
|
|
info_richtext_ = convertLaTeXCommands(processRichtext(info_, true));
|
|
return info_richtext_;
|
|
}
|
|
|
|
info_ = convertLaTeXCommands(processRichtext(info_, false));
|
|
return info_;
|
|
}
|
|
|
|
|
|
docstring const BibTeXInfo::getLabel(BibTeXInfo const * const xref,
|
|
Buffer const & buf, docstring const & format, bool richtext,
|
|
docstring const & before, docstring const & after,
|
|
docstring const & dialog, bool next) const
|
|
{
|
|
docstring loclabel;
|
|
|
|
int counter = 0;
|
|
loclabel = expandFormat(format, xref, counter, buf,
|
|
before, after, dialog, next);
|
|
|
|
if (!loclabel.empty() && !next) {
|
|
loclabel = processRichtext(loclabel, richtext);
|
|
loclabel = convertLaTeXCommands(loclabel);
|
|
}
|
|
|
|
return loclabel;
|
|
}
|
|
|
|
|
|
docstring const & BibTeXInfo::operator[](docstring const & field) const
|
|
{
|
|
BibTeXInfo::const_iterator it = find(field);
|
|
if (it != end())
|
|
return it->second;
|
|
static docstring const empty_value = docstring();
|
|
return empty_value;
|
|
}
|
|
|
|
|
|
docstring const & BibTeXInfo::operator[](string const & field) const
|
|
{
|
|
return operator[](from_ascii(field));
|
|
}
|
|
|
|
|
|
docstring BibTeXInfo::getValueForKey(string const & oldkey, Buffer const & buf,
|
|
docstring const & before, docstring const & after, docstring const & dialog,
|
|
BibTeXInfo const * const xref, size_t maxsize) const
|
|
{
|
|
// anything less is pointless
|
|
LASSERT(maxsize >= 16, maxsize = 16);
|
|
string key = oldkey;
|
|
bool cleanit = false;
|
|
if (prefixIs(oldkey, "clean:")) {
|
|
key = oldkey.substr(6);
|
|
cleanit = true;
|
|
}
|
|
|
|
docstring ret = operator[](key);
|
|
if (ret.empty() && xref)
|
|
ret = (*xref)[key];
|
|
if (ret.empty()) {
|
|
// some special keys
|
|
// FIXME: dialog, textbefore and textafter have nothing to do with this
|
|
if (key == "dialog")
|
|
ret = dialog;
|
|
else if (key == "entrytype")
|
|
ret = entry_type_;
|
|
else if (key == "key")
|
|
ret = bib_key_;
|
|
else if (key == "label")
|
|
ret = label_;
|
|
else if (key == "modifier" && modifier_ != 0)
|
|
ret = modifier_;
|
|
else if (key == "numericallabel")
|
|
ret = cite_number_;
|
|
else if (key == "abbrvauthor")
|
|
// Special key to provide abbreviated author names.
|
|
ret = getAbbreviatedAuthor(buf, false);
|
|
else if (key == "shortauthor")
|
|
// When shortauthor is not defined, jurabib automatically
|
|
// provides jurabib-style abbreviated author names. We do
|
|
// this as well.
|
|
ret = getAbbreviatedAuthor(buf, true);
|
|
else if (key == "shorttitle") {
|
|
// When shorttitle is not defined, jurabib uses for `article'
|
|
// and `periodical' entries the form `journal volume [year]'
|
|
// and for other types of entries it uses the `title' field.
