extractorhtml.html

一个开源的网页爬虫一个开源的网页爬虫一个开源的网页爬虫一个开源的网页爬虫一个开源的网页爬虫一个开源的网页爬虫

<a name="406" href="#406">406</a> <em>     * </em>
<a name="407" href="#407">407</a> <em>     * @param curi</em>
<a name="408" href="#408">408</a> <em>     * @param value</em>
<a name="409" href="#409">409</a> <em>     * @param context</em>
<a name="410" href="#410">410</a> <em>     */</em>
<a name="411" href="#411">411</a>     <strong>protected</strong> <strong>void</strong> processLink(<a href="../../../../org/archive/crawler/datamodel/CrawlURI.html">CrawlURI</a> curi, <strong>final</strong> CharSequence value,
<a name="412" href="#412">412</a>             CharSequence context) {
<a name="413" href="#413">413</a>         <strong>if</strong> (TextUtils.matches(JAVASCRIPT, value)) {
<a name="414" href="#414">414</a>             processScriptCode(curi, value. subSequence(11, value.length()));
<a name="415" href="#415">415</a>         } <strong>else</strong> {    
<a name="416" href="#416">416</a>             <strong>if</strong> (logger.isLoggable(Level.FINEST)) {
<a name="417" href="#417">417</a>                 logger.finest(<span class="string">"link: "</span> + value.toString() + <span class="string">" from "</span> + curi);
<a name="418" href="#418">418</a>             }
<a name="419" href="#419">419</a>             addLinkFromString(curi,
<a name="420" href="#420">420</a>                 (value instanceof String)?
<a name="421" href="#421">421</a>                     (String)value: value.toString(),
<a name="422" href="#422">422</a>                 context, Link.NAVLINK_HOP);
<a name="423" href="#423">423</a>             <strong>this</strong>.numberOfLinksExtracted++;
<a name="424" href="#424">424</a>         }
<a name="425" href="#425">425</a>     }
<a name="426" href="#426">426</a> 
<a name="427" href="#427">427</a>     <strong>private</strong> <strong>void</strong> addLinkFromString(<a href="../../../../org/archive/crawler/datamodel/CrawlURI.html">CrawlURI</a> curi, String uri,
<a name="428" href="#428">428</a>             CharSequence context, <strong>char</strong> hopType) {
<a name="429" href="#429">429</a>         <strong>try</strong> {
<a name="430" href="#430">430</a>             <em class="comment">// We do a 'toString' on context because its a sequence from</em>
<a name="431" href="#431">431</a>             <em class="comment">// the underlying ReplayCharSequence and the link its about</em>
<a name="432" href="#432">432</a>             <em class="comment">// to become a part of is expected to outlive the current</em>
<a name="433" href="#433">433</a>             <em class="comment">// ReplayCharSequence.</em>
<a name="434" href="#434">434</a>             curi.createAndAddLinkRelativeToBase(uri, context.toString(),
<a name="435" href="#435">435</a>                 hopType);
<a name="436" href="#436">436</a>         } <strong>catch</strong> (URIException e) {
<a name="437" href="#437">437</a>             <strong>if</strong> (getController() != <strong>null</strong>) {
<a name="438" href="#438">438</a>                 getController().logUriError(e, curi.getUURI(), uri);
<a name="439" href="#439">439</a>             } <strong>else</strong> {
<a name="440" href="#440">440</a>                 logger.info(<span class="string">"Failed createAndAddLinkRelativeToBase "</span> +
<a name="441" href="#441">441</a>                     curi + <span class="string">", "</span> + uri + <span class="string">", "</span> + context + <span class="string">", "</span> + hopType +
<a name="442" href="#442">442</a>                     <span class="string">": "</span> + e);
<a name="443" href="#443">443</a>             }
<a name="444" href="#444">444</a>         }
<a name="445" href="#445">445</a>     }
<a name="446" href="#446">446</a> 
<a name="447" href="#447">447</a>     <strong>protected</strong> <strong>final</strong> <strong>void</strong> processEmbed(<a href="../../../../org/archive/crawler/datamodel/CrawlURI.html">CrawlURI</a> curi, CharSequence value,
<a name="448" href="#448">448</a>             CharSequence context) {
<a name="449" href="#449">449</a>         processEmbed(curi, value, context, Link.EMBED_HOP);
<a name="450" href="#450">450</a>     }
<a name="451" href="#451">451</a> 
<a name="452" href="#452">452</a>     <strong>protected</strong> <strong>void</strong> processEmbed(<a href="../../../../org/archive/crawler/datamodel/CrawlURI.html">CrawlURI</a> curi, <strong>final</strong> CharSequence value,
<a name="453" href="#453">453</a>             CharSequence context, <strong>char</strong> hopType) {
<a name="454" href="#454">454</a>         <strong>if</strong> (logger.isLoggable(Level.FINEST)) {
<a name="455" href="#455">455</a>             logger.finest(<span class="string">"embed ("</span> + hopType + <span class="string">"): "</span> + value.toString() +
<a name="456" href="#456">456</a>                 <span class="string">" from "</span> + curi);
<a name="457" href="#457">457</a>         }
<a name="458" href="#458">458</a>         addLinkFromString(curi,
<a name="459" href="#459">459</a>             (value instanceof String)?
