Roll sfntly 111

third_party/sfntly/src/subsetter/subsetter_impl.cc

Index: third_party/sfntly/src/subsetter/subsetter_impl.cc
diff --git a/third_party/sfntly/src/subsetter/subsetter_impl.cc b/third_party/sfntly/src/subsetter/subsetter_impl.cc
index bed1fc1d2cf5215d73cc2fc12a671c105cd3978b..4eb3e8323f3f41668cd56d1e7c91d9a74810ad11 100644
--- a/third_party/sfntly/src/subsetter/subsetter_impl.cc
+++ b/third_party/sfntly/src/subsetter/subsetter_impl.cc
@@ -21,6 +21,14 @@
 #include <map>
 #include <set>
 
+#include "third_party/sfntly/src/sfntly/table/bitmap/eblc_table.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/ebdt_table.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/index_sub_table.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/index_sub_table_format1.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/index_sub_table_format2.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/index_sub_table_format3.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/index_sub_table_format4.h"
+#include "third_party/sfntly/src/sfntly/table/bitmap/index_sub_table_format5.h"
 #include "third_party/sfntly/src/sfntly/table/core/name_table.h"
 #include "third_party/sfntly/src/sfntly/table/truetype/glyph_table.h"
 #include "third_party/sfntly/src/sfntly/table/truetype/loca_table.h"
@@ -29,10 +37,21 @@
 #include "third_party/sfntly/src/sfntly/port/memory_input_stream.h"
 #include "third_party/sfntly/src/sfntly/port/memory_output_stream.h"
 
+#if defined U_USING_ICU_NAMESPACE
+  U_NAMESPACE_USE
+#endif

[Nico, 2011/11/30 18:19:05]

Are you sure you need this? 

In chrome, this bit from third_party/icu/icu.gyp should set things up correctly
already:

  'target_defaults': {
    'direct_dependent_settings': {
      'defines': [
        # Tell ICU to not insert |using namespace icu;| into its headers,
        # so that chrome's source explicitly has to use |icu::|.
        'U_USING_ICU_NAMESPACE=0',
      ],
    },

[arthurhsu, 2011/11/30 18:30:01]

It's not done so because upstream sfntly does not have the support (and yes,
this code you see here actually has a copy in upstream's test).

On 2011/11/30 18:19:05, Nico wrote:

+
+namespace {
+
+// The bitmap tables must be greater than 16KB to trigger bitmap subsetter.
+static const int BITMAP_SIZE_THRESHOLD = 16384;
+
+}
+
 namespace sfntly {
 
 void ConstructName(UChar* name_part, UnicodeString* name, int32_t name_id) {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

Shouldn't this be in an anonymous namespace?  The same goes for all the
non-class member functions.

[arthurhsu, 2011/12/01 00:59:35]

Upstream has test cases testing the functions so namespace sfntly is used.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

Can we put the ones with explicit unit tests into a namespace sftly_unittested
(or similar) and move the rest to the anonymous namespace?

[arthurhsu, 2011/12/02 00:06:47]

Moved those are not used for now.  namespace hierarchy change will need
stuartg/bstell to okay it.

-  switch(name_id) {
+  switch (name_id) {
     case NameId::kFullFontName:
       *name = name_part;
       break;
@@ -71,80 +90,7 @@ int32_t HashCode(int32_t platform_id, int32_t encoding_id, int32_t language_id,
   return result;
 }
 
