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

src/pdf/SkPDFFont.cpp

Index: src/pdf/SkPDFFont.cpp
===================================================================
--- src/pdf/SkPDFFont.cpp	(revision 11071)
+++ src/pdf/SkPDFFont.cpp	(working copy)
@@ -342,7 +342,9 @@
 
 }  // namespace
 
-static void append_tounicode_header(SkDynamicMemoryWStream* cmap) {
+static void append_tounicode_header(SkDynamicMemoryWStream* cmap,
+                                    uint16_t firstGlypthID,
+                                    uint16_t lastGlypthID) {
     // 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 =
@@ -365,17 +367,20 @@
 
     // 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", firstGlypthID, lastGlypthID);
+    cmap->writeText(range.c_str());
+    
+    const char* kTypeInfoFooter = "endcodespacerange\n";
+    cmap->writeText(kTypeInfoFooter);
 }
 
 static void append_cmap_footer(SkDynamicMemoryWStream* cmap) {
@@ -399,11 +404,32 @@
 };
 
 static void append_bfchar_section(const SkTDArray<BFChar>& bfchar,
-                                  SkDynamicMemoryWStream* cmap) {
+                                  SkDynamicMemoryWStream* cmap,
+                                  uint16_t firstGlypthID,
+                                  uint16_t lastGlypthID) {    
+    int32_t i = 0;
+    while (i < bfchar.count() && bfchar[i].fGlyphId < firstGlypthID) {

[vandebo (ex-Chrome), 2013/09/04 22:07:53]

While the end point probably doesn't take a lot of time, this one could. 
Probably worth the effort to figure out SkTSearch and use it here.  At that
point you might as well use it below as well.  Or you could do both outside the
loop and note the start and end indices. 

[edisonn, 2013/09/06 18:54:38]

N/A - moved filtering in the caller - fastest way to do something is not to do
it at all.

On 2013/09/04 22:07:53, vandebo wrote:

+        i++;
+    }
+    
     // PDF spec defines that every bf* list can have at most 100 entries.
-    for (int i = 0; i < bfchar.count(); i += 100) {
+    for (; i < bfchar.count(); i += 100) {
+        if (bfchar[i].fGlyphId > lastGlypthID) {
+            return;
+        }
+
         int count = bfchar.count() - i;
         count = SkMin32(count, 100);
+        // TODO(edisonn): perf, might be better to do a bsearch, but I think
+        // most of the times, we will have enough data to stop imediately.
+        while (count > 0 && bfchar[i + count - 1].fGlyphId > lastGlypthID) {
+            count--;
+        }
+
+        if (count == 0) {

[vandebo (ex-Chrome), 2013/09/04 22:07:53]

Not possible, count must be >= 1 because of the check on line 417

[edisonn, 2013/09/06 18:54:38]

On 2013/09/04 22:07:53, vandebo wrote:
dead code

Done.

+            return;
+        }
+        
         cmap->writeDecAsText(count);
         cmap->writeText(" beginbfchar\n");
         for (int j = 0; j < count; ++j) {
@@ -418,18 +444,36 @@
 }
 
 static void append_bfrange_section(const SkTDArray<BFRange>& bfrange,
-                                   SkDynamicMemoryWStream* cmap) {
+                                   SkDynamicMemoryWStream* cmap,
+                                   uint16_t firstGlypthID,
+                                   uint16_t lastGlypthID) {
+    int32_t i = 0;
+    while (i < bfrange.count() && bfrange[i].fEnd < firstGlypthID) {

[vandebo (ex-Chrome), 2013/09/04 22:07:53]

Same here re SkTSearch

[edisonn, 2013/09/06 18:54:38]

dead code
On 2013/09/04 22:07:53, vandebo wrote:

+        i++;
+    }
+
     // PDF spec defines that every bf* list can have at most 100 entries.
-    for (int i = 0; i < bfrange.count(); i += 100) {
+    for (; i < bfrange.count(); i += 100) {
+        if (bfrange[i].fStart > lastGlypthID) {
+            return;
+        }
+
         int count = bfrange.count() - i;
         count = SkMin32(count, 100);
         cmap->writeDecAsText(count);
         cmap->writeText(" beginbfrange\n");
+
+        // first range might be chopped.
+        uint16_t start = SkMax32(firstGlypthID, bfrange[i].fStart);
+
         for (int j = 0; j < count; ++j) {
+            if (bfrange[i + j].fStart > lastGlypthID) {
+                break;
+            }
             cmap->writeText("<");
-            cmap->writeHexAsText(bfrange[i + j].fStart, 4);
+            cmap->writeHexAsText(j == 0 ? start : bfrange[i + j].fStart, 4);

[vandebo (ex-Chrome), 2013/09/04 22:07:53]

It's probably the same computational cost to do the trinary operator as it is to
just inline the SkMax32 here.

[edisonn, 2013/09/06 18:54:38]

dead code

             cmap->writeText("> <");
-            cmap->writeHexAsText(bfrange[i + j].fEnd, 4);
+            cmap->writeHexAsText(SkMin32(bfrange[i + j].fEnd, lastGlypthID), 4);
             cmap->writeText("> <");
             cmap->writeHexAsText(bfrange[i + j].fUnicode, 4);
             cmap->writeText(">\n");
@@ -469,11 +513,15 @@
 // ( see caller in tests/ToUnicode.cpp )
 void append_cmap_sections(const SkTDArray<SkUnichar>& glyphToUnicode,
                           const SkPDFGlyphSet* subset,
-                          SkDynamicMemoryWStream* cmap);
+                          SkDynamicMemoryWStream* cmap,
+                          uint16_t firstGlypthID = 0x0000,

[vandebo (ex-Chrome), 2013/09/04 22:07:53]

No need for defaults, these args are always passed in.

[edisonn, 2013/09/06 18:54:38]

Done.

+                          uint16_t lastGlypthID  = 0xffff);
 
 void append_cmap_sections(const SkTDArray<SkUnichar>& glyphToUnicode,
                           const SkPDFGlyphSet* subset,
-                          SkDynamicMemoryWStream* cmap) {
+                          SkDynamicMemoryWStream* cmap,
+                          uint16_t firstGlypthID,
+                          uint16_t lastGlypthID) {
     if (glyphToUnicode.isEmpty()) {
         return;
     }
@@ -520,16 +568,19 @@
 
     // The spec requires all bfchar entries for a font must come before bfrange
     // entries.
-    append_bfchar_section(bfcharEntries, cmap);
-    append_bfrange_section(bfrangeEntries, cmap);
+    append_bfchar_section(bfcharEntries, cmap, firstGlypthID, lastGlypthID);
+    append_bfrange_section(bfrangeEntries, cmap, firstGlypthID, lastGlypthID);
 }
 
 static SkPDFStream* generate_tounicode_cmap(
         const SkTDArray<SkUnichar>& glyphToUnicode,
-        const SkPDFGlyphSet* subset) {
+        const SkPDFGlyphSet* subset,
+        uint16_t firstGlypthID,
+        uint16_t lastGlypthID) {
     SkDynamicMemoryWStream cmap;
-    append_tounicode_header(&cmap);
-    append_cmap_sections(glyphToUnicode, subset, &cmap);
+    append_tounicode_header(&cmap, firstGlypthID, lastGlypthID);
+    append_cmap_sections(glyphToUnicode, subset, &cmap,
+                         firstGlypthID, lastGlypthID);
     append_cmap_footer(&cmap);
     SkAutoTUnref<SkMemoryStream> cmapStream(new SkMemoryStream());
     cmapStream->setData(cmap.copyToData())->unref();
@@ -1014,7 +1065,8 @@
         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();
 }