package com.ctc.wstx.dtd;
import java.util.StringTokenizer;
import javax.xml.stream.XMLStreamException;
import com.ctc.wstx.ent.EntityDecl;
import com.ctc.wstx.io.WstxInputData;
import com.ctc.wstx.sr.InputProblemReporter;
import com.ctc.wstx.util.PrefixedName;
Specific attribute class for attributes that contain (unique)
identifiers.
/**
* Specific attribute class for attributes that contain (unique)
* identifiers.
*/
public final class DTDEntitiesAttr
extends DTDAttribute
{
/*
///////////////////////////////////////////////////
// Life-cycle
///////////////////////////////////////////////////
*/
Main constructor. Note that id attributes can never have
default values.
/**
* Main constructor. Note that id attributes can never have
* default values.
*/
public DTDEntitiesAttr(PrefixedName name, DefaultAttrValue defValue, int specIndex,
boolean nsAware, boolean xml11)
{
super(name, defValue, specIndex, nsAware, xml11);
}
@Override
public DTDAttribute cloneWith(int specIndex) {
return new DTDEntitiesAttr(mName, mDefValue, specIndex, mCfgNsAware, mCfgXml11);
}
/*
///////////////////////////////////////////////////
// Public API
///////////////////////////////////////////////////
*/
@Override
public int getValueType() {
return TYPE_ENTITIES;
}
/*
///////////////////////////////////////////////////
// Public API, validation
///////////////////////////////////////////////////
*/
Method called by the DTDValidatorBase
to let the attribute do necessary normalization and/or validation for the value. /**
* Method called by the {@link DTDValidatorBase}
* to let the attribute do necessary normalization and/or validation
* for the value.
*
*/
@Override
public String validate(DTDValidatorBase v, char[] cbuf, int start, int end, boolean normalize)
throws XMLStreamException
{
/* Let's skip leading/trailing white space, even if we are not
* to normalize visible attribute value. This allows for better
* round-trip handling (no changes for physical value caller
* gets), but still allows succesful validation.
*/
while (start < end && WstxInputData.isSpaceChar(cbuf[start])) {
++start;
}
// Empty value?
if (start >= end) {
return reportValidationProblem(v, "Empty ENTITIES value");
}
--end; // so that it now points to the last char
while (end > start && WstxInputData.isSpaceChar(cbuf[end])) {
--end;
}
// Ok; now start points to first, last to last char (both inclusive)
String idStr = null;
StringBuilder sb = null;
while (start <= end) {
// Ok, need to check char validity, and also calc hash code:
char c = cbuf[start];
if (!WstxInputData.isNameStartChar(c, mCfgNsAware, mCfgXml11)) {
return reportInvalidChar(v, c, "not valid as the first ENTITIES character");
}
int i = start+1;
for (; i <= end; ++i) {
c = cbuf[i];
if (WstxInputData.isSpaceChar(c)) {
break;
}
if (!WstxInputData.isNameChar(c, mCfgNsAware, mCfgXml11)) {
return reportInvalidChar(v, c, "not valid as an ENTITIES character");
}
}
EntityDecl ent = findEntityDecl(v, cbuf, start, (i - start));
// only returns if entity was found...
// Can skip the trailing space char (if there was one)
start = i+1;
/* When normalizing, we can possibly share id String, or
* alternatively, compose normalized String if multiple
*/
if (normalize) {
if (idStr == null) { // first idref
idStr = ent.getName();
} else {
if (sb == null) {
sb = new StringBuilder(idStr);
}
idStr = ent.getName();
sb.append(' ');
sb.append(idStr);
}
}
// Ok, any white space to skip?
while (start <= end && WstxInputData.isSpaceChar(cbuf[start])) {
++start;
}
}
if (normalize) {
if (sb != null) {
idStr = sb.toString();
}
return idStr;
}
return null;
}
Method called by the validator object
to ask attribute to verify that the default it has (if any) is
valid for such type.
/**
* Method called by the validator object
* to ask attribute to verify that the default it has (if any) is
* valid for such type.
*/
@Override
public void validateDefault(InputProblemReporter rep, boolean normalize)
throws XMLStreamException
{
String normStr = validateDefaultNames(rep, true);
if (normalize) {
mDefValue.setValue(normStr);
}
// Ok, but were they declared?
/* Performance really shouldn't be critical here (only called when
* parsing DTDs, which get cached) -- let's just
* tokenize using standard StringTokenizer
*/
StringTokenizer st = new StringTokenizer(normStr);
/* !!! 03-Dec-2004, TSa: This is rather ugly -- need to know we
* actually really get a DTD reader, and DTD reader needs
* to expose a special method... but it gets things done.
*/
MinimalDTDReader dtdr = (MinimalDTDReader) rep;
while (st.hasMoreTokens()) {
String str = st.nextToken();
EntityDecl ent = dtdr.findEntity(str);
// Needs to exists, and be an unparsed entity...
checkEntity(rep, normStr, ent);
}
}
}