|
|
if (entry_type_ == "article" || entry_type_ == "periodical")
|
|
ret = operator[]("journal") + " " + operator[]("volume")
|
|
+ " [" + operator[]("year") + "]";
|
|
else
|
|
ret = operator[]("title");
|
|
} else if (key == "bibentry") {
|
|
// Special key to provide the full bibliography entry: see getInfo()
|
|
CiteEngineType const engine_type = buf.params().citeEngineType();
|
|
DocumentClass const & dc = buf.params().documentClass();
|
|
docstring const & format =
|
|
from_utf8(dc.getCiteFormat(engine_type, to_utf8(entry_type_)));
|
|
int counter = 0;
|
|
ret = expandFormat(format, xref, counter, buf,
|
|
docstring(), docstring(), docstring(), false);
|
|
} else if (key == "textbefore")
|
|
ret = before;
|
|
else if (key == "textafter")
|
|
ret = after;
|
|
else if (key == "year")
|
|
ret = getYear();
|
|
}
|
|
|
|
if (cleanit)
|
|
ret = html::cleanAttr(ret);
|
|
|
|
// make sure it is not too big
|
|
if (ret.size() > maxsize)
|
|
ret = ret.substr(0, maxsize - 3) + from_ascii("...");
|
|
return ret;
|
|
}
|
|
|
|
|
|
//////////////////////////////////////////////////////////////////////
|
|
//
|
|
// BiblioInfo
|
|
//
|
|
//////////////////////////////////////////////////////////////////////
|
|
|
|
namespace {
|
|
|
|
// A functor for use with sort, leading to case insensitive sorting
|
|
class compareNoCase: public binary_function<docstring, docstring, bool>
|
|
{
|
|
public:
|
|
bool operator()(docstring const & s1, docstring const & s2) const {
|
|
return compare_no_case(s1, s2) < 0;
|
|
}
|
|
};
|
|
|
|
} // namespace anon
|
|
|
|
|
|
vector<docstring> const BiblioInfo::getKeys() const
|
|
{
|
|
vector<docstring> bibkeys;
|
|
BiblioInfo::const_iterator it = begin();
|
|
for (; it != end(); ++it)
|
|
bibkeys.push_back(it->first);
|
|
sort(bibkeys.begin(), bibkeys.end(), compareNoCase());
|
|
return bibkeys;
|
|
}
|
|
|
|
|
|
vector<docstring> const BiblioInfo::getFields() const
|
|
{
|
|
vector<docstring> bibfields;
|
|
set<docstring>::const_iterator it = field_names_.begin();
|
|
set<docstring>::const_iterator end = field_names_.end();
|
|
for (; it != end; ++it)
|
|
bibfields.push_back(*it);
|
|
sort(bibfields.begin(), bibfields.end());
|
|
return bibfields;
|
|
}
|
|
|
|
|
|
vector<docstring> const BiblioInfo::getEntries() const
|
|
{
|
|
vector<docstring> bibentries;
|
|
set<docstring>::const_iterator it = entry_types_.begin();
|
|
set<docstring>::const_iterator end = entry_types_.end();
|
|
for (; it != end; ++it)
|
|
bibentries.push_back(*it);
|
|
sort(bibentries.begin(), bibentries.end());
|
|
return bibentries;
|
|
}
|
|
|
|
|
|
docstring const BiblioInfo::getAbbreviatedAuthor(docstring const & key, Buffer const & buf) const
|
|
{
|
|
BiblioInfo::const_iterator it = find(key);
|
|
if (it == end())
|
|
return docstring();
|
|
BibTeXInfo const & data = it->second;
|
|
return data.getAbbreviatedAuthor(buf, false);
|
|
}
|
|
|
|
|
|
docstring const BiblioInfo::getCiteNumber(docstring const & key) const
|
|
{
|
|
BiblioInfo::const_iterator it = find(key);
|
|
if (it == end())
|
|
return docstring();
|
|
BibTeXInfo const & data = it->second;
