SkPDF: Optionally output PDF/A-2b archive format.

src/doc/SkDocument_PDF.cpp

 /*
  * Copyright 2011 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkDocument.h"
 #include "SkPDFCanon.h"
 #include "SkPDFDevice.h"
 #include "SkPDFFont.h"
 #include "SkPDFStream.h"
 #include "SkPDFTypes.h"
 #include "SkPDFUtils.h"
 #include "SkStream.h"
+#include "SkPDFMetadata.h"
 
 class SkPDFDict;
 
 static void emit_pdf_header(SkWStream* stream) {
     stream->writeText("%PDF-1.4\n%");
     // The PDF spec recommends including a comment with four bytes, all
     // with their high bits set.  This is "Skia" with the high bits set.
     stream->write32(0xD3EBE9E1);
     stream->writeText("\n");
 }
 
 static void emit_pdf_footer(SkWStream* stream,
                             const SkPDFObjNumMap& objNumMap,
                             const SkPDFSubstituteMap& substitutes,
                             SkPDFObject* docCatalog,
                             int64_t objCount,
                             int32_t xRefFileOffset,
-                            SkPDFDict* info) {
+                            SkPDFObject* info /* take ownership */,
+                            SkPDFObject* id /* take ownership */) {
     SkPDFDict trailerDict;
-    // TODO(vandebo): Linearized format will take a Prev entry too.
-    // TODO(vandebo): PDF/A requires an ID entry.
+    // TODO(http://crbug.com/80908): Linearized format will take a
+    //                               Prev entry too.
     trailerDict.insertInt("Size", int(objCount));
     trailerDict.insertObjRef("Root", SkRef(docCatalog));
-    SkASSERT(info);
-    trailerDict.insertObjRef("Info", SkRef(info));
-
+    if (info) {
+        trailerDict.insertObjRef("Info", info);

[tomhudson, 2015/10/09 15:31:37]

You made info optional?

[hal.canary, 2015/10/09 19:13:27]

Done.

+    }
+    if (id) {
+        trailerDict.insertObject("ID", id);

[tomhudson, 2015/10/09 15:31:37]

Both info and id are noted "take ownership", but for one you insert a ref and
the other you insert the object. Consistent?
(Confession: I haven't got your memory model internalized right now, although
you've explained it very well in the past. No mention at
https://skia.org/dev/design/pdftheory?)

[hal.canary, 2015/10/09 19:13:27]

a ref just means "don't inline this object"

+    }
     stream->writeText("trailer\n");
     trailerDict.emitObject(stream, objNumMap, substitutes);
     stream->writeText("\nstartxref\n");
     stream->writeBigDecAsText(xRefFileOffset);
     stream->writeText("\n%%EOF");
 }
 
 static void perform_font_subsetting(
         const SkTDArray<const SkPDFDevice*>& pageDevices,
         SkPDFSubstituteMap* substituteMap) {
@@ skipping 103 matching lines @@
         nextRoundNodes.rewind();
         treeCapacity *= kNodeSize;
     } while (curNodes.count() > 1);
 
     pageTree->push(curNodes[0]);  // Transfer reference.
     if (rootNode) {
         *rootNode = curNodes[0];
     }
 }
 
