Java 5 + Writer4LaTeX + bugfixes
git-svn-id: svn://svn.code.sf.net/p/writer2latex/code/trunk@11 f0f2a975-2e09-46c8-9428-3b39399b9f3c
This commit is contained in:
parent
be54e842f4
commit
9241a44f6c
83 changed files with 2373 additions and 631 deletions
|
@ -83,27 +83,27 @@ public class TextConverter extends ConverterHelper {
|
|||
private int nLastSplitLevel = 1; // The outline level at which the last split occured
|
||||
private int nDontSplitLevel = 0; // if > 0 splitting is forbidden
|
||||
boolean bAfterHeading=false; // last element was a top level heading
|
||||
protected Stack sections = new Stack(); // stack of nested sections
|
||||
protected Stack<Node> sections = new Stack<Node>(); // stack of nested sections
|
||||
Element[] currentHeading = new Element[7]; // Last headings (repeated when splitting)
|
||||
|
||||
// Counters for generated numbers
|
||||
private ListCounter outlineNumbering;
|
||||
private Hashtable listCounters = new Hashtable();
|
||||
private Hashtable<String, ListCounter> listCounters = new Hashtable<String, ListCounter>();
|
||||
private String sCurrentListLabel = null;
|
||||
|
||||
// Mode used to handle floats (depends on source doc type and config)
|
||||
private int nFloatMode;
|
||||
|
||||
// Data used for index bookkeeping
|
||||
private Vector indexes = new Vector();
|
||||
private Vector<IndexData> indexes = new Vector<IndexData>();
|
||||
|
||||
// Data used to handle Alphabetical Index
|
||||
Vector index = new Vector(); // All words for the index
|
||||
Vector<AlphabeticalEntry> index = new Vector<AlphabeticalEntry>(); // All words for the index
|
||||
private int nIndexIndex = -1; // Current index used for id's (of form idxN)
|
||||
private int nAlphabeticalIndex = -1; // File containing alphabetical index
|
||||
|
||||
// Data used to handle Table of Contents
|
||||
private Vector tocEntries = new Vector(); // All potential(!) toc items
|
||||
private Vector<TocEntry> tocEntries = new Vector<TocEntry>(); // All potential(!) toc items
|
||||
private int nTocFileIndex = -1; // file index for main toc
|
||||
private Element currentChapter = null; // Node for the current chapter (level 1) heading
|
||||
private int nTocIndex = -1; // Current index for id's (of form tocN)
|
||||
|
@ -116,8 +116,8 @@ public class TextConverter extends ConverterHelper {
|
|||
private String sEntCitStyle = null;
|
||||
|
||||
// Gather the footnotes and endnotes
|
||||
private LinkedList footnotes = new LinkedList();
|
||||
private LinkedList endnotes = new LinkedList();
|
||||
private LinkedList<Node> footnotes = new LinkedList<Node>();
|
||||
private LinkedList<Node> endnotes = new LinkedList<Node>();
|
||||
|
||||
// Sometimes we have to create an inlinenode in a block context
|
||||
// (labels for footnotes and endnotes)
|
||||
|
@ -169,7 +169,7 @@ public class TextConverter extends ConverterHelper {
|
|||
// Generate all indexes
|
||||
int nIndexCount = indexes.size();
|
||||
for (int i=0; i<nIndexCount; i++) {
|
||||
generateToc((IndexData) indexes.get(i));
|
||||
generateToc(indexes.get(i));
|
||||
}
|
||||
|
||||
// Generate navigation links
|
||||
|
@ -214,7 +214,7 @@ public class TextConverter extends ConverterHelper {
|
|||
// Get the last heading of level <= split level for this file
|
||||
TocEntry entryCurrent = null;
|
||||
for (int i=nLen-1; i>=0; i--) {
|
||||
TocEntry entry = (TocEntry) tocEntries.get(i);
|
||||
TocEntry entry = tocEntries.get(i);
|
||||
if (XMLString.TEXT_H.equals(entry.onode.getTagName()) && entry.nFileIndex==nIndex && entry.nOutlineLevel<=nSplit) {
|
||||
entryCurrent = entry; break;