|
|
return data.citeNumber();
|
|
}
|
|
|
|
|
|
docstring const BiblioInfo::getYear(docstring const & key, bool use_modifier) const
|
|
{
|
|
BiblioInfo::const_iterator it = find(key);
|
|
if (it == end())
|
|
return docstring();
|
|
BibTeXInfo const & data = it->second;
|
|
docstring year = data.getYear();
|
|
if (year.empty()) {
|
|
// let's try the crossref
|
|
docstring const xref = data.getXRef();
|
|
if (xref.empty())
|
|
// no luck
|
|
return docstring();
|
|
BiblioInfo::const_iterator const xrefit = find(xref);
|
|
if (xrefit == end())
|
|
// no luck again
|
|
return docstring();
|
|
BibTeXInfo const & xref_data = xrefit->second;
|
|
year = xref_data.getYear();
|
|
}
|
|
if (use_modifier && data.modifier() != 0)
|
|
year += data.modifier();
|
|
return year;
|
|
}
|
|
|
|
|
|
docstring const BiblioInfo::getYear(docstring const & key, Buffer const & buf, bool use_modifier) const
|
|
{
|
|
docstring const year = getYear(key, use_modifier);
|
|
if (year.empty())
|
|
return buf.B_("No year");
|
|
return year;
|
|
}
|
|
|
|
|
|
docstring const BiblioInfo::getInfo(docstring const & key,
|
|
Buffer const & buf, bool richtext) const
|
|
{
|
|
BiblioInfo::const_iterator it = find(key);
|
|
if (it == end())
|
|
return docstring(_("Bibliography entry not found!"));
|
|
BibTeXInfo const & data = it->second;
|
|
BibTeXInfo const * xrefptr = 0;
|
|
docstring const xref = data.getXRef();
|
|
if (!xref.empty()) {
|
|
BiblioInfo::const_iterator const xrefit = find(xref);
|
|
if (xrefit != end())
|
|
xrefptr = &(xrefit->second);
|
|
}
|
|
return data.getInfo(xrefptr, buf, richtext);
|
|
}
|
|
|
|
|
|
docstring const BiblioInfo::getLabel(vector<docstring> keys,
|
|
Buffer const & buf, string const & style, bool for_xhtml,
|
|
size_t max_size, docstring const & before, docstring const & after,
|
|
docstring const & dialog) const
|
|
{
|
|
// shorter makes no sense
|
|
LASSERT(max_size >= 16, max_size = 16);
|
|
|
|
// we can't display more than 10 of these, anyway
|
|
bool const too_many_keys = keys.size() > 10;
|
|
if (too_many_keys)
|
|
keys.resize(10);
|
|
|
|
CiteEngineType const engine_type = buf.params().citeEngineType();
|
|
DocumentClass const & dc = buf.params().documentClass();
|
|
docstring const & format = from_utf8(dc.getCiteFormat(engine_type, style, "cite"));
|
|
docstring ret = format;
|
|
vector<docstring>::const_iterator key = keys.begin();
|
|
vector<docstring>::const_iterator ken = keys.end();
|
|
for (; key != ken; ++key) {
|
|
BiblioInfo::const_iterator it = find(*key);
|
|
BibTeXInfo empty_data;
|
|
empty_data.key(*key);
|
|
BibTeXInfo & data = empty_data;
|
|
BibTeXInfo const * xrefptr = 0;
|
|
if (it != end()) {
|
|
data = it->second;
|
|
docstring const xref = data.getXRef();
|
|
if (!xref.empty()) {
|
|
BiblioInfo::const_iterator const xrefit = find(xref);
|
|
if (xrefit != end())
|
|
xrefptr = &(xrefit->second);
|
|
}
|
|
}
|
|
ret = data.getLabel(xrefptr, buf, ret, for_xhtml,
|
|
before, after, dialog, key + 1 != ken);
|
|
}
|
|
|
|
if (ret.size() > max_size) {
|
|
ret.resize(max_size - 3);
|
|
ret += "...";
|
|
} else if (too_many_keys) {
