pdfviewer: load files with missing xref (we need in order to help with testing, as most good pdfx i…

experimental/PdfViewer/pdfparser/native/SkNativeParsedPDF.cpp

 #include "SkNativeParsedPDF.h"
 #include "SkPdfNativeTokenizer.h"
 #include "SkPdfBasics.h"
 #include "SkPdfObject.h"
 
 #include <stdio.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 
@@ skipping 105 matching lines @@
 
     if (strcmp((char*)xrefstartKeywordLine, "startxref") != 0) {
         // TODO(edisonn): report/issue
     }
 
     long xrefByteOffset = atol((const char*)xrefByteOffsetLine);
 
     bool storeCatalog = true;
     while (xrefByteOffset >= 0) {
         const unsigned char* trailerStart = readCrossReferenceSection(fFileContent + xrefByteOffset, xrefstartKeywordLine);
-        xrefByteOffset = readTrailer(trailerStart, xrefstartKeywordLine, storeCatalog);
+        readTrailer(trailerStart, xrefstartKeywordLine, storeCatalog, &xrefByteOffset, false);
         storeCatalog = false;
     }
 
     // TODO(edisonn): warn/error expect fObjects[fRefCatalogId].fGeneration == fRefCatalogGeneration
     // TODO(edisonn): security, verify that SkPdfCatalogDictionary is indeed using mapper
     // load catalog
 
     if (fRootCatalogRef) {
         fRootCatalog = (SkPdfCatalogDictionary*)resolveReference(fRootCatalogRef);
         if (fRootCatalog->isDictionary() && fRootCatalog->valid()) {
             SkPdfPageTreeNodeDictionary* tree = fRootCatalog->Pages(this);
             if (tree && tree->isDictionary() && tree->valid()) {
                 fillPages(tree);
             }
         }
     }
 
+    // TODO(edisonn): clean up this doc, or better, let the caller call again and build a new doc
+    // caller should be a static function.
+    if (pages() == 0) {
+        loadWithoutXRef();
+    }
+
     // TODO(edisonn): corrupted pdf, read it from beginning and rebuild (xref, trailer, or just reall all objects)
     // 0 pages
 
     // now actually read all objects if we want, or do it lazyly
     // and resolve references?... or not ...
 }
 
+void SkNativeParsedPDF::loadWithoutXRef() {
+    const unsigned char* current = fFileContent;
+    const unsigned char* end = fFileContent + fContentLength;
+
+    // TODO(edisonn): read pdf version
+    current = ignoreLine(current, end);
+
+    current = skipPdfWhiteSpaces(0, current, end);
+    while (current < end) {
+        SkPdfObject token;
+        current = nextObject(0, current, end, &token, NULL, NULL);
+        if (token.isInteger()) {
+            int id = (int)token.intValue();
+
+            token.reset();
+            current = nextObject(0, current, end, &token, NULL, NULL);
+            // int generation = (int)token.intValue();  // TODO(edisonn): ignored for now
+
+            token.reset();
+            current = nextObject(0, current, end, &token, NULL, NULL);
+            // TODO(edisonn): must be obj, return error if not? ignore ?
+            if (!token.isKeyword("obj")) {
+                continue;
+            }
+
+            while (fObjects.count() < id + 1) {
+                reset(fObjects.append());
+            }
+
+            fObjects[id].fOffset = current - fFileContent;
+
+            SkPdfObject* obj = fAllocator->allocObject();
+            current = nextObject(0, current, end, obj, fAllocator, this);
+
+            fObjects[id].fResolvedReference = obj;
+            fObjects[id].fObj = obj;
+
+            // set objects
+        } else if (token.isKeyword("trailer")) {
+            long dummy;
+            current = readTrailer(current, end, true, &dummy, true);
+        } else if (token.isKeyword("startxref")) {
+            token.reset();
+            current = nextObject(0, current, end, &token, NULL, NULL);  // ignore
+        }
+
+        current = skipPdfWhiteSpaces(0, current, end);
+    }
+
+    if (fRootCatalogRef) {
+        fRootCatalog = (SkPdfCatalogDictionary*)resolveReference(fRootCatalogRef);
+        if (fRootCatalog->isDictionary() && fRootCatalog->valid()) {
+            SkPdfPageTreeNodeDictionary* tree = fRootCatalog->Pages(this);
+            if (tree && tree->isDictionary() && tree->valid()) {
+                fillPages(tree);
+            }
+        }
+    }
+
+}
+
 // TODO(edisonn): NYI
 SkNativeParsedPDF::~SkNativeParsedPDF() {
     sk_free((void*)fFileContent);
     delete fAllocator;
 }
 
 const unsigned char* SkNativeParsedPDF::readCrossReferenceSection(const unsigned char* xrefStart, const unsigned char* trailerEnd) {
     const unsigned char* current = ignoreLine(xrefStart, trailerEnd);  // TODO(edisonn): verify next keyord is "xref", use nextObject here
 
