All Downloads are FREE. Search and download functionalities are using the official Maven repository.

com.thaiopensource.xml.dtd.parse.Entity Maven / Gradle / Ivy

The newest version!
package com.thaiopensource.xml.dtd.parse;

import java.util.Vector;

import com.thaiopensource.xml.tok.Tokenizer;
import com.thaiopensource.xml.em.ExternalId;
import com.thaiopensource.xml.dtd.om.*;

class Entity {
  static class Reference {
    Reference(Entity entity, int start, int end) {
      this.entity = entity;
      this.start = start;
      this.end = end;
    }
    final Entity entity;
    final int start;
    final int end;
  }

  final String name;
  final boolean isParameter;
  Entity(String name, boolean isParameter) {
    this.name = name;
    this.isParameter = isParameter;
  }
  char[] text;
  String entityValue;
  String systemId;
  String publicId;
  String baseUri;
  String encoding;
  String uri;
  Entity overrides;
  boolean overridden;
  // Which parts of text came from references?
  Reference[] references;
  boolean open;
  String notationName;
  Vector atoms;
  boolean mustReparse;
  static final int NO_PROBLEM = 0;
  static final int INCONSISTENT_LEVEL_PROBLEM = 1;
  static final int INCONSISTENT_PARSE_PROBLEM = 2;
  static final int UNEXPAND_PROBLEM = 3;
  static final int UNKNOWN_SEMANTIC_PROBLEM = 4;
  static final int EMPTY_PARTICLE_PROBLEM = 5;
  static final int REPARSE_PROBLEM = 6;
  static final int UNREFERENCED_PROBLEM = 7;
  static final int INTERNAL_DECL_PROBLEM = 8;
  static final int NOT_PARSED_PROBLEM = 9;

  int problem = UNREFERENCED_PROBLEM;

  static final int INCONSISTENT_LEVEL = -1;
  static final int NO_LEVEL = 0;
  static final int DECL_LEVEL = 1;
  static final int PARAM_LEVEL = 2;
  static final int PARTICLE_LEVEL = 3;

  int referenceLevel = NO_LEVEL;

  static final int GROUP_CONTAINS_OR = 01;
  static final int GROUP_CONTAINS_SEQ = 02;
  static final int GROUP_CONTAINS_PCDATA = 04;
  static final int GROUP_CONTAINS_GROUP = 010;
  static final int GROUP_CONTAINS_ELEMENT_NAME = 020;
  static final int GROUP_CONTAINS_NMTOKEN = 040;

  int groupFlags = 0;

  static final int SEMANTIC_NONE = 0;
  static final int SEMANTIC_MODEL_GROUP = 1;
  static final int SEMANTIC_ATTRIBUTE_GROUP = 2;
  static final int SEMANTIC_ENUM_GROUP = 3;
  static final int SEMANTIC_DATATYPE = 4;
  static final int SEMANTIC_FLAG = 5;
  static final int SEMANTIC_NAME_SPEC = 6;
  static final int SEMANTIC_ATTRIBUTE_DEFAULT = 7;

  int semantic = SEMANTIC_NONE;

  boolean maybeNameSpec;

  ModelGroup modelGroup;
  AttributeGroup attributeGroup;
  EnumGroup enumGroup;
  Datatype datatype;
  Flag flag;
  NameSpec nameSpec;
  AttributeDefault attributeDefault;

  Decl decl;

  Vector parsed;

  void inconsistentParse() {
    parsed = null;
    referenceLevel = INCONSISTENT_LEVEL;
    problem = INCONSISTENT_PARSE_PROBLEM;
  }