-SubsetterImpl::SubsetterImpl() {
-}
-
-SubsetterImpl::~SubsetterImpl() {
-}
-
-bool SubsetterImpl::LoadFont(const char* font_name,
-                             const unsigned char* original_font,
-                             size_t font_size) {
-  MemoryInputStream mis;
-  mis.Attach(original_font, font_size);
-  if (factory_ == NULL) {
-    factory_.Attach(FontFactory::GetInstance());
-  }
-
-  FontArray font_array;
-  factory_->LoadFonts(&mis, &font_array);
-  font_ = FindFont(font_name, font_array);
-  if (font_ == NULL) {
-    return false;
-  }
-
-  return true;
-}
-
-int SubsetterImpl::SubsetFont(const unsigned int* glyph_ids,
-                              size_t glyph_count,
-                              unsigned char** output_buffer) {
-  if (factory_ == NULL || font_ == NULL) {
-    return -1;
-  }
-
-  IntegerSet glyph_id_processed;
-  if (!ResolveCompositeGlyphs(glyph_ids, glyph_count, &glyph_id_processed) ||
-      glyph_id_processed.empty()) {
-    return 0;
-  }
-
-  FontPtr new_font;
-  new_font.Attach(Subset(glyph_id_processed));
-  if (new_font == NULL) {
-    return 0;
-  }
-
-  MemoryOutputStream output_stream;
-  factory_->SerializeFont(new_font, &output_stream);
-  int length = static_cast<int>(output_stream.Size());
-  if (length > 0) {
-    *output_buffer = new unsigned char[length];
-    memcpy(*output_buffer, output_stream.Get(), length);
-  }
-
-  return length;
-}
-
-Font* SubsetterImpl::FindFont(const char* font_name,
-                              const FontArray& font_array) {
-  if (font_array.empty() || font_array[0] == NULL) {
-    return NULL;
-  }
-
-  if (font_name && strlen(font_name)) {
-    for (FontArray::const_iterator b = font_array.begin(), e = font_array.end();
-         b != e; ++b) {
-      if (HasName(font_name, (*b).p_)) {
-        return (*b).p_;
-      }
-    }
-  }
-
-  return font_array[0].p_;
-}
-
-bool SubsetterImpl::HasName(const char* font_name, Font* font) {
+bool HasName(const char* font_name, Font* font) {
   UnicodeString font_string = UnicodeString::fromUTF8(font_name);
   if (font_string.isEmpty())
     return false;
@@ -160,7 +106,7 @@ bool SubsetterImpl::HasName(const char* font_name, Font* font) {
   }
 
   for (int32_t i = 0; i < name_table->NameCount(); ++i) {
-    switch(name_table->NameId(i)) {
+    switch (name_table->NameId(i)) {
       case NameId::kFontFamilyName:
       case NameId::kFontSubfamilyName:
       case NameId::kFullFontName:
@@ -193,22 +139,37 @@ bool SubsetterImpl::HasName(const char* font_name, Font* font) {
   return false;
 }
 
-bool SubsetterImpl::ResolveCompositeGlyphs(const unsigned int* glyph_ids,
-                                           size_t glyph_count,
-                                           IntegerSet* glyph_id_processed) {
-  if (glyph_ids == NULL || glyph_count == 0 || glyph_id_processed == NULL) {
-    return false;
+Font* FindFont(const char* font_name, const FontArray& font_array) {
+  if (font_array.empty() || font_array[0] == NULL) {
+    return NULL;
   }
 
-  // Find glyf and loca table.
-  GlyphTablePtr glyph_table =
-      down_cast<GlyphTable*>(font_->GetTable(Tag::glyf));
-  LocaTablePtr loca_table = down_cast<LocaTable*>(font_->GetTable(Tag::loca));
-  if (glyph_table == NULL || loca_table == NULL) {
-    // The font is invalid.
+  if (font_name && strlen(font_name)) {
+    for (FontArray::const_iterator b = font_array.begin(), e = font_array.end();

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: b -> i, e -> end ?

[arthurhsu, 2011/12/01 00:59:35]

e is used throughout sfntly to represent end.  b->i done.

+         b != e; ++b) {
+      if (HasName(font_name, (*b).p_)) {
+        return (*b).p_;

[Nico, 2011/11/30 18:19:05]

b->p_ instead of (*b).p_?

[arthurhsu, 2011/11/30 18:30:01]

Done.