-struct Metadata {
-    SkTArray<SkDocument::Attribute> fInfo;
-    SkAutoTDelete<const SkTime::DateTime> fCreation;
-    SkAutoTDelete<const SkTime::DateTime> fModified;
-};
-
-static SkString pdf_date(const SkTime::DateTime& dt) {
-    int timeZoneMinutes = SkToInt(dt.fTimeZoneMinutes);
-    char timezoneSign = timeZoneMinutes >= 0 ? '+' : '-';
-    int timeZoneHours = SkTAbs(timeZoneMinutes) / 60;
-    timeZoneMinutes = SkTAbs(timeZoneMinutes) % 60;
-    return SkStringPrintf(
-            "D:%04u%02u%02u%02u%02u%02u%c%02d'%02d'",
-            static_cast<unsigned>(dt.fYear), static_cast<unsigned>(dt.fMonth),
-            static_cast<unsigned>(dt.fDay), static_cast<unsigned>(dt.fHour),
-            static_cast<unsigned>(dt.fMinute),
-            static_cast<unsigned>(dt.fSecond), timezoneSign, timeZoneHours,
-            timeZoneMinutes);
-}
-
-SkPDFDict* create_document_information_dict(const Metadata& metadata) {
-    SkAutoTUnref<SkPDFDict> dict(new SkPDFDict);
-    static const char* keys[] = {
-        "Title", "Author", "Subject", "Keywords", "Creator" };
-    for (const char* key : keys) {
-        for (const SkDocument::Attribute& keyValue : metadata.fInfo) {
-            if (keyValue.fKey.equals(key)) {
-                dict->insertString(key, keyValue.fValue);
-            }
-        }
-    }
-    dict->insertString("Producer", "Skia/PDF");
-    if (metadata.fCreation) {
-        dict->insertString("CreationDate", pdf_date(*metadata.fCreation.get()));
-    }
-    if (metadata.fModified) {
-        dict->insertString("ModDate", pdf_date(*metadata.fModified.get()));
-    }
-    return dict.detach();
-}
-
 static bool emit_pdf_document(const SkTDArray<const SkPDFDevice*>& pageDevices,
-                              const Metadata& metadata,
+                              const SkPDFMetadata& metadata,
                               SkWStream* stream) {
     if (pageDevices.isEmpty()) {
         return false;
     }
 
     SkTDArray<SkPDFDict*> pages;
     SkAutoTUnref<SkPDFDict> dests(new SkPDFDict);
 
     for (int i = 0; i < pageDevices.count(); i++) {
         SkASSERT(pageDevices[i]);
         SkASSERT(i == 0 ||
                  pageDevices[i - 1]->getCanon() == pageDevices[i]->getCanon());
         SkAutoTUnref<SkPDFDict> page(create_pdf_page(pageDevices[i]));
         pageDevices[i]->appendDestinations(dests, page.get());
         pages.push(page.detach());
     }
 
-    SkTDArray<SkPDFDict*> pageTree;
     SkAutoTUnref<SkPDFDict> docCatalog(new SkPDFDict("Catalog"));
 
+    SkAutoTUnref<SkPDFObject> infoDict(
+            metadata.createDocumentInformationDict());
+
+    SkAutoTUnref<SkPDFObject> id, xmp;
+    #ifdef SK_PDF_GENERATE_PDFA
+    SkPDFMetadata::UUID uuid = metadata.uuid();
+    id.reset(SkPDFMetadata::CreatePdfId(uuid, uuid));

[tomhudson, 2015/10/09 15:31:37]

CreatePdfId(uuid, uuid) calls the first one "doc" and the second one "instance".
If you're passing the same UUID for both parameters, it's really not clear how
they're being used.

[hal.canary, 2015/10/09 15:49:11]

    // We use the same UUID for Document ID and Instance ID since this
    // is the first revision of this document (and Skia does not
    // support revising existing PDF documents).

[tomhudson, 2015/10/09 19:21:37]

Acknowledged.

+    xmp.reset(metadata.createXMPObject(uuid, uuid));
+    docCatalog->insertObjRef("Metadata", xmp.detach());
+
+    SkAutoTUnref<SkPDFDict> outputIntent(new SkPDFDict("OutputIntent"));
+    outputIntent->insertName("S", "GTS_PDFA1");
+    outputIntent->insertString("RegistryName", "http://www.color.org");
+    outputIntent->insertString("OutputConditionIdentifier",
+                               "sRGB IEC61966-2.1");

[tomhudson, 2015/10/09 15:31:37]

I thought we don't actually support sRGB yet (and Mike doesn't want to invest
headcount in it in the short-term?)

[tomhudson, 2015/10/09 15:31:37]

Creating these intents even when we aren't writing a PDFA? Because it's so small
it doesn't much matter?

[hal.canary, 2015/10/09 15:49:11]

HTML, CSS, and SVG all specify that they use sRGB.  So the colors that come in
from those sources are already sRGB.  We just don't *blend* in a non-linear way.

[hal.canary, 2015/10/09 15:49:11]

1. I want to save 139 bytes.
2. Nobody cares outside of the PDF/A request.

[tomhudson, 2015/10/09 19:21:37]

Never mind, my question was wrong - this is inside an #ifdef.

HOWEVER, the #ifdef/#endif aren't 0-aligned.

