Apache Tika源码研究（七）

tika怎样加载Parser实现类的，怎样根据文档的mime类型调用相应的Parser实现类,本文接着分析

先熟悉一下tika的解析类的相关接口和类的UML模型：

Parser接口的源码如下：

/**
 * Tika parser interface.
 */
public interface Parser extends Serializable {

    /**
     * Returns the set of media types supported by this parser when used
     * with the given parse context.
     *
     * @since Apache Tika 0.7
     * @param context parse context
     * @return immutable set of media types
     */
    Set<MediaType> getSupportedTypes(ParseContext context);

    /**
     * Parses a document stream into a sequence of XHTML SAX events.
     * Fills in related document metadata in the given metadata object.
     * <p>
     * The given document stream is consumed but not closed by this method.
     * The responsibility to close the stream remains on the caller.
     * <p>
     * Information about the parsing context can be passed in the context
     * parameter. See the parser implementations for the kinds of context
     * information they expect.
     *
     * @since Apache Tika 0.5
     * @param stream the document stream (input)
     * @param handler handler for the XHTML SAX events (output)
     * @param metadata document metadata (input and output)
     * @param context parse context
     * @throws IOException if the document stream could not be read
     * @throws SAXException if the SAX events could not be processed
     * @throws TikaException if the document could not be parsed
     */
    void parse(
            InputStream stream, ContentHandler handler,
            Metadata metadata, ParseContext context)
            throws IOException, SAXException, TikaException;

}

该接口只提供void parse(InputStream stream, ContentHandler handler,Metadata metadata, ParseContext context)方法供其他类调用

抽象类AbstractParser实现了Parser接口，其源码如下：

/**
 * Abstract base class for new parsers. This method implements the old
 * deprecated parse method so subclasses won't have to.
 *
 * @since Apache Tika 0.10
 */
public abstract class AbstractParser implements Parser {

    /**
     * Serial version UID.
     */
    private static final long serialVersionUID = 7186985395903074255L;

    /**
     * Calls the
     * {@link Parser#parse(InputStream, ContentHandler, Metadata, ParseContext)}
     * method with an empty {@link ParseContext}. This method exists as a
     * leftover from Tika 0.x when the three-argument parse() method still
     * existed in the {@link Parser} interface. No new code should call this
     * method anymore, it's only here for backwards compatibility.
     *
     * @deprecated use the {@link Parser#parse(InputStream, ContentHandler, Metadata, ParseContext)} method instead
     */
    public void parse(
            InputStream stream, ContentHandler handler, Metadata metadata)
            throws IOException, SAXException, TikaException {
        parse(stream, handler, metadata, new ParseContext());
    }

}

新增void parse(InputStream stream, ContentHandler handler, Metadata metadata)方法，提供模板方法功能

下面接着贴出CompositeParser的源码，它继承自抽象类AbstractParser

/**
 * Composite parser that delegates parsing tasks to a component parser
 * based on the declared content type of the incoming document. A fallback
 * parser is defined for cases where a parser for the given content type is
 * not available.
 */
public class CompositeParser extends AbstractParser {

    /** Serial version UID */
    private static final long serialVersionUID = 2192845797749627824L;

    /**
     * Media type registry.
     */
    private MediaTypeRegistry registry;

    /**
     * List of component parsers.
     */
    private List<Parser> parsers;

    /**
     * The fallback parser, used when no better parser is available.
     */
    private Parser fallback = new EmptyParser();

    public CompositeParser(MediaTypeRegistry registry, List<Parser> parsers) {
        this.parsers = parsers;
        this.registry = registry;
    }

    public CompositeParser(MediaTypeRegistry registry, Parser... parsers) {
        this(registry, Arrays.asList(parsers));
    }

    public CompositeParser() {
        this(new MediaTypeRegistry());
    }

    public Map<MediaType, Parser> getParsers(ParseContext context) {
        Map<MediaType, Parser> map = new HashMap<MediaType, Parser>();
        for (Parser parser : parsers) {
            for (MediaType type : parser.getSupportedTypes(context)) {
                map.put(registry.normalize(type), parser);
            }
        }
        return map;
    }