+      }
+    }
+  }
+
+  return font_array[0].p_;
+}
+
+bool ResolveCompositeGlyphs(GlyphTable* glyf,
+                            LocaTable* loca,
+                            const unsigned int* glyph_ids,
+                            size_t glyph_count,
+                            IntegerSet* glyph_id_processed) {
+  if (glyf == NULL || loca == NULL || glyph_ids == NULL || glyph_count == 0 ||
+      glyph_id_processed == NULL) {
     return false;
   }
 
+  // Find glyf and loca table.
+  GlyphTablePtr glyph_table = glyf;
+  LocaTablePtr loca_table = loca;
+
   // Sort and uniquify glyph ids.
   IntegerSet glyph_id_remaining;
   glyph_id_remaining.insert(0);  // Always include glyph id 0.
@@ -263,23 +224,26 @@ bool SubsetterImpl::ResolveCompositeGlyphs(const unsigned int* glyph_ids,
   return true;
 }
 
-CALLER_ATTACH Font* SubsetterImpl::Subset(const IntegerSet& glyph_ids) {
-  // The tables are already checked in ResolveCompositeGlyphs().
-  GlyphTablePtr glyph_table =
-      down_cast<GlyphTable*>(font_->GetTable(Tag::glyf));
-  LocaTablePtr loca_table = down_cast<LocaTable*>(font_->GetTable(Tag::loca));
+bool SetupGlyfBuilders(Font::Builder* builder,

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: SetUp

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

FYI: http://notaverb.com/setup

+                       GlyphTable* glyf,
+                       LocaTable* loca,
+                       const IntegerSet& glyph_ids) {
+  if (!builder || !glyf || !loca) {
+    return false;
+  }
 
-  // Setup font builders we need.
-  FontBuilderPtr font_builder;
-  font_builder.Attach(factory_->NewFontBuilder());
+  // The tables are already checked in ResolveCompositeGlyphs().
+  GlyphTablePtr glyph_table = glyf;
+  LocaTablePtr loca_table = loca;
 
+  FontBuilderPtr font_builder = builder;
   GlyphTableBuilderPtr glyph_table_builder =
       down_cast<GlyphTable::Builder*>(font_builder->NewTableBuilder(Tag::glyf));
   LocaTableBuilderPtr loca_table_builder =
       down_cast<LocaTable::Builder*>(font_builder->NewTableBuilder(Tag::loca));
   if (glyph_table_builder == NULL || loca_table_builder == NULL) {
     // Out of memory.
-    return NULL;
+    return false;
   }
 
   // Extract glyphs and setup loca list.
@@ -320,12 +284,543 @@ CALLER_ATTACH Font* SubsetterImpl::Subset(const IntegerSet& glyph_ids) {
   }
   loca_table_builder->SetLocaList(&loca_list);
 
+  return true;
+}
+
+bool HasOverlap(int32_t range_begin, int32_t range_end,
+                const IntegerSet& glyph_ids) {
+  if (range_begin == range_end) {
+    return glyph_ids.find(range_begin) != glyph_ids.end();
+  } else if (range_end > range_begin) {
+    IntegerSet::const_iterator left = glyph_ids.lower_bound(range_begin);
+    IntegerSet::const_iterator right = glyph_ids.lower_bound(range_end);
+    return right != left;
+  }
+  return false;
+}
+
+// Initialize builder, returns false if glyph_id subset is not covered.
+bool ShallSubset(EbdtTable::Builder* ebdt, EblcTable::Builder* eblc,

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

better name? SetUpBitmapBuilder?

[arthurhsu, 2011/12/01 00:59:35]

Changed to InitializeBitmapBuilder

+                 const IntegerSet& glyph_ids) {
+  EblcTableBuilderPtr eblc_builder = eblc;

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

Do you need a smart pointer here?  The caller isn't going to give up its
reference in the lifetime of this function.

[arthurhsu, 2011/12/01 00:59:35]

It's more consistent to always use smart pointers to host objects ported from
Java.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

But it's not free.  Every time you use a smart pointer, it requires that the ref
count of the object be updated.  That usually requires an atomic instruction,
which can be quiet expensive on some platforms.  In this case (and several other
in this file), we are sure that the reference count of an object won't reach
zero while in the body of a function because the caller has a reference.  In
those cases, I'd prefer we not take another reference.

[arthurhsu, 2011/12/02 00:06:47]

Done.

+  EbdtTableBuilderPtr ebdt_builder = ebdt;
+
+  BitmapLocaList loca_list;
+  BitmapSizeTableBuilderList* strikes = eblc_builder->BitmapSizeBuilders();
+
+  // Note: Do not call eblc_builder->GenerateLocaList(&loca_list) and then
+  //       ebdt_builder->SetLoca(loca_list).  For fonts like SimSun, there are
+  //       >28K glyphs inside, where a typical usage will be <1K glyphs.  Doing
+  //       the calls improperly will result in creation of >100K objects that
+  //       will be destroyed immediately, inducing significant slowness.
+  IntegerList removed_strikes;
+  for (size_t i = 0; i < strikes->size(); i++) {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

I didn't follow all the details here, but this looks like an n^3 algorithm.  Is
there any easy way to reduce the complexity?

[arthurhsu, 2011/12/01 00:59:35]

It has to be n^3 unfortunately.  We need to go over all subtables and extract
the glyph ids that we need.  Since index subtables do not overlap each other, we
are actually ruling out most of them because we have sparse glyph id
distributions.

+    if (!HasOverlap((*strikes)[i]->StartGlyphIndex(),
+                    (*strikes)[i]->EndGlyphIndex(), glyph_ids)) {
+      removed_strikes.push_back(i);
+      continue;
+    }
+
+    IndexSubTableBuilderList* index_builders =
+        (*strikes)[i]->IndexSubTableBuilders();
+    IntegerList removed_indexes;
+    BitmapGlyphInfoMap info_map;
+    for (size_t j = 0; j < index_builders->size(); ++j) {
+      int32_t first_glyph_id = (*index_builders)[j]->first_glyph_index();
+      int32_t last_glyph_id = (*index_builders)[j]->last_glyph_index();
+      if (!HasOverlap(first_glyph_id, last_glyph_id, glyph_ids)) {
+        removed_indexes.push_back(j);
+        continue;
+      }
+      for (IntegerSet::const_iterator gid = glyph_ids.begin(),
+                                      gid_end = glyph_ids.end();
+                                      gid != gid_end; gid++) {
+        if (*gid < first_glyph_id) {
+          continue;
+        }
+        if (*gid > last_glyph_id) {
+          break;
+        }
+        BitmapGlyphInfoPtr info;
+        info.Attach((*index_builders)[j]->GlyphInfo(*gid));
+        if (info && info->length()) {  // Do not include gid without bitmap
+          info_map[*gid] = info;
+        }
+      }
+    }
+    if (!info_map.empty()) {
+      loca_list.push_back(info_map);
+    } else {
+      removed_strikes.push_back(i);  // Detected null entries.
+    }
+
+    // Remove unused index sub tables
+    for (IntegerList::reverse_iterator j = removed_indexes.rbegin(),
+                                       e = removed_indexes.rend();

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: e -> end

+                                       j != e; j++) {
+      index_builders->erase(index_builders->begin() + *j);
+    }
+  }
+  if (removed_strikes.size() == strikes->size() || loca_list.empty()) {
+    return false;  // All strikes shall be gone.

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: remove comment, redundant

[arthurhsu, 2011/12/01 00:59:35]

Done.

+  }
+
+  // Remove unused strikes

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: Remove comment

[arthurhsu, 2011/12/01 00:59:35]

Done.

+  for (IntegerList::reverse_iterator j = removed_strikes.rbegin(),

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: j -> i

[arthurhsu, 2011/12/01 00:59:35]

Done.

+                                     e = removed_strikes.rend(); j != e; j++) {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: e -> end