+    SkAutoTUnref<SkPDFArray> intentArray(new SkPDFArray);
+    intentArray->appendObject(outputIntent.detach());
+    docCatalog->insertObject("OutputIntents", intentArray.detach());
+    #endif
+
+    SkTDArray<SkPDFDict*> pageTree;
     SkPDFDict* pageTreeRoot;
     generate_page_tree(pages, &pageTree, &pageTreeRoot);
     docCatalog->insertObjRef("Pages", SkRef(pageTreeRoot));
 
     if (dests->size() > 0) {
         docCatalog->insertObjRef("Dests", dests.detach());
     }
 
-    /* TODO(vandebo): output intent
-    SkAutoTUnref<SkPDFDict> outputIntent = new SkPDFDict("OutputIntent");
-    outputIntent->insertName("S", "GTS_PDFA1");
-    outputIntent->insertString("OutputConditionIdentifier", "sRGB");
-    SkAutoTUnref<SkPDFArray> intentArray(new SkPDFArray);
-    intentArray->appendObject(SkRef(outputIntent.get()));
-    docCatalog->insertObject("OutputIntent", intentArray.detach());
-    */
-
     // Build font subsetting info before proceeding.
     SkPDFSubstituteMap substitutes;
     perform_font_subsetting(pageDevices, &substitutes);
 
-    SkAutoTUnref<SkPDFDict> infoDict(
-            create_document_information_dict(metadata));
     SkPDFObjNumMap objNumMap;
-    if (objNumMap.addObject(infoDict)) {
-        infoDict->addResources(&objNumMap, substitutes);
-    }
-    if (objNumMap.addObject(docCatalog.get())) {
-        docCatalog->addResources(&objNumMap, substitutes);
-    }
+    objNumMap.addObjectRecursively(infoDict, substitutes);
+    objNumMap.addObjectRecursively(docCatalog.get(), substitutes);
     size_t baseOffset = stream->bytesWritten();
     emit_pdf_header(stream);
     SkTDArray<int32_t> offsets;
     for (int i = 0; i < objNumMap.objects().count(); ++i) {
         SkPDFObject* object = objNumMap.objects()[i];
         size_t offset = stream->bytesWritten();
         // This assert checks that size(pdf_header) > 0 and that
         // the output stream correctly reports bytesWritten().
         SkASSERT(offset > baseOffset);
         offsets.push(SkToS32(offset - baseOffset));
@@ skipping 11 matching lines @@
 
     stream->writeText("xref\n0 ");
     stream->writeDecAsText(objCount);
     stream->writeText("\n0000000000 65535 f \n");
     for (int i = 0; i < offsets.count(); i++) {
         SkASSERT(offsets[i] > 0);
         stream->writeBigDecAsText(offsets[i], 10);
         stream->writeText(" 00000 n \n");
     }
     emit_pdf_footer(stream, objNumMap, substitutes, docCatalog.get(), objCount,
-                    xRefFileOffset, infoDict);
+                    xRefFileOffset, infoDict.detach(), id.detach());
 
     // The page tree has both child and parent pointers, so it creates a
     // reference cycle.  We must clear that cycle to properly reclaim memory.
     for (int i = 0; i < pageTree.count(); i++) {
         pageTree[i]->clear();
     }
     pageTree.safeUnrefAll();
     pages.unrefAll();
     return true;
 }
@@ skipping 97 matching lines @@
         fMetadata.fInfo = info;
         fMetadata.fCreation.reset(clone(creationDate));
         fMetadata.fModified.reset(clone(modifiedDate));
     }
 
 private:
     SkPDFCanon fCanon;
     SkTDArray<const SkPDFDevice*> fPageDevices;
     SkAutoTUnref<SkCanvas> fCanvas;
     SkScalar fRasterDpi;
-    Metadata fMetadata;
+    SkPDFMetadata fMetadata;
 };
 }  // namespace
 ///////////////////////////////////////////////////////////////////////////////
 
 SkDocument* SkDocument::CreatePDF(SkWStream* stream, SkScalar dpi) {
     return stream ? new SkDocument_PDF(stream, nullptr, dpi) : nullptr;
 }
 
 SkDocument* SkDocument::CreatePDF(const char path[], SkScalar dpi) {
     SkFILEWStream* stream = new SkFILEWStream(path);
     if (!stream->isValid()) {
         delete stream;
         return nullptr;
     }
     auto delete_wstream = [](SkWStream* stream, bool) { delete stream; };
     return new SkDocument_PDF(stream, delete_wstream, dpi);
 }