  void setParsed(int level, Vector v, int start, int end) {
    if (referenceLevel < 0)
      return;
    if (level == referenceLevel) {
      if (!sliceEqual(parsed, v, start, end))
	inconsistentParse();
      return;
    }
    if (referenceLevel == NO_LEVEL) {
      parsed = new Vector();
      appendSlice(parsed, v, start, end);
      referenceLevel = level;
      return;
    }
    if (parsed.size() == end - start) {
      if (level == PARAM_LEVEL && referenceLevel == PARTICLE_LEVEL) {
	if (paramsParticlesConsistent(v, start, parsed, 0, end - start)) {
	  // For element name case, otherwise particle will be
	  // ambiguous with model group.
	  referenceLevel = PARAM_LEVEL;
	  parsed = new Vector();
	  appendSlice(parsed, v, start, end);
	  return;
	}
      }
      else if (level == PARTICLE_LEVEL && referenceLevel == PARAM_LEVEL) {
	if (paramsParticlesConsistent(parsed, 0, v, start, end - start))
	  return;
      }
    }
    problem = INCONSISTENT_LEVEL_PROBLEM;
    parsed = null;
    referenceLevel = INCONSISTENT_LEVEL;
  }

  static boolean paramsParticlesConsistent(Vector params, int i,
					   Vector particles, int j,
					   int n) {
    for (int k = 0; k < n; k++)
      if (!paramParticleConsistent((Param)params.elementAt(i + k),
				   (Particle)particles.elementAt(j + k)))
	return false;
    return true;
  }

  static boolean paramParticleConsistent(Param param, Particle particle) {
    switch (param.type) {
    case Param.MODEL_GROUP:
      return param.group.equals(particle);
    case Param.ELEMENT_NAME:
      return particle.type == Particle.ELEMENT_NAME;
    case Param.REFERENCE:
      return particle.type == Particle.REFERENCE;
    case Param.REFERENCE_END:
      return particle.type == Particle.REFERENCE_END;
    }
    return false;
  }

  int textIndexToAtomIndexOccur(int ti) {
    if (ti == text.length)
      return -1;
    switch (text[ti]) {
    case '*':
    case '?':
    case '+':
      break;
    default:
      return -1;
    }
    return textIndexToAtomIndex(ti + 1);
  }

  int textIndexToAtomIndex(int ti) {
    int nAtoms = atoms.size();
    int len = 0;
    int atomIndex = 0;      
    for (;;) {
      if (len == ti)
	return atomIndex;
      if (atomIndex >= nAtoms)
	break;
      Atom a = (Atom)atoms.elementAt(atomIndex);
      len += a.getToken().length();
      if (len > ti)
	break;
      atomIndex++;
    }
    return -1;
  }

  void unexpandEntities() {
    if (references == null || atoms == null)
      return;
    Vector newAtoms = null;
    int nCopiedAtoms = 0;
    for (int i = 0; i < references.length; i++) {
      int start = textIndexToAtomIndex(references[i].start);
      int end = textIndexToAtomIndex(references[i].end);
      if (start >= 0
	  && end >= 0
	  && !(start == end && atomMaybePasted(start))
	  && atomsAreProperlyNested(start, end, true)) {
	if (newAtoms == null)
	  newAtoms = new Vector();
	appendSlice(newAtoms, atoms, nCopiedAtoms, start);
	newAtoms.addElement(new Atom(references[i].entity));
	if (references[i].entity.atoms == null) {
	  Vector tem = new Vector();
	  references[i].entity.atoms = tem;
	  appendSlice(tem, atoms, start, end);
	  references[i].entity.unexpandEntities();
	}
	nCopiedAtoms = end;
      }
      else if (start >= 0
	       && (end = textIndexToAtomIndexOccur(references[i].end)) >= 0
	       && atomsAreProperlyNested(start, end, false)) {
	// This deals with a case like %foo;* by turning it into (%foo;)*.
	if (newAtoms == null)
	  newAtoms = new Vector();
	Atom[] split = splitAtom((Atom)atoms.elementAt(end - 1));
	appendSlice(newAtoms, atoms, nCopiedAtoms, start);
	newAtoms.addElement(new Atom(Tokenizer.TOK_OPEN_PAREN, "("));
	newAtoms.addElement(new Atom(references[i].entity));
	newAtoms.addElement(split[1]);
	if (references[i].entity.atoms == null) {
	  Vector tem = new Vector();
	  references[i].entity.atoms = tem;
	  appendSlice(tem, atoms, start, end - 1);
	  tem.addElement(split[0]);
	  references[i].entity.unexpandEntities();
	}
	nCopiedAtoms = end;
      }
      else if (!overridden)
	references[i].entity.problem = UNEXPAND_PROBLEM;
    }
    if (newAtoms == null)
      return;
    appendSlice(newAtoms, atoms, nCopiedAtoms, atoms.size());
    atoms = newAtoms;
    references = null;
    if (overrides != null)
      overrides.unexpandEntities();
  }
  