|
||||
}
|
||||
|
@ -237,7 +237,7 @@ public class TextConverter extends ConverterHelper {
|
|||
|
||||
int nPrevFileIndex = 0;
|
||||
for (int i=0; i<nLen; i++) {
|
||||
TocEntry entry = (TocEntry) tocEntries.get(i);
|
||||
TocEntry entry = tocEntries.get(i);
|
||||
|
||||
if (entry.nFileIndex>nPrevFileIndex+1) {
|
||||
// Skipping a file index means we have passed an index
|
||||
|
@ -660,7 +660,7 @@ public class TextConverter extends ConverterHelper {
|
|||
else if (style!=null) {
|
||||
// Get existing or create new counter
|
||||
if (listCounters.containsKey(style.getName())) {
|
||||
return (ListCounter) listCounters.get(style.getName());
|
||||
return listCounters.get(style.getName());
|
||||
}
|
||||
else {
|
||||
ListCounter counter = new ListCounter(style);
|
||||
|
@ -1066,7 +1066,7 @@ public class TextConverter extends ConverterHelper {
|
|||
// Find the chapter
|
||||
if (tocReader.isByChapter() && chapter!=null) {
|
||||
for (int i=0; i<nLen; i++) {
|
||||
TocEntry entry = (TocEntry) tocEntries.get(i);
|
||||
TocEntry entry = tocEntries.get(i);
|
||||
if (entry.onode==chapter) { nStart=i; break; }
|
||||
}
|
||||
|
||||
|
@ -1074,7 +1074,7 @@ public class TextConverter extends ConverterHelper {
|
|||
|
||||
// Generate entries
|
||||
for (int i=nStart; i<nLen; i++) {
|
||||
TocEntry entry = (TocEntry) tocEntries.get(i);
|
||||
TocEntry entry = tocEntries.get(i);
|
||||
String sNodeName = entry.onode.getTagName();
|
||||
if (XMLString.TEXT_H.equals(sNodeName)) {
|
||||
int nLevel = getOutlineLevel(entry.onode);
|
||||
|
@ -1219,8 +1219,8 @@ public class TextConverter extends ConverterHelper {
|
|||
}
|
||||
for (int i = 0; i<=nIndexIndex; i++) {
|
||||
for (int j = i+1; j<=nIndexIndex ; j++) {
|
||||
AlphabeticalEntry entryi = (AlphabeticalEntry) index.get(i);
|
||||
AlphabeticalEntry entryj = (AlphabeticalEntry) index.get(j);
|
||||
AlphabeticalEntry entryi = index.get(i);
|
||||
AlphabeticalEntry entryj = index.get(j);
|
||||
if (collator.compare(entryi.sWord, entryj.sWord) > 0) {
|
||||
index.set(i,entryj);
|
||||
index.set(j,entryi);
|
||||
|
@ -1243,7 +1243,7 @@ public class TextConverter extends ConverterHelper {
|
|||
int nColIndex = -1;
|
||||
for (int i=0; i<=nIndexIndex; i++) {
|
||||
if (i%nColEntries==0) { nColIndex++; }
|
||||
AlphabeticalEntry entry = (AlphabeticalEntry) index.get(i);
|
||||
AlphabeticalEntry entry = index.get(i);
|
||||
Element p = createParagraph(td[nColIndex],sEntryStyleName);
|
||||
Element a = converter.createLink("idx"+entry.nIndex);
|
||||
p.appendChild(a);
|
||||
|
@ -1517,7 +1517,7 @@ public class TextConverter extends ConverterHelper {
|
|||
public void insertFootnotes(Node hnode) {
|
||||
int n = footnotes.size();
|
||||
for (int i=0; i<n; i++) {
|
||||
Node footnote = (Node) footnotes.get(i);
|
||||
Node footnote = footnotes.get(i);
|
||||
String sId = Misc.getAttribute(footnote,XMLString.TEXT_ID);
|
||||
Node citation = Misc.getChildByTagName(footnote,XMLString.TEXT_FOOTNOTE_CITATION);
|
||||
if (citation==null) { // try oasis
|
||||
|
@ -1552,7 +1552,7 @@ public class TextConverter extends ConverterHelper {
|
|||
int n = endnotes.size();
|
||||
if (nSplit>0 && n>0) { hnode = converter.nextOutFile(); }
|
||||
for (int i=0; i<n; i++) {
|
||||
Node endnote = (Node) endnotes.get(i);
|
||||
Node endnote = endnotes.get(i);
|
||||
String sId = Misc.getAttribute(endnote,XMLString.TEXT_ID);
|
||||
Node citation = Misc.getChildByTagName(endnote,XMLString.TEXT_ENDNOTE_CITATION);
|
||||
if (citation==null) { // try oasis
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue