pdf: write only ToUnicode mappings needed by the font, trimming anything out of [firstChar, lastCha…

src/pdf/SkPDFFont.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 <ctype.h>
 
 #include "SkData.h"
@@ skipping 324 matching lines @@
                 appendAdvance(advanceInfo->fAdvance[0], emSize, result);
                 break;
             }
         }
     }
     return result;
 }
 
 }  // namespace
 
-static void append_tounicode_header(SkDynamicMemoryWStream* cmap) {
+static void append_tounicode_header(SkDynamicMemoryWStream* cmap,
+                                    uint16_t firstGlyphID,
+                                    uint16_t lastGlyphID) {
     // 12 dict begin: 12 is an Adobe-suggested value. Shall not change.
     // It's there to prevent old version Adobe Readers from malfunctioning.
     const char* kHeader =
         "/CIDInit /ProcSet findresource begin\n"
         "12 dict begin\n"
         "begincmap\n";
     cmap->writeText(kHeader);
 
     // The /CIDSystemInfo must be consistent to the one in
     // SkPDFFont::populateCIDFont().
     // We can not pass over the system info object here because the format is
     // different. This is not a reference object.
     const char* kSysInfo =
         "/CIDSystemInfo\n"
         "<<  /Registry (Adobe)\n"
         "/Ordering (UCS)\n"
         "/Supplement 0\n"
         ">> def\n";
     cmap->writeText(kSysInfo);
 
     // The CMapName must be consistent to /CIDSystemInfo above.
     // /CMapType 2 means ToUnicode.
-    // We specify codespacerange from 0x0000 to 0xFFFF because we convert our
-    // code table from unsigned short (16-bits). Codespace range just tells the
-    // PDF processor the valid range. It does not matter whether a complete
-    // mapping is provided or not.
-    const char* kTypeInfo =
+    // Codespace range just tells the PDF processor the valid range.
+    const char* kTypeInfoHeader =
         "/CMapName /Adobe-Identity-UCS def\n"
         "/CMapType 2 def\n"
-        "1 begincodespacerange\n"
-        "<0000> <FFFF>\n"
-        "endcodespacerange\n";
-    cmap->writeText(kTypeInfo);
+        "1 begincodespacerange\n";
+    cmap->writeText(kTypeInfoHeader);
+    
+    // e.g.     "<0000> <FFFF>\n"
+    SkString range;
+    range.appendf("<%04X> <%04X>\n", firstGlyphID, lastGlyphID);
+    cmap->writeText(range.c_str());
+    
+    const char* kTypeInfoFooter = "endcodespacerange\n";
+    cmap->writeText(kTypeInfoFooter);
 }
 
 static void append_cmap_footer(SkDynamicMemoryWStream* cmap) {
     const char* kFooter =
         "endcmap\n"
         "CMapName currentdict /CMap defineresource pop\n"
         "end\n"
         "end";
     cmap->writeText(kFooter);
 }
@@ skipping 73 matching lines @@
 // the spec does not mention how will this kind of conflict being resolved.
 //
 // For the worst case (having 65536 continuous unicode and we use every other
 // one of them), the possible savings by aggressive optimization is 416KB
 // pre-compressed and does not provide enough motivation for implementation.
 
 // FIXME: this should be in a header so that it is separately testable
 // ( see caller in tests/ToUnicode.cpp )
 void append_cmap_sections(const SkTDArray<SkUnichar>& glyphToUnicode,
                           const SkPDFGlyphSet* subset,
-                          SkDynamicMemoryWStream* cmap);
+                          SkDynamicMemoryWStream* cmap,
+                          uint16_t firstGlyphID,
+                          uint16_t lastGlyphID);
 
 void append_cmap_sections(const SkTDArray<SkUnichar>& glyphToUnicode,
                           const SkPDFGlyphSet* subset,
-                          SkDynamicMemoryWStream* cmap) {
+                          SkDynamicMemoryWStream* cmap,
+                          uint16_t firstGlyphID,
+                          uint16_t lastGlyphID) {
     if (glyphToUnicode.isEmpty()) {
         return;
     }
 
     SkTDArray<BFChar> bfcharEntries;
     SkTDArray<BFRange> bfrangeEntries;
 
     BFRange currentRangeEntry = {0, 0, 0};
     bool rangeEmpty = true;
-    const int count = glyphToUnicode.count();
+    // We must until lastGlyphID (inclusive) plus one more char.
+    const int limit = SkMin32(lastGlyphID, glyphToUnicode.count()) + 2;

[vandebo (ex-Chrome), 2013/09/09 20:00:29]

These changes seem to work for me and are closer to what's already here:

const int limit = SkMin32(lastGlyphID, glyphToUnicode.count());

[edisonn, 2013/09/10 14:50:11]

It does not work with the new tests I added. Please apply the ToUnitode.cpp
test.
the issue is that count requires a < operator, while lastGlyph requires a <=
operator. 