  private static Atom[] splitAtom(Atom atom) {
    Atom[] split = new Atom[2];
    switch (atom.getTokenType()) {
    case Tokenizer.TOK_NAME_QUESTION:
      split[1] = new Atom(Tokenizer.TOK_CLOSE_PAREN_QUESTION, ")?");
      break;
    case Tokenizer.TOK_NAME_ASTERISK:
      split[1] = new Atom(Tokenizer.TOK_CLOSE_PAREN_ASTERISK, ")*");
      break;
    case Tokenizer.TOK_NAME_PLUS:
      split[1] = new Atom(Tokenizer.TOK_CLOSE_PAREN_PLUS, ")+");
      break;
    case Tokenizer.TOK_CLOSE_PAREN_QUESTION:
    case Tokenizer.TOK_CLOSE_PAREN_ASTERISK:
    case Tokenizer.TOK_CLOSE_PAREN_PLUS:
      split[0] = new Atom(Tokenizer.TOK_CLOSE_PAREN, ")");
      split[1] = atom;
      return split;
    }
    split[0] = new Atom(Tokenizer.TOK_NAME,
			atom.getToken().substring(0,
						  atom.getToken().length() - 1));
    return split;
  }
	
  private boolean atomsAreProperlyNested(int start,
					 int end,
					 boolean allowConnectors) {
    int level = 0;
    for (int i = start; i < end; i++)
      switch (((Atom)atoms.elementAt(i)).getTokenType()) {
      case Tokenizer.TOK_COND_SECT_OPEN:
      case Tokenizer.TOK_OPEN_PAREN:
      case Tokenizer.TOK_OPEN_BRACKET:
      case Tokenizer.TOK_DECL_OPEN:
	level++;
	break;
      case Tokenizer.TOK_CLOSE_PAREN:
      case Tokenizer.TOK_CLOSE_PAREN_ASTERISK:
      case Tokenizer.TOK_CLOSE_PAREN_QUESTION:
      case Tokenizer.TOK_CLOSE_PAREN_PLUS:
      case Tokenizer.TOK_CLOSE_BRACKET:
      case Tokenizer.TOK_DECL_CLOSE:
	if (--level < 0)
	  return false;
	break;
      case Tokenizer.TOK_COND_SECT_CLOSE:
	if ((level -= 2) < 0)
	  return false;
	break;
      case Tokenizer.TOK_OR:
      case Tokenizer.TOK_COMMA:
	if (!allowConnectors && level == 0)
	  return false;
	break;
      }
    return level == 0;
  }

  private boolean atomMaybePasted(int i) {
    if (i > 0) {
      switch (((Atom)atoms.elementAt(i - 1)).getTokenType()) {
      case Tokenizer.TOK_NAME:
      case Tokenizer.TOK_PREFIXED_NAME:
      case Tokenizer.TOK_NMTOKEN:
	return true;
      }
    }
    if (i < atoms.size()) {
      switch (((Atom)atoms.elementAt(i)).getTokenType()) {
      case Tokenizer.TOK_NAME:
      case Tokenizer.TOK_PREFIXED_NAME:
      case Tokenizer.TOK_NAME_QUESTION:
      case Tokenizer.TOK_NAME_ASTERISK:
      case Tokenizer.TOK_NAME_PLUS:
      case Tokenizer.TOK_NMTOKEN:
	return true;
      }
    }
    return false;
  }