+    strikes->erase(strikes->begin() + *j);
+  }
+
+  if (strikes->empty()) {  // no glyph covered, can safely drop the builders.
+    return false;
+  }
+
+  ebdt_builder->SetLoca(&loca_list);
+  ebdt_builder->GlyphBuilders();  // Initialize the builder.
+  return true;
+}
+
+void CopyBigGlyphMetrics(BigGlyphMetrics::Builder* source,
+                         BigGlyphMetrics::Builder* target) {
+  target->SetHeight(static_cast<byte_t>(source->Height()));
+  target->SetWidth(static_cast<byte_t>(source->Width()));
+  target->SetHoriBearingX(static_cast<byte_t>(source->HoriBearingX()));
+  target->SetHoriBearingY(static_cast<byte_t>(source->HoriBearingY()));
+  target->SetHoriAdvance(static_cast<byte_t>(source->HoriAdvance()));
+  target->SetVertBearingX(static_cast<byte_t>(source->VertBearingX()));
+  target->SetVertBearingY(static_cast<byte_t>(source->VertBearingY()));
+  target->SetVertAdvance(static_cast<byte_t>(source->VertAdvance()));
+}
+
+CALLER_ATTACH IndexSubTable::Builder*
+ConstructIndexFormat4(IndexSubTable::Builder* b, const BitmapGlyphInfoMap& loca,
+                      int32_t* image_data_offset) {
+  IndexSubTableFormat4BuilderPtr builder4;
+  builder4.Attach(IndexSubTableFormat4::Builder::CreateBuilder());
+  CodeOffsetPairBuilderList offset_pairs;
+
+  size_t offset = 0;
+  int32_t lower_bound = b->first_glyph_index();
+  int32_t upper_bound = b->last_glyph_index();
+  bool lower_bound_reached = false;
+  bool upper_bound_reached = false;
+  int32_t last_gid = -1;
+  BitmapGlyphInfoMap::const_iterator last_element = loca.end();
+  --last_element;
+  for (BitmapGlyphInfoMap::const_iterator i = loca.begin(), e = loca.end();

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: e -> end

+                                          i != e; i++) {
+    int32_t gid = i->first;
+    if (gid < lower_bound) {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

Could you search for lower_bound outside the loop and start the loop there? Or
does BitmapGlyphInfoMap not let you search?

[arthurhsu, 2011/12/01 00:59:35]

Done.

+      continue;
+    }
+    if (!lower_bound_reached) {
+      builder4->set_first_glyph_index(gid);
+      builder4->set_image_format(b->image_format());
+      builder4->set_image_data_offset(*image_data_offset);
+      last_gid = gid;
+      lower_bound_reached = true;
+    }
+    if (gid > upper_bound) {
+      upper_bound_reached = true;
+    }
+    if (!upper_bound_reached) {
+      offset_pairs.push_back(
+          IndexSubTableFormat4::CodeOffsetPairBuilder(gid, offset));
+      offset += i->second->length();
+      last_gid = gid;
+      if (i == last_element) {
+        upper_bound_reached = true;
+      }
+    }
+    if (upper_bound_reached) {
+      offset_pairs.push_back(

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

If you pull this to after the loop, I think you can get rid of last_element and
upper_bound_reached.

[arthurhsu, 2011/12/01 00:59:35]

upper_bound_reached is still needed to avoid adding out-of-range glyph id into
index table.  The rest is done.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

Not the way I read the code, do I misunderstand?

[arthurhsu, 2011/12/02 00:06:47]

|gid| can't be greater than |upper_bound|.  Using std::lower_bound(upper_bound)
here will still return a |gid| greater or equal to the |upper_bound|.

+          IndexSubTableFormat4::CodeOffsetPairBuilder(-1, offset));
+      builder4->set_last_glyph_index(last_gid);
+      *image_data_offset += offset;
+      break;
+    }
+  }
+  builder4->SetOffsetArray(offset_pairs);
+
+  return builder4.Detach();
+}
+
+CALLER_ATTACH IndexSubTable::Builder*
+ConstructIndexFormat5(IndexSubTable::Builder* b, const BitmapGlyphInfoMap& loca,
+                      int32_t* image_data_offset) {
+  IndexSubTableFormat5BuilderPtr new_builder;
+  new_builder.Attach(IndexSubTableFormat5::Builder::CreateBuilder());
+
+  // Copy BigMetrics
+  int32_t image_size = 0;
+  if (b->index_format() == IndexSubTable::Format::FORMAT_2) {
+    IndexSubTableFormat2BuilderPtr builder2 =
+      down_cast<IndexSubTableFormat2::Builder*>(b);
+    CopyBigGlyphMetrics(builder2->BigMetrics(), new_builder->BigMetrics());
+    image_size = builder2->ImageSize();
+  } else {
+    IndexSubTableFormat5BuilderPtr builder5 =
+      down_cast<IndexSubTableFormat5::Builder*>(b);
+    BigGlyphMetricsBuilderPtr metrics_builder;
+    CopyBigGlyphMetrics(builder5->BigMetrics(), new_builder->BigMetrics());
+    image_size = builder5->ImageSize();
+  }
+
+  IntegerList* glyph_array = new_builder->GlyphArray();
+  size_t offset = 0;
+  int32_t lower_bound = b->first_glyph_index();
+  int32_t upper_bound = b->last_glyph_index();
+  bool lower_bound_reached = false;
+  bool upper_bound_reached = false;
+  int32_t last_gid = -1;
+  BitmapGlyphInfoMap::const_iterator last_element = loca.end();
+  --last_element;
+  for (BitmapGlyphInfoMap::const_iterator i = loca.begin(), e = loca.end();

