TIKA-4744 - sweep based on copilot feedback

Author: tballison

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-microsoft-module/src/main/java/org/apache/tika/parser/microsoft/ooxml/OOXMLTikaBodyPartHandler.java

@@ -386,14 +386,21 @@ private void inlineNoteContent(byte[] xml, String cssClass) throws SAXException
         // from the footnote/endnote parts (needed for picture resolution)
         Map<String, String> noteRelationships = inlinePartMap.getLinkedRelationships();
         xhtml.startElement("div", "class", cssClass);
+        // Track the inner handler so we can call its closeAnyPending() if
+        // the inline-note parseSAX aborts mid-element. Without the drain
+        // the surrounding </div> mismatches whatever the inner handler
+        // left on the SAX stack (<p>/<td>/etc.) and StrictXHTMLValidator
+        // propagates a misleading error.
+        OOXMLTikaBodyPartHandler innerHandler = new OOXMLTikaBodyPartHandler(xhtml);
         try {
             XMLReaderUtils.parseSAX(new ByteArrayInputStream(xml),
                     new EmbeddedContentHandler(
                             new OOXMLWordAndPowerPointTextHandler(
-                                    new OOXMLTikaBodyPartHandler(xhtml),
+                                    innerHandler,
                                     noteRelationships)),
                     parseContext);
-        } catch (TikaException | IOException e) {
+        } catch (TikaException | IOException | SAXException e) {
+            innerHandler.closeAnyPending();
             xhtml.characters("[" + cssClass + " parse error]");
         }
         xhtml.endElement("div");

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-microsoft-module/src/main/java/org/apache/tika/parser/microsoft/ooxml/XSSFExcelExtractorDecorator.java

@@ -239,6 +239,13 @@ protected void buildXHTML(XHTMLContentHandler xhtml)
                     // the </tbody></table></div> emitted below land in the
                     // right place.
                     sheetExtractor.closeAnyPending();
+                } catch (IOException e) {
+                    // Truncated stream — same risk: partial <tr>/<td> still
+                    // open. Close them so the surrounding </tbody></table>
+                    // stays balanced, record the failure, and keep going.
+                    metadata.add(TikaCoreProperties.TIKA_META_EXCEPTION_WARNING,
+                            ExceptionUtils.getStackTrace(e));
+                    sheetExtractor.closeAnyPending();
                 }
                 try {
                     getThreadedComments(container, sheetPart, xhtml);

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-microsoft-module/src/main/java/org/apache/tika/parser/microsoft/rtf/RTFParser.java

@@ -104,7 +104,11 @@ public void parse(TikaInputStream tis, ContentHandler handler, Metadata metadata
             tagged.throwIfCauseOf(e);
             balancer.drainOpenElements();
             throw new TikaException("Error parsing an RTF document", e);
-        } catch (SAXException e) {
+        } catch (SAXException | TikaException e) {
+            // Drain on any exception escaping parseInline. parseInline can
+            // throw TikaException too (memory limits, embedded extraction,
+            // etc.); without the drain, the finally's xhtml.endDocument()
+            // throws on the unbalanced stack and masks the real error.
             balancer.drainOpenElements();
             throw e;
         } finally {

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-microsoft-module/src/main/java/org/apache/tika/parser/microsoft/rtf/TextExtractor.java

@@ -1501,6 +1501,10 @@ private void processGroupEnd() throws IOException, SAXException, TikaException {
                 embObjHandler.handleCompletedObject();
             } catch (TikaException | IOException e) {
                 EmbeddedDocumentUtil.recordException(e, metadata);
+            } catch (SecurityException e) {
+                // Security-relevant exceptions must always propagate
+                // immediately -- never swallow them as a warning.
+                throw e;
             } catch (RuntimeException e) {
                 // POI dispatch on a zero-byte embedded object throws
                 // EmptyFileException; other malformed embedded payloads

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-microsoft-module/src/main/java/org/apache/tika/parser/microsoft/xml/AbstractXML2003Parser.java

@@ -111,6 +111,12 @@ public void parse(TikaInputStream tis, ContentHandler handler, Metadata metadata
             // Close anything the aborted parse left open, then propagate.
             balancer.drainOpenElements();
             throw new TikaException("XML parse error", e);
+        } catch (IOException e) {
+            // Truncated streams etc. -- same problem as the SAX path: without
+            // a drain, the finally's xhtml.endDocument() throws on the
+            // unbalanced stack and masks the real IO error.
+            balancer.drainOpenElements();
+            throw e;
         } finally {
             tis.removeCloseShield();
             xhtml.endDocument();

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-miscoffice-module/src/main/java/org/apache/tika/parser/epub/EpubParser.java

@@ -248,6 +248,11 @@ private Set<String> bufferedParseZipFile(ZipFile zipFile, ContentHandler bodyHan
                             normalizer.drainOpenElements();
                         } catch (IOException ioe) {
                             LOG.trace("epub spine read IOException on {}: {}", path, ioe.toString());
+                            // Same risk as the SAX path: the partial parse may
+                            // have left elements open. Drain before rethrow so
+                            // subsequent spine items and the outer </body>
+                            // don't land on a corrupted stack.
+                            normalizer.drainOpenElements();
                             throw ioe;
                         } finally {
                             processed.add(id);
@@ -351,6 +356,9 @@ private Set<String> fallbackParseAllHtmlEntries(ZipFile zipFile,
             } catch (IOException e) {
                 failed++;
                 LOG.trace("epub fallback: IO failure on {}: {}", entry.getName(), e.toString());
+                // Same drain need as the SAX path: a partial parse before the
+                // IO failure may have left elements open for the next iter.
+                normalizer.drainOpenElements();
             }
         }
         LOG.trace("epub fallback summary: parsed={} failed={}", parsed, failed);

tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-miscoffice-module/src/main/java/org/apache/tika/parser/odf/FlatOpenDocumentParser.java

@@ -119,7 +119,11 @@ public void parse(TikaInputStream tis, ContentHandler handler, Metadata metadata
             if (detected != null) {
                 metadata.set(Metadata.CONTENT_TYPE, detected.toString());
             }
-        } catch (SAXException e) {
+        } catch (SAXException | IOException | TikaException e) {
+            // Drain the balancer on ANY exception that escapes the parse
+            // loop. Without this, the finally's xhtml.endDocument() throws
+            // on the unbalanced stack and the masked </body> vs <p>/<b>
+            // hides the real failure (truncated input, malformed XML, etc.).
             balancer.drainOpenElements();
             throw e;
         } finally {
@@ -228,9 +232,13 @@ public void startElement(String namespaceURI, String localName, String qName,
                 }
             } else if (outerSectionDepth > 0
                     && (META.equals(localName) || BODY.equals(localName)
-                        || SCRIPTS.equals(localName))) {
+                        || (extractMacros && SCRIPTS.equals(localName)))) {
                 // Track nested occurrences so the matching end doesn't
                 // prematurely flip currentHandler back to defaultHandler.
+                // SCRIPTS is gated on extractMacros to match the decrement
+                // condition in endElement -- otherwise a <office:scripts>
+                // inside an outer section would inflate the counter without
+                // a matching decrement, stranding currentHandler.
                 outerSectionDepth++;
             }
             currentHandler.startElement(namespaceURI, localName, qName, attrs);