To make it work I can use SkMin32(lastGlyphID + 1, glyphToUnicode.count());

This works and makes the code simpler, but I do not like it because the code is
based on a technicality, and the idea why we use +1 is lost in the code, and the
readability is lost from the code, it is like a compiled code for perf and
beauty

 
On 2013/09/09 20:00:29, vandebo wrote:

[vandebo (ex-Chrome), 2013/09/12 22:23:39]

Fair, I think this can be fixed by using another variable or two and working on
variable names...

const int lastUnicodeIndex = glyphToUnicode.count() - 1;
const int endIndex = SkMin32(lastGlyphID, lastUnicodeIndex) + 1;

// Loop around once more after endIndex in order to close any open range
for (int i ...

?

 
-    for (int i = 0; i < count + 1; ++i) {
-        bool inSubset = i < count && (subset == NULL || subset->has(i));
+    for (int i = firstGlyphID; i < limit; ++i) {

[vandebo (ex-Chrome), 2013/09/09 20:00:29]

i < limit + 1

[edisonn, 2013/09/10 14:50:11]

Done.

+        bool inSubset = i < glyphToUnicode.count() &&

[vandebo (ex-Chrome), 2013/09/09 20:00:29]

i < limit &&

+                        i <= lastGlyphID &&
+                        (subset == NULL || subset->has(i));
         if (!rangeEmpty) {
             // PDF spec requires bfrange not changing the higher byte,
             // e.g. <1035> <10FF> <2222> is ok, but
             //      <1035> <1100> <2222> is no good
             bool inRange =
                 i == currentRangeEntry.fEnd + 1 &&
                 i >> 8 == currentRangeEntry.fStart >> 8 &&
-                i < count &&
+                i < glyphToUnicode.count() && i <= lastGlyphID &&

[vandebo (ex-Chrome), 2013/09/09 20:00:29]

i < limit &&

[edisonn, 2013/09/10 14:50:11]

Done.

                 glyphToUnicode[i] == currentRangeEntry.fUnicode + i -
                                          currentRangeEntry.fStart;
             if (!inSubset || !inRange) {
                 if (currentRangeEntry.fEnd > currentRangeEntry.fStart) {
                     bfrangeEntries.push(currentRangeEntry);
                 } else {
                     BFChar* entry = bfcharEntries.append();
                     entry->fGlyphId = currentRangeEntry.fStart;
                     entry->fUnicode = currentRangeEntry.fUnicode;
                 }
@@ skipping 11 matching lines @@
     }
 
     // The spec requires all bfchar entries for a font must come before bfrange
     // entries.
     append_bfchar_section(bfcharEntries, cmap);
     append_bfrange_section(bfrangeEntries, cmap);
 }
 
 static SkPDFStream* generate_tounicode_cmap(
         const SkTDArray<SkUnichar>& glyphToUnicode,
-        const SkPDFGlyphSet* subset) {
+        const SkPDFGlyphSet* subset,
+        uint16_t firstGlyphID,
+        uint16_t lastGlyphID) {
     SkDynamicMemoryWStream cmap;
-    append_tounicode_header(&cmap);
-    append_cmap_sections(glyphToUnicode, subset, &cmap);
+    append_tounicode_header(&cmap, firstGlyphID, lastGlyphID);
+    append_cmap_sections(glyphToUnicode, subset, &cmap,
+                         firstGlyphID, lastGlyphID);
     append_cmap_footer(&cmap);
     SkAutoTUnref<SkMemoryStream> cmapStream(new SkMemoryStream());
     cmapStream->setData(cmap.copyToData())->unref();
     return new SkPDFStream(cmapStream.get());
 }
 
 #if defined (SK_SFNTLY_SUBSETTER)
 static void sk_delete_array(const void* ptr, size_t, void*) {
     // Use C-style cast to cast away const and cast type simultaneously.
     delete[] (unsigned char*)ptr;
@@ skipping 465 matching lines @@
     : fFont(font),
       fFontID(fontID),
       fGlyphID(glyphID) {
 }
 
 void SkPDFFont::populateToUnicodeTable(const SkPDFGlyphSet* subset) {
     if (fFontInfo == NULL || fFontInfo->fGlyphToUnicode.begin() == NULL) {
         return;
     }
     SkAutoTUnref<SkPDFStream> pdfCmap(
-        generate_tounicode_cmap(fFontInfo->fGlyphToUnicode, subset));
+        generate_tounicode_cmap(fFontInfo->fGlyphToUnicode, subset,
+                                firstGlyphID(), lastGlyphID()));
     addResource(pdfCmap.get());
     insert("ToUnicode", new SkPDFObjRef(pdfCmap.get()))->unref();
 }
 
 ///////////////////////////////////////////////////////////////////////////////
 // class SkPDFType0Font
 ///////////////////////////////////////////////////////////////////////////////
 
 SkPDFType0Font::SkPDFType0Font(SkAdvancedTypefaceMetrics* info,
                                SkTypeface* typeface)
@@ skipping 380 matching lines @@
 
     insert("FontBBox", makeFontBBox(bbox, 1000))->unref();
     insertInt("FirstChar", firstGlyphID());
     insertInt("LastChar", lastGlyphID());
     insert("Widths", widthArray.get());
     insertName("CIDToGIDMap", "Identity");
 
     populateToUnicodeTable(NULL);
     return true;
 }