<a name="460" href="#460">460</a>                 (String)value: value.toString(),
<a name="461" href="#461">461</a>             context, hopType);
<a name="462" href="#462">462</a>         <strong>this</strong>.numberOfLinksExtracted++;
<a name="463" href="#463">463</a>     }
<a name="464" href="#464">464</a> 
<a name="465" href="#465">465</a>     <strong>public</strong> <strong>void</strong> extract(<a href="../../../../org/archive/crawler/datamodel/CrawlURI.html">CrawlURI</a> curi) {
<a name="466" href="#466">466</a>         <strong>if</strong> (!isHttpTransactionContentToProcess(curi) ||
<a name="467" href="#467">467</a>                 ! (isExpectedMimeType(curi.getContentType(), <span class="string">"text/html"</span>)
<a name="468" href="#468">468</a>                    || isExpectedMimeType(curi.getContentType(), <span class="string">"application/xhtml"</span>))) {
<a name="469" href="#469">469</a>             <strong>return</strong>;
<a name="470" href="#470">470</a>         }
<a name="471" href="#471">471</a> 
<a name="472" href="#472">472</a>         <strong>final</strong> <strong>boolean</strong> ignoreUnexpectedHTML =
<a name="473" href="#473">473</a>              ((Boolean)getUncheckedAttribute(curi, 
<a name="474" href="#474">474</a>                  ATTR_IGNORE_UNEXPECTED_HTML)).booleanValue();        
<a name="475" href="#475">475</a> 
<a name="476" href="#476">476</a>         <strong>if</strong> (ignoreUnexpectedHTML) {
<a name="477" href="#477">477</a>             <strong>try</strong> {
<a name="478" href="#478">478</a>                 <strong>if</strong>(!isHtmlExpectedHere(curi)) {
<a name="479" href="#479">479</a>                     <em class="comment">// HTML was not expected (eg a GIF was expected) so ignore</em>
<a name="480" href="#480">480</a>                     <em class="comment">// (as if a soft 404)</em>
<a name="481" href="#481">481</a>                     <strong>return</strong>;
<a name="482" href="#482">482</a>                 }
<a name="483" href="#483">483</a>             } <strong>catch</strong> (URIException e) {
<a name="484" href="#484">484</a>                 logger.severe(<span class="string">"Failed expectedHTML test: "</span> + e.getMessage());
<a name="485" href="#485">485</a>             }
<a name="486" href="#486">486</a>         }
<a name="487" href="#487">487</a> 
<a name="488" href="#488">488</a>         <strong>this</strong>.numberOfCURIsHandled++;
<a name="489" href="#489">489</a> 
<a name="490" href="#490">490</a>         <a href="../../../../org/archive/io/ReplayCharSequence.html">ReplayCharSequence</a> cs = <strong>null</strong>;
<a name="491" href="#491">491</a>         
<a name="492" href="#492">492</a>         <strong>try</strong> {
<a name="493" href="#493">493</a>            <a href="../../../../org/archive/util/HttpRecorder.html">HttpRecorder</a> hr = curi.getHttpRecorder();
<a name="494" href="#494">494</a>            <strong>if</strong> (hr == <strong>null</strong>) {
<a name="495" href="#495">495</a>                <strong>throw</strong> <strong>new</strong> IOException(<span class="string">"Why is recorder null here?"</span>);
<a name="496" href="#496">496</a>            }
<a name="497" href="#497">497</a>            cs = hr.getReplayCharSequence();
<a name="498" href="#498">498</a>         } <strong>catch</strong> (IOException e) {
<a name="499" href="#499">499</a>             curi.addLocalizedError(<strong>this</strong>.getName(), e,
<a name="500" href="#500">500</a>                 <span class="string">"Failed get of replay char sequence "</span> + curi.toString() +