    /**
     * Utility method that goes through all the component parsers and finds
     * all media types for which more than one parser declares support. This
     * is useful in tracking down conflicting parser definitions.
     *
     * @since Apache Tika 0.10
     * @see <a href="https://issues.apache.org/jira/browse/TIKA-660">TIKA-660</a>
     * @param context parsing context
     * @return media types that are supported by at least two component parsers
     */
    public Map<MediaType, List<Parser>> findDuplicateParsers(
            ParseContext context) {
        Map<MediaType, Parser> types = new HashMap<MediaType, Parser>();
        Map<MediaType, List<Parser>> duplicates =
            new HashMap<MediaType, List<Parser>>();
        for (Parser parser : parsers) {
            for (MediaType type : parser.getSupportedTypes(context)) {
                MediaType canonicalType = registry.normalize(type);
                if (types.containsKey(canonicalType)) {
                    List<Parser> list = duplicates.get(canonicalType);
                    if (list == null) {
                        list = new ArrayList<Parser>();
                        list.add(types.get(canonicalType));
                        duplicates.put(canonicalType, list);
                    }
                    list.add(parser);
                } else {
                    types.put(canonicalType, parser);
                }
            }
        }
        return duplicates;
    }

    /**
     * Returns the media type registry used to infer type relationships.
     *
     * @since Apache Tika 0.8
     * @return media type registry
     */
    public MediaTypeRegistry getMediaTypeRegistry() {
        return registry;
    }

    /**
     * Sets the media type registry used to infer type relationships.
     *
     * @since Apache Tika 0.8
     * @param registry media type registry
     */
    public void setMediaTypeRegistry(MediaTypeRegistry registry) {
        this.registry = registry;
    }

    /**
     * Returns the component parsers.
     *
     * @return component parsers, keyed by media type
     */
    public Map<MediaType, Parser> getParsers() {
        return getParsers(new ParseContext());
    }

    /**
     * Sets the component parsers.
     *
     * @param parsers component parsers, keyed by media type
     */
    public void setParsers(Map<MediaType, Parser> parsers) {
        this.parsers = new ArrayList<Parser>(parsers.size());
        for (Map.Entry<MediaType, Parser> entry : parsers.entrySet()) {
            this.parsers.add(ParserDecorator.withTypes(
                    entry.getValue(), Collections.singleton(entry.getKey())));
        }
    }

    /**
     * Returns the fallback parser.
     *
     * @return fallback parser
     */
    public Parser getFallback() {
        return fallback;
    }

    /**
     * Sets the fallback parser.
     *
     * @param fallback fallback parser
     */
    public void setFallback(Parser fallback) {
        this.fallback = fallback;
    }

    /**
     * Returns the parser that best matches the given metadata. By default
     * looks for a parser that matches the content type metadata property,
     * and uses the fallback parser if a better match is not found. The
     * type hierarchy information included in the configured media type
     * registry is used when looking for a matching parser instance.
     * <p>
     * Subclasses can override this method to provide more accurate
     * parser resolution.
     *
     * @param metadata document metadata
     * @return matching parser
     */
    protected Parser getParser(Metadata metadata) {
        return getParser(metadata, new ParseContext());
    }

    protected Parser getParser(Metadata metadata, ParseContext context) {
        Map<MediaType, Parser> map = getParsers(context);
        MediaType type = MediaType.parse(metadata.get(Metadata.CONTENT_TYPE));
        if (type != null) {
           // We always work on the normalised, canonical form
           type = registry.normalize(type);
        }
        
        while (type != null) {
            // Try finding a parser for the type
            Parser parser = map.get(type);
            if (parser != null) {
                return parser;
            }
            
            // Failing that, try for the parent of the type
            type = registry.getSupertype(type);
        }
        return fallback;
    }

    public Set<MediaType> getSupportedTypes(ParseContext context) {
        return getParsers(context).keySet();
    }

    /**
     * Delegates the call to the matching component parser.
     * <p>
     * Potential {@link RuntimeException}s, {@link IOException}s and
     * {@link SAXException}s unrelated to the given input stream and content
     * handler are automatically wrapped into {@link TikaException}s to better
     * honor the {@link Parser} contract.
     */
    public void parse(
            InputStream stream, ContentHandler handler,
            Metadata metadata, ParseContext context)
            throws IOException, SAXException, TikaException {
        Parser parser = getParser(metadata);
        TemporaryResources tmp = new TemporaryResources();
        try {
            TikaInputStream taggedStream = TikaInputStream.get(stream, tmp);
            TaggedContentHandler taggedHandler = new TaggedContentHandler(handler);
            try {
                parser.parse(taggedStream, taggedHandler, metadata, context);
            } catch (RuntimeException e) {
                throw new TikaException(
                        "Unexpected RuntimeException from " + parser, e);
            } catch (IOException e) {
                taggedStream.throwIfCauseOf(e);
                throw new TikaException(
                        "TIKA-198: Illegal IOException from " + parser, e);
            } catch (SAXException e) {
                taggedHandler.throwIfCauseOf(e);
                throw new TikaException(
                        "TIKA-237: Illegal SAXException from " + parser, e);
            }
        } finally {
            tmp.dispose();
        }
    }

}

 该类的注释很清楚，相当于将解析任务委托给了其他的解析组件，而自身提供的parser方法供其他类调用

且分析CompositeParser类是怎样将解析任务委托给其他解析组件的，关键是parser方法的这行代码 Parser parser = getParser(metadata);

