| /* | |
| * Copyright 2010 The Android Open Source Project | |
| * | |
| * Use of this source code is governed by a BSD-style license that can be | |
| * found in the LICENSE file. | |
| */ | |
| #include "SkData.h" | |
| #include "SkPDFCatalog.h" | |
| #include "SkPDFDevice.h" | |
| #include "SkPDFPage.h" | |
| #include "SkPDFResourceDict.h" | |
| SkPDFPage::SkPDFPage(SkPDFDevice* content) | |
| : SkPDFDict("Page"), | |
| fDevice(content) { | |
| SkSafeRef(content); | |
| } | |
| SkPDFPage::~SkPDFPage() {} | |
| void SkPDFPage::finalizePage(SkPDFCatalog* catalog, bool firstPage, | |
| const SkTSet<SkPDFObject*>& knownResourceObjects, | |
| SkTSet<SkPDFObject*>* newResourceObjects) { | |
| SkPDFResourceDict* resourceDict = fDevice->getResourceDict(); | |
| if (fContentStream.get() == NULL) { | |
| insert("Resources", resourceDict); | |
| SkSafeUnref(this->insert("MediaBox", fDevice->copyMediaBox())); | |
| if (!SkToBool(catalog->getDocumentFlags() & | |
| SkPDFDocument::kNoLinks_Flags)) { | |
| SkPDFArray* annots = fDevice->getAnnotations(); | |
| if (annots && annots->size() > 0) { | |
| insert("Annots", annots); | |
| } | |
| } | |
| SkAutoTUnref<SkData> content(fDevice->copyContentToData()); | |
| fContentStream.reset(new SkPDFStream(content.get())); | |
| insert("Contents", new SkPDFObjRef(fContentStream.get()))->unref(); | |
| } | |
| catalog->addObject(fContentStream.get(), firstPage); | |
| resourceDict->getReferencedResources(knownResourceObjects, | |
| newResourceObjects, | |
| true); | |
| } | |
| off_t SkPDFPage::getPageSize(SkPDFCatalog* catalog, off_t fileOffset) { | |
| SkASSERT(fContentStream.get() != NULL); | |
| catalog->setFileOffset(fContentStream.get(), fileOffset); | |
| return fContentStream->getOutputSize(catalog, true); | |
| } | |
| void SkPDFPage::emitPage(SkWStream* stream, SkPDFCatalog* catalog) { | |
| SkASSERT(fContentStream.get() != NULL); | |
| fContentStream->emitObject(stream, catalog, true); | |
| } | |
| // static | |
| void SkPDFPage::GeneratePageTree(const SkTDArray<SkPDFPage*>& pages, | |
| SkPDFCatalog* catalog, | |
| SkTDArray<SkPDFDict*>* pageTree, | |
| SkPDFDict** rootNode) { | |
| // PDF wants a tree describing all the pages in the document. We arbitrary | |
| // choose 8 (kNodeSize) as the number of allowed children. The internal | |
| // nodes have type "Pages" with an array of children, a parent pointer, and | |
| // the number of leaves below the node as "Count." The leaves are passed | |
| // into the method, have type "Page" and need a parent pointer. This method | |
| // builds the tree bottom up, skipping internal nodes that would have only | |
| // one child. | |
| static const int kNodeSize = 8; | |
| SkAutoTUnref<SkPDFName> kidsName(new SkPDFName("Kids")); | |
| SkAutoTUnref<SkPDFName> countName(new SkPDFName("Count")); | |
| SkAutoTUnref<SkPDFName> parentName(new SkPDFName("Parent")); | |
| // curNodes takes a reference to its items, which it passes to pageTree. | |
| SkTDArray<SkPDFDict*> curNodes; | |
| curNodes.setReserve(pages.count()); | |
| for (int i = 0; i < pages.count(); i++) { | |
| SkSafeRef(pages[i]); | |
| curNodes.push(pages[i]); | |
| } | |
| // nextRoundNodes passes its references to nodes on to curNodes. | |
| SkTDArray<SkPDFDict*> nextRoundNodes; | |
| nextRoundNodes.setReserve((pages.count() + kNodeSize - 1)/kNodeSize); | |
| int treeCapacity = kNodeSize; | |
| do { | |
| for (int i = 0; i < curNodes.count(); ) { | |
| if (i > 0 && i + 1 == curNodes.count()) { | |
| nextRoundNodes.push(curNodes[i]); | |
| break; | |
| } | |
| SkPDFDict* newNode = new SkPDFDict("Pages"); | |
| SkAutoTUnref<SkPDFObjRef> newNodeRef(new SkPDFObjRef(newNode)); | |
| SkAutoTUnref<SkPDFArray> kids(new SkPDFArray); | |
| kids->reserve(kNodeSize); | |
| int count = 0; | |
| for (; i < curNodes.count() && count < kNodeSize; i++, count++) { | |
| curNodes[i]->insert(parentName.get(), newNodeRef.get()); | |
| kids->append(new SkPDFObjRef(curNodes[i]))->unref(); | |
| // TODO(vandebo): put the objects in strict access order. | |
| // Probably doesn't matter because they are so small. | |
| if (curNodes[i] != pages[0]) { | |
| pageTree->push(curNodes[i]); // Transfer reference. | |
| catalog->addObject(curNodes[i], false); | |
| } else { | |
| SkSafeUnref(curNodes[i]); | |
| catalog->addObject(curNodes[i], true); | |
| } | |
| } | |
| // treeCapacity is the number of leaf nodes possible for the | |
| // current set of subtrees being generated. (i.e. 8, 64, 512, ...). | |
| // It is hard to count the number of leaf nodes in the current | |
| // subtree. However, by construction, we know that unless it's the | |
| // last subtree for the current depth, the leaf count will be | |
| // treeCapacity, otherwise it's what ever is left over after | |
| // consuming treeCapacity chunks. | |
| int pageCount = treeCapacity; | |
| if (i == curNodes.count()) { | |
| pageCount = ((pages.count() - 1) % treeCapacity) + 1; | |
| } | |
| newNode->insert(countName.get(), new SkPDFInt(pageCount))->unref(); | |
| newNode->insert(kidsName.get(), kids.get()); | |
| nextRoundNodes.push(newNode); // Transfer reference. | |
| } | |
| curNodes = nextRoundNodes; | |
| nextRoundNodes.rewind(); | |
| treeCapacity *= kNodeSize; | |
| } while (curNodes.count() > 1); | |
| pageTree->push(curNodes[0]); // Transfer reference. | |
| catalog->addObject(curNodes[0], false); | |
| if (rootNode) { | |
| *rootNode = curNodes[0]; | |
| } | |
| } | |
| const SkTDArray<SkPDFFont*>& SkPDFPage::getFontResources() const { | |
| return fDevice->getFontResources(); | |
| } | |
| const SkPDFGlyphSetMap& SkPDFPage::getFontGlyphUsage() const { | |
| return fDevice->getFontGlyphUsage(); | |
| } | |
| void SkPDFPage::appendDestinations(SkPDFDict* dict) { | |
| fDevice->appendDestinations(dict, this); | |
| } |