<a name="501" href="#501">501</a>                     <span class="string">" "</span> + e.getMessage());
<a name="502" href="#502">502</a>             logger.log(Level.SEVERE,<span class="string">"Failed get of replay char sequence in "</span> +
<a name="503" href="#503">503</a>                 Thread.currentThread().getName(), e);
<a name="504" href="#504">504</a>         }
<a name="505" href="#505">505</a>         
<a name="506" href="#506">506</a>         <strong>if</strong> (cs == <strong>null</strong>) {
<a name="507" href="#507">507</a>             <strong>return</strong>;
<a name="508" href="#508">508</a>         }
<a name="509" href="#509">509</a> 
<a name="510" href="#510">510</a>         <em class="comment">// We have a ReplayCharSequence open.  Wrap all in finally so we</em>
<a name="511" href="#511">511</a>         <em class="comment">// for sure close it before we leave.</em>
<a name="512" href="#512">512</a>         <strong>try</strong> {
<a name="513" href="#513">513</a>             <em class="comment">// Extract all links from the charsequence</em>
<a name="514" href="#514">514</a>             extract(curi, cs);
<a name="515" href="#515">515</a>             <em class="comment">// Set flag to indicate that link extraction is completed.</em>
<a name="516" href="#516">516</a>             curi.linkExtractorFinished();
<a name="517" href="#517">517</a>         } <strong>finally</strong> {
<a name="518" href="#518">518</a>             <strong>if</strong> (cs != <strong>null</strong>) {
<a name="519" href="#519">519</a>                 <strong>try</strong> {
<a name="520" href="#520">520</a>                     cs.close();
<a name="521" href="#521">521</a>                 } <strong>catch</strong> (IOException ioe) {
<a name="522" href="#522">522</a>                     logger.warning(TextUtils.exceptionToString(
<a name="523" href="#523">523</a>                         <span class="string">"Failed close of ReplayCharSequence."</span>, ioe));
<a name="524" href="#524">524</a>                 }
<a name="525" href="#525">525</a>             }
<a name="526" href="#526">526</a>         }
<a name="527" href="#527">527</a>     }
<a name="528" href="#528">528</a> 
<a name="529" href="#529">529</a>     <em>/**<em>*</em></em>
<a name="530" href="#530">530</a> <em>     * Run extractor.</em>
<a name="531" href="#531">531</a> <em>     * This method is package visible to ease testing.</em>
<a name="532" href="#532">532</a> <em>     * @param curi CrawlURI we're processing.</em>
<a name="533" href="#533">533</a> <em>     * @param cs Sequence from underlying ReplayCharSequence. This</em>
<a name="534" href="#534">534</a> <em>     * is TRANSIENT data. Make a copy if you want the data to live outside</em>
<a name="535" href="#535">535</a> <em>     * of this extractors' lifetime.</em>
<a name="536" href="#536">536</a> <em>     */</em>
<a name="537" href="#537">537</a>     <strong>void</strong> extract(<a href="../../../../org/archive/crawler/datamodel/CrawlURI.html">CrawlURI</a> curi, CharSequence cs) {
<a name="538" href="#538">538</a>         Matcher tags = TextUtils.getMatcher(RELEVANT_TAG_EXTRACTOR, cs);
<a name="539" href="#539">539</a>         <strong>while</strong>(tags.find()) {
<a name="540" href="#540">540</a>             <strong>if</strong>(Thread.interrupted()){
<a name="541" href="#541">541</a>                 <strong>break</strong>;
<a name="542" href="#542">542</a>             }
<a name="543" href="#543">543</a>             <strong>if</strong> (tags.start(8) > 0) {