|
|
if (ret.size() > max_size - 3)
|
|
ret.resize(max_size - 3);
|
|
ret += "...";
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
|
|
bool BiblioInfo::isBibtex(docstring const & key) const
|
|
{
|
|
docstring key1;
|
|
split(key, key1, ',');
|
|
BiblioInfo::const_iterator it = find(key1);
|
|
if (it == end())
|
|
return false;
|
|
return it->second.isBibTeX();
|
|
}
|
|
|
|
|
|
vector<docstring> const BiblioInfo::getCiteStrings(
|
|
vector<docstring> const & keys, vector<CitationStyle> const & styles,
|
|
Buffer const & buf, docstring const & before,
|
|
docstring const & after, docstring const & dialog, size_t max_size) const
|
|
{
|
|
if (empty())
|
|
return vector<docstring>();
|
|
|
|
string style;
|
|
vector<docstring> vec(styles.size());
|
|
for (size_t i = 0; i != vec.size(); ++i) {
|
|
style = styles[i].cmd;
|
|
vec[i] = getLabel(keys, buf, style, false, max_size, before, after, dialog);
|
|
}
|
|
|
|
return vec;
|
|
}
|
|
|
|
|
|
void BiblioInfo::mergeBiblioInfo(BiblioInfo const & info)
|
|
{
|
|
bimap_.insert(info.begin(), info.end());
|
|
field_names_.insert(info.field_names_.begin(), info.field_names_.end());
|
|
entry_types_.insert(info.entry_types_.begin(), info.entry_types_.end());
|
|
}
|
|
|
|
|
|
namespace {
|
|
|
|
// used in xhtml to sort a list of BibTeXInfo objects
|
|
bool lSorter(BibTeXInfo const * lhs, BibTeXInfo const * rhs)
|
|
{
|
|
docstring const lauth = lhs->getAbbreviatedAuthor();
|
|
docstring const rauth = rhs->getAbbreviatedAuthor();
|
|
docstring const lyear = lhs->getYear();
|
|
docstring const ryear = rhs->getYear();
|
|
docstring const ltitl = lhs->operator[]("title");
|
|
docstring const rtitl = rhs->operator[]("title");
|
|
return (lauth < rauth)
|
|
|| (lauth == rauth && lyear < ryear)
|
|
|| (lauth == rauth && lyear == ryear && ltitl < rtitl);
|
|
}
|
|
|
|
}
|
|
|
|
|
|
void BiblioInfo::collectCitedEntries(Buffer const & buf)
|
|
{
|
|
cited_entries_.clear();
|
|
// We are going to collect all the citation keys used in the document,
|
|
// getting them from the TOC.
|
|
// FIXME We may want to collect these differently, in the first case,
|
|
// so that we might have them in order of appearance.
|
|
set<docstring> citekeys;
|
|
Toc const & toc = buf.tocBackend().toc("citation");
|
|
Toc::const_iterator it = toc.begin();
|
|
Toc::const_iterator const en = toc.end();
|
|
for (; it != en; ++it) {
|
|
if (it->str().empty())
|
|
continue;
|
|
vector<docstring> const keys = getVectorFromString(it->str());
|
|
citekeys.insert(keys.begin(), keys.end());
|
|
}
|
|
if (citekeys.empty())
|
|
return;
|
|
|
|
// We have a set of the keys used in this document.
|
|
// We will now convert it to a list of the BibTeXInfo objects used in
|
|
// this document...
|
|
vector<BibTeXInfo const *> bi;
|
|
set<docstring>::const_iterator cit = citekeys.begin();
|
|
set<docstring>::const_iterator const cen = citekeys.end();
|
|
for (; cit != cen; ++cit) {
|
|
BiblioInfo::const_iterator const bt = find(*cit);
|
|
if (bt == end() || !bt->second.isBibTeX())
|
|
continue;
|
|
bi.push_back(&(bt->second));
|
|
}
|
|
// ...and sort it.