它调用了下面的方法：

    protected Parser getParser(Metadata metadata) {
        return getParser(metadata, new ParseContext());
    }

    protected Parser getParser(Metadata metadata, ParseContext context) {
        Map<MediaType, Parser> map = getParsers(context);
        MediaType type = MediaType.parse(metadata.get(Metadata.CONTENT_TYPE));
        if (type != null) {
           // We always work on the normalised, canonical form
           type = registry.normalize(type);
        }
        
        while (type != null) {
            // Try finding a parser for the type
            Parser parser = map.get(type);
            if (parser != null) {
                return parser;
            }
            
            // Failing that, try for the parent of the type
            type = registry.getSupertype(type);
        }
        return fallback;
    }

执行流程是首先获取mime类型跟相应的Parser实现类的映射Map<MediaType, Parser> ，然后根据Metadata的Metadata.CONTENT_TYPE属性得到MediaType类型，最后从Map<MediaType, Parser>获取相应的Parser实现类

 上面的代码Map<MediaType, Parser> map = getParsers(context)是获取Map<MediaType, Parser>映射

public Map<MediaType, Parser> getParsers(ParseContext context) {
        Map<MediaType, Parser> map = new HashMap<MediaType, Parser>();
        for (Parser parser : parsers) {
            for (MediaType type : parser.getSupportedTypes(context)) {
                map.put(registry.normalize(type), parser);
            }
        }
        return map;
    }

即根据构造方法初始化的List<Parser> parsers组件集合，这里注意的是如果该组件类集合中的成员之一为CompositeParser本身的类型，则该成员提供的可以支持的mime类型同时又来自于该成员的解析组件集合（这里也许是CompositeParser命名的原因，这里用到了Composite模式），我们可以看到它Set<MediaType> getSupportedTypes(ParseContext context)方法：

    public Set<MediaType> getSupportedTypes(ParseContext context) {
        return getParsers(context).keySet();
    }

 Composite模式的简要UML模型图如下：

我们接下来分析DefaultParser的源码，该类继承自CompositeParser类，用于初始化CompositeParser类的相关成员变量

/**
 * A composite parser based on all the {@link Parser} implementations
 * available through the
 * {@link javax.imageio.spi.ServiceRegistry service provider mechanism}.
 *
 * @since Apache Tika 0.8
 */
public class DefaultParser extends CompositeParser {

    /** Serial version UID */
    private static final long serialVersionUID = 3612324825403757520L;

    /**
     * Finds all statically loadable parsers and sort the list by name,
     * rather than discovery order. CompositeParser takes the last
     * parser for any given media type, so put the Tika parsers first
     * so that non-Tika (user supplied) parsers can take precedence.
     *
     * @param loader service loader
     * @return ordered list of statically loadable parsers
     */
    private static List<Parser> getDefaultParsers(ServiceLoader loader) {
        List<Parser> parsers =
                loader.loadStaticServiceProviders(Parser.class);
        Collections.sort(parsers, new Comparator<Parser>() {
            public int compare(Parser p1, Parser p2) {
                String n1 = p1.getClass().getName();
                String n2 = p2.getClass().getName();
                boolean t1 = n1.startsWith("org.apache.tika.");
                boolean t2 = n2.startsWith("org.apache.tika.");
                if (t1 == t2) {
                    return n1.compareTo(n2);
                } else if (t1) {
                    return -1;
                } else {
                    return 1;
                }
            }
        });
        return parsers;
    }

    private transient final ServiceLoader loader;

    public DefaultParser(MediaTypeRegistry registry, ServiceLoader loader) {
        super(registry, getDefaultParsers(loader));
        this.loader = loader;
    }

    public DefaultParser(MediaTypeRegistry registry, ClassLoader loader) {
        this(registry, new ServiceLoader(loader));
    }

    public DefaultParser(ClassLoader loader) {
        this(MediaTypeRegistry.getDefaultRegistry(), new ServiceLoader(loader));
    }

    public DefaultParser(MediaTypeRegistry registry) {
        this(registry, new ServiceLoader());
    }

    public DefaultParser() {
        this(MediaTypeRegistry.getDefaultRegistry());
    }

    @Override
    public Map<MediaType, Parser> getParsers(ParseContext context) {
        Map<MediaType, Parser> map = super.getParsers(context);

        if (loader != null) {
            // Add dynamic parser service (they always override static ones)
            MediaTypeRegistry registry = getMediaTypeRegistry();
            for (Parser parser
                    : loader.loadDynamicServiceProviders(Parser.class)) {
                for (MediaType type : parser.getSupportedTypes(context)) {
                    map.put(registry.normalize(type), parser);
                }
            }
        }

        return map;
    }

}

该类主要是为基类初始化MediaTypeRegistry registry成员与List<Parser> parsers成员，它本身并没有覆盖void parse(InputStream stream, ContentHandler handler,Metadata metadata, ParseContext context)方法，为的是执行基类的方法（根据mime类型执行具体parser类的parse方法）