     SkPdfObject token;
@@ skipping 40 matching lines @@
                 return current;
             }
 
             addCrossSectionInfo(startId + i, generation, offset, *token.c_str() == 'f');
         }
     }
     // TODO(edisonn): it should never get here? there is no trailer?
     return current;
 }
 
-long SkNativeParsedPDF::readTrailer(const unsigned char* trailerStart, const unsigned char* trailerEnd, bool storeCatalog) {
-    SkPdfObject trailerKeyword;
-    // TODO(edisonn): use null allocator, and let it just fail if memory
-    // needs allocated (but no crash)!
-    const unsigned char* current =
-            nextObject(0, trailerStart, trailerEnd, &trailerKeyword, NULL, NULL);
+const unsigned char* SkNativeParsedPDF::readTrailer(const unsigned char* trailerStart, const unsigned char* trailerEnd, bool storeCatalog, long* prev, bool skipKeyword) {
+    *prev = -1;
 
-    if (!trailerKeyword.isKeyword() || strlen("trailer") != trailerKeyword.lenstr() ||
-        strncmp(trailerKeyword.c_str(), "trailer", strlen("trailer")) != 0) {
-        // TODO(edisonn): report warning, rebuild trailer from objects.
-        return -1;
+    const unsigned char* current = trailerStart;
+    if (!skipKeyword) {
+        SkPdfObject trailerKeyword;
+        // TODO(edisonn): use null allocator, and let it just fail if memory
+        // needs allocated (but no crash)!
+        current = nextObject(0, current, trailerEnd, &trailerKeyword, NULL, NULL);
+
+        if (!trailerKeyword.isKeyword() || strlen("trailer") != trailerKeyword.lenstr() ||
+            strncmp(trailerKeyword.c_str(), "trailer", strlen("trailer")) != 0) {
+            // TODO(edisonn): report warning, rebuild trailer from objects.
+            return current;
+        }
     }
 
     SkPdfObject token;
     current = nextObject(0, current, trailerEnd, &token, fAllocator, NULL);
     if (!token.isDictionary()) {
-        return -1;
+        return current;
     }
     SkPdfFileTrailerDictionary* trailer = (SkPdfFileTrailerDictionary*)&token;
     if (!trailer->valid()) {
-        return -1;
+        return current;
     }
 
     if (storeCatalog) {
         const SkPdfObject* ref = trailer->Root(NULL);
         if (ref == NULL || !ref->isReference()) {
             // TODO(edisonn): oops, we have to fix the corrup pdf file
-            return -1;
+            return current;
         }
         fRootCatalogRef = ref;
     }
 
     if (trailer->has_Prev()) {
-        return (long)trailer->Prev(NULL);
+        *prev = (long)trailer->Prev(NULL);
     }
 
-    return -1;
+    return current;
 }
 
 void SkNativeParsedPDF::addCrossSectionInfo(int id, int generation, int offset, bool isFreed) {
     // TODO(edisonn): security here
     while (fObjects.count() < id + 1) {
         reset(fObjects.append());
     }
 
     fObjects[id].fOffset = offset;
     fObjects[id].fObj = NULL;
+    fObjects[id].fResolvedReference = NULL;
 }
 
 SkPdfObject* SkNativeParsedPDF::readObject(int id/*, int expectedGeneration*/) {
     long startOffset = fObjects[id].fOffset;
     //long endOffset = fObjects[id].fOffsetEnd;
     // TODO(edisonn): use hinted endOffset
     // TODO(edisonn): current implementation will result in a lot of memory usage
     // to decrease memory usage, we wither need to be smart and know where objects end, and we will
     // alocate only the chancks needed, or the tokenizer will not make copies, but then it needs to
     // cache the results so it does not go twice on the same buffer
@@ skipping 192 matching lines @@
     return (SkPdfObject*)ref;
 }
 
 size_t SkNativeParsedPDF::bytesUsed() const {
     return fAllocator->bytesUsed() +
            fContentLength +
            fObjects.count() * sizeof(PublicObjectEntry) +
            fPages.count() * sizeof(SkPdfPageObjectDictionary*) +
            sizeof(*this);
 }