  static boolean sliceEqual(Vector v1, Vector v2, int start, int end) {
    int n = v1.size();
    if (end - start != n)
      return false;
    for (int i = 0; i < n; i++)
      if (!v1.elementAt(i).equals(v2.elementAt(start + i)))
	return false;
    return true;
  }

  static void appendSlice(Vector to, Vector from, int start, int end) {
    for (; start < end; start++)
      to.addElement(from.elementAt(start));
  }

  void analyzeSemantic() {
    if (problem != NO_PROBLEM)
      return;
    switch (referenceLevel) {
    case PARAM_LEVEL:
      analyzeSemanticParam();
      break;
    case PARTICLE_LEVEL:
      analyzeSemanticParticle();
      break;
    case DECL_LEVEL:
      problem = INTERNAL_DECL_PROBLEM;
      break;
    case NO_LEVEL:
      problem = NOT_PARSED_PROBLEM;
      break;
    }
    for (Entity e = overrides; e != null; e = e.overrides)
      e.analyzeSemanticOverride(this);
  }

  private void analyzeSemanticOverride(Entity orig) {
    if (parsed == null || problem != NO_PROBLEM)
      return;
    switch (referenceLevel) {
    case PARAM_LEVEL:
      analyzeSemanticParam();
      break;
    case PARTICLE_LEVEL:
      groupFlags = orig.groupFlags;
      analyzeSemanticParticle();
      break;
    }
    if (semantic != orig.semantic)
      semantic = SEMANTIC_NONE;
  }

  private void analyzeSemanticParam() {
    if (isAttributeGroup())
      semantic = SEMANTIC_ATTRIBUTE_GROUP;
    else if (isDatatype())
      semantic = SEMANTIC_DATATYPE;
    else if (isFlag())
      semantic = SEMANTIC_FLAG;
    else if (isModelGroup())
      semantic = SEMANTIC_MODEL_GROUP;
    else if (isNameSpec())
      semantic = SEMANTIC_NAME_SPEC;
    else if (isAttributeDefault())
      semantic = SEMANTIC_ATTRIBUTE_DEFAULT;
    else if (isEnumGroup())
      semantic = SEMANTIC_ENUM_GROUP;
    else
      problem = UNKNOWN_SEMANTIC_PROBLEM;
  }

  private boolean isAttributeGroup() {
    ParamStream ps = new ParamStream(parsed);
    if (!ps.advance())
      return false;
    do {
      if (ps.type != Param.EMPTY_ATTRIBUTE_GROUP
	  && (ps.type != Param.ATTRIBUTE_NAME
	      || !ps.advance()
	      || (ps.type == Param.ATTRIBUTE_TYPE_NOTATION && !ps.advance())
	      || !ps.advance()
	      || (ps.type == Param.FIXED && !ps.advance())))
	return false;
    } while (ps.advance());
    return true;
  }

  private boolean isDatatype() {
    ParamStream ps = new ParamStream(parsed);
    return (ps.advance()
	    && (ps.type == Param.ATTRIBUTE_TYPE
		|| ps.type == Param.ATTRIBUTE_VALUE_GROUP
		|| (ps.type == Param.ATTRIBUTE_TYPE_NOTATION
		    && ps.advance()))
	    && !ps.advance());
  }

  private boolean isAttributeDefault() {
    ParamStream ps = new ParamStream(parsed);
    return (ps.advance()
	    && (ps.type == Param.DEFAULT_ATTRIBUTE_VALUE
		|| ps.type == Param.IMPLIED
		|| ps.type == Param.REQUIRED
		|| (ps.type == Param.FIXED && ps.advance()))
	    && !ps.advance());
  }

  private boolean isFlag() {
    ParamStream ps = new ParamStream(parsed);
    return (ps.advance()
	    && (ps.type == Param.INCLUDE
		|| ps.type == Param.IGNORE)
	    && !ps.advance());
  }

  private boolean isModelGroup() {
    ParamStream ps = new ParamStream(parsed);
    return (ps.advance()
	    && (ps.type == Param.MODEL_GROUP
		|| ps.type == Param.EMPTY
		|| ps.type == Param.EMPTY)
	    && !ps.advance());
  }