最后来分析AutoDetectParser类的源码，它也继承自CompositeParser类：

public class AutoDetectParser extends CompositeParser {

    /** Serial version UID */
    private static final long serialVersionUID = 6110455808615143122L;

    /**
     * The type detector used by this parser to auto-detect the type
     * of a document.
     */
    private Detector detector; // always set in the constructor

    /**
     * Creates an auto-detecting parser instance using the default Tika
     * configuration.
     */
    public AutoDetectParser() {
        this(TikaConfig.getDefaultConfig());
    }

    public AutoDetectParser(Detector detector) {
        this(TikaConfig.getDefaultConfig());
        setDetector(detector);
    }

    /**
     * Creates an auto-detecting parser instance using the specified set of parser.
     * This allows one to create a Tika configuration where only a subset of the
     * available parsers have their 3rd party jars included, as otherwise the
     * use of the default TikaConfig will throw various "ClassNotFound" exceptions.
     * 
     * @param detector Detector to use
     * @param parsers
     */
    public AutoDetectParser(Parser...parsers) {
        this(new DefaultDetector(), parsers);
    }

    public AutoDetectParser(Detector detector, Parser...parsers) {
        super(MediaTypeRegistry.getDefaultRegistry(), parsers);
        setDetector(detector);
    }

    public AutoDetectParser(TikaConfig config) {
        super(config.getMediaTypeRegistry(), config.getParser());
        setDetector(config.getDetector());
    }

    /**
     * Returns the type detector used by this parser to auto-detect the type
     * of a document.
     *
     * @return type detector
     * @since Apache Tika 0.4
     */
    public Detector getDetector() {
        return detector;
    }

    /**
     * Sets the type detector used by this parser to auto-detect the type
     * of a document.
     *
     * @param detector type detector
     * @since Apache Tika 0.4
     */
    public void setDetector(Detector detector) {
        this.detector = detector;
    }

    public void parse(
            InputStream stream, ContentHandler handler,
            Metadata metadata, ParseContext context)
            throws IOException, SAXException, TikaException {
        TemporaryResources tmp = new TemporaryResources();
        try {
            TikaInputStream tis = TikaInputStream.get(stream, tmp);

            // Automatically detect the MIME type of the document
            MediaType type = detector.detect(tis, metadata);
            metadata.set(Metadata.CONTENT_TYPE, type.toString());

            // TIKA-216: Zip bomb prevention
            SecureContentHandler sch = new SecureContentHandler(handler, tis);
            try {
                // Parse the document
                super.parse(tis, sch, metadata, context);
            } catch (SAXException e) {
                // Convert zip bomb exceptions to TikaExceptions
                sch.throwIfCauseOf(e);
                throw e;
            }
        } finally {
            tmp.dispose();
        }
    }

    public void parse(
            InputStream stream, ContentHandler handler, Metadata metadata)
            throws IOException, SAXException, TikaException {
        ParseContext context = new ParseContext();
        context.set(Parser.class, this);
        parse(stream, handler, metadata, context);
    }

}

 该类也初始化基类的MediaTypeRegistry registry成员与List<Parser> parsers成员，不过这里的List<Parser> parsers成员有TikaConfig类提供，后者默认提供的Parser实现类为DefaultParser

它的void parse(InputStream stream, ContentHandler handler,Metadata metadata, ParseContext context)方法首先检测文档的mime类型，然后将解析处理委托给CompositeParser基类执行，自身对外提供接口

所以整个解析的流程是

一、AutoDetectParser的parse方法：首先检测文件的mime类型，然后将解析任务交给基类CompositeParser的parse方法

二、AutoDetectParser的基类CompositeParser的parse方法：根据参数里面的mime类型获取解析类DefaultParser（支持所有已经注册的mime类型，由List<Parser> parsers成员提供）

三、调用DefaultParser的parse方法（DefaultParser默认执行父类的parse方法），即基类CompositeParser的parse方法，根据参数里面的mime类型获取具体解析类

四、最后执行具体解析类的parse方法

这里第一次AutoDetectParser初始化基类CompositeParser的parser组件集合是DefaultParser，基类CompositeParser的parse方法委托给DefaultParser

第二次DefaultParser初始化基类的CompositeParser的parser组件集合是具体的parser实现类集合，基类CompositeParser的parse方法委托给具体的parser实现类

这里体现的是Composite模式的运用。