|
|
sort(bi.begin(), bi.end(), lSorter);
|
|
|
|
// Now we can write the sorted keys
|
|
vector<BibTeXInfo const *>::const_iterator bit = bi.begin();
|
|
vector<BibTeXInfo const *>::const_iterator ben = bi.end();
|
|
for (; bit != ben; ++bit)
|
|
cited_entries_.push_back((*bit)->key());
|
|
}
|
|
|
|
|
|
void BiblioInfo::makeCitationLabels(Buffer const & buf)
|
|
{
|
|
collectCitedEntries(buf);
|
|
CiteEngineType const engine_type = buf.params().citeEngineType();
|
|
bool const numbers = (engine_type & ENGINE_TYPE_NUMERICAL);
|
|
|
|
int keynumber = 0;
|
|
char modifier = 0;
|
|
// used to remember the last one we saw
|
|
// we'll be comparing entries to see if we need to add
|
|
// modifiers, like "1984a"
|
|
map<docstring, BibTeXInfo>::iterator last;
|
|
|
|
vector<docstring>::const_iterator it = cited_entries_.begin();
|
|
vector<docstring>::const_iterator const en = cited_entries_.end();
|
|
for (; it != en; ++it) {
|
|
map<docstring, BibTeXInfo>::iterator const biit = bimap_.find(*it);
|
|
// this shouldn't happen, but...
|
|
if (biit == bimap_.end())
|
|
// ...fail gracefully, anyway.
|
|
continue;
|
|
BibTeXInfo & entry = biit->second;
|
|
if (numbers) {
|
|
docstring const num = convert<docstring>(++keynumber);
|
|
entry.setCiteNumber(num);
|
|
} else {
|
|
if (it != cited_entries_.begin()
|
|
&& entry.getAbbreviatedAuthor() == last->second.getAbbreviatedAuthor()
|
|
// we access the year via getYear() so as to get it from the xref,
|
|
// if we need to do so
|
|
&& getYear(entry.key()) == getYear(last->second.key())) {
|
|
if (modifier == 0) {
|
|
// so the last one should have been 'a'
|
|
last->second.setModifier('a');
|
|
modifier = 'b';
|
|
} else if (modifier == 'z')
|
|
modifier = 'A';
|
|
else
|
|
modifier++;
|
|
} else {
|
|
modifier = 0;
|
|
}
|
|
entry.setModifier(modifier);
|
|
// remember the last one
|
|
last = biit;
|
|
}
|
|
}
|
|
// Set the labels
|
|
it = cited_entries_.begin();
|
|
for (; it != en; ++it) {
|
|
map<docstring, BibTeXInfo>::iterator const biit = bimap_.find(*it);
|
|
// this shouldn't happen, but...
|
|
if (biit == bimap_.end())
|
|
// ...fail gracefully, anyway.
|
|
continue;
|
|
BibTeXInfo & entry = biit->second;
|
|
if (numbers) {
|
|
entry.label(entry.citeNumber());
|
|
} else {
|
|
docstring const auth = entry.getAbbreviatedAuthor(buf, false);
|
|
// we do it this way so as to access the xref, if necessary
|
|
// note that this also gives us the modifier
|
|
docstring const year = getYear(*it, buf, true);
|
|
if (!auth.empty() && !year.empty())
|
|
entry.label(auth + ' ' + year);
|
|
else
|
|
entry.label(entry.key());
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
//////////////////////////////////////////////////////////////////////
|
|
//
|
|
// CitationStyle
|
|
//
|
|
//////////////////////////////////////////////////////////////////////
|
|
|
|
|
|
CitationStyle citationStyleFromString(string const & command)
|
|
{
|
|
CitationStyle cs;
|
|
if (command.empty())
|
|
return cs;
|
|
|
|
string cmd = command;
|
|
if (cmd[0] == 'C') {
|
|
cs.forceUpperCase = true;
|
|
cmd[0] = 'c';
|
|
}
|
|
|
|
size_t const n = cmd.size() - 1;
|
|
if (cmd[n] == '*') {
|
|
cs.fullAuthorList = true;
|
|
cmd = cmd.substr(0, n);
|
|
}
|
|
|
|
cs.cmd = cmd;
|
|
return cs;
|
|
}
|
|
|
|
|
|
string citationStyleToString(const CitationStyle & cs)
|
|
{
|
|
string cmd = cs.cmd;
|
|
if (cs.forceUpperCase)
|
|
cmd[0] = 'C';
|
|
if (cs.fullAuthorList)
|
|
cmd += '*';
|
|
return cmd;
|
|
}
|
|
|
|
} // namespace lyx
|