  private boolean isNameSpec() {
    ParamStream ps = new ParamStream(parsed);
    return (ps.advance()
	    && (ps.type == Param.ELEMENT_NAME
		|| ps.type == Param.ATTRIBUTE_NAME)
	    && !ps.advance());
  }

  private boolean isEnumGroup() {
    ParamStream ps = new ParamStream(parsed);
    return (ps.advance() && ps.type == Param.NOTATION_GROUP && !ps.advance());
  }

  private void analyzeSemanticParticle() {
    if (maybeNameSpec) {
      semantic = SEMANTIC_NAME_SPEC;
      return;
    }
    int n = parsed.size();
    if (n == 0) {
      analyzeEmptySemanticParticle();
      return;
    }
    for (int i = 0; i < n; i++) {
      switch (((Particle)parsed.elementAt(i)).type) {
      case Particle.GROUP:
      case Particle.ELEMENT_NAME:
      case Particle.PCDATA:
	semantic = SEMANTIC_MODEL_GROUP;
	return;
      case Particle.NMTOKEN:
	semantic = SEMANTIC_ENUM_GROUP;
	return;
      }
    }
    problem = UNKNOWN_SEMANTIC_PROBLEM;
  }

  static final int GROUP_MODEL_GROUP_FLAGS 
    = GROUP_CONTAINS_PCDATA|GROUP_CONTAINS_GROUP|GROUP_CONTAINS_ELEMENT_NAME;

  private void analyzeEmptySemanticParticle() {
    if ((groupFlags & GROUP_MODEL_GROUP_FLAGS) == 0) {
      semantic = SEMANTIC_ENUM_GROUP;
      return;
    }
    if ((groupFlags & GROUP_CONTAINS_NMTOKEN) == 0) {
      switch (groupFlags & (GROUP_CONTAINS_SEQ|GROUP_CONTAINS_OR)) {
      case 0:
	groupFlags |= GROUP_CONTAINS_OR;
	// fall through
      case GROUP_CONTAINS_SEQ:
      case GROUP_CONTAINS_OR:
	semantic = SEMANTIC_MODEL_GROUP;
	return;
      }
    }
    problem = EMPTY_PARTICLE_PROBLEM;
  }

  ModelGroup toModelGroup() {
    if (referenceLevel == PARAM_LEVEL)
      return Param.paramsToModelGroup(parsed);
    if (parsed.size() == 0) {
      if ((groupFlags & GROUP_CONTAINS_SEQ) != 0)
	return new Sequence(new ModelGroup[0]);
      else
	return new Choice(new ModelGroup[0]);
    }
    return Particle.particlesToModelGroup(parsed);
  }

  EnumGroup toEnumGroup() {
    if (referenceLevel == PARTICLE_LEVEL)
      return Particle.particlesToEnumGroup(parsed);
    else
      return Particle.particlesToEnumGroup(((Param)parsed.elementAt(0)).group.particles);
  }

  NameSpec toNameSpec() {
    if (referenceLevel == PARTICLE_LEVEL)
      return new Name(((Particle)parsed.elementAt(0)).value);
    else
      return Param.paramsToNameSpec(parsed);
  }

  ExternalId getExternalId() {
    return new ExternalId(systemId, publicId, baseUri);
  }

  void noteReferenced() {
    if (problem == UNREFERENCED_PROBLEM)
      problem = NO_PROBLEM;
    if (overrides != null)
      overrides.noteReferenced();
  }

  String ambiguousNameSpec() {
    if (problem != NO_PROBLEM
	|| referenceLevel != PARTICLE_LEVEL
	|| parsed.size() != 1)
      return null;
    Particle p = (Particle)parsed.elementAt(0);
    if (p.type != Particle.ELEMENT_NAME)
      return null;
    maybeNameSpec = true;
    return p.value;
  }
}





© 2015 - 2024 Weber Informatics LLC | Privacy Policy