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

Similar comments apply to this loop.

[arthurhsu, 2011/12/01 00:59:35]

Done.

+                                          i != e; i++) {
+    int32_t gid = i->first;
+    if (gid < lower_bound) {
+      continue;
+    }
+    if (!lower_bound_reached) {
+      new_builder->set_first_glyph_index(gid);
+      new_builder->set_image_format(b->image_format());
+      new_builder->set_image_data_offset(*image_data_offset);
+      new_builder->SetImageSize(image_size);
+      last_gid = gid;
+      lower_bound_reached = true;
+    }
+    if (gid > upper_bound || i == last_element) {
+      upper_bound_reached = true;
+    }
+    if (!upper_bound_reached || i == last_element) {
+      glyph_array->push_back(gid);
+      offset += i->second->length();
+      last_gid = gid;
+    }
+    if (upper_bound_reached) {
+      new_builder->set_last_glyph_index(last_gid);
+      *image_data_offset += offset;
+      break;
+    }
+  }
+  return new_builder.Detach();
+}
+
+CALLER_ATTACH IndexSubTable::Builder*
+SubsetIndexSubTable(IndexSubTable::Builder* builder,
+                    const BitmapGlyphInfoMap& loca,
+                    int32_t* image_data_offset) {
+  switch (builder->index_format()) {
+    case IndexSubTable::Format::FORMAT_1:
+    case IndexSubTable::Format::FORMAT_3:
+    case IndexSubTable::Format::FORMAT_4:
+      return ConstructIndexFormat4(builder, loca, image_data_offset);
+    case IndexSubTable::Format::FORMAT_2:
+    case IndexSubTable::Format::FORMAT_5:
+      return ConstructIndexFormat5(builder, loca, image_data_offset);
+    default:
+      assert(false);  // Shall not be here.

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

Remove comment.

[arthurhsu, 2011/12/01 00:59:35]

Done.

+      break;
+  }
+  return NULL;
+}
+
+void SubsetEBLC(EblcTable::Builder* eblc, const BitmapLocaList& new_loca) {
+  EblcTableBuilderPtr eblc_builder = eblc;

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

No need for a smart pointer here - the caller holds a reference.

[arthurhsu, 2011/12/02 00:06:47]

Done.

+  BitmapSizeTableBuilderList* size_builders = eblc->BitmapSizeBuilders();
+  if (size_builders == NULL) {
+    return;  // No valid EBLC.

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

redundant comment

[arthurhsu, 2011/12/01 00:59:35]

Done.

+  }
+
+  int32_t image_data_offset = EbdtTable::Offset::kHeaderLength;
+  for (size_t strike = 0; strike < size_builders->size(); ++strike) {
+    IndexSubTableBuilderList* index_builders =
+        (*size_builders)[strike]->IndexSubTableBuilders();
+    for (size_t index = 0; index < index_builders->size(); ++index) {
+      IndexSubTable::Builder* new_builder_raw =
+          SubsetIndexSubTable((*index_builders)[index], new_loca[strike],
+                              &image_data_offset);
+      if (NULL != new_builder_raw) {
+        (*index_builders)[index].Attach(new_builder_raw);
+      }
+    }
+  }
+}
+
+/******************************************************************************

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

// style comments are more common than /* */ style.  The style guide says to use
one or the other.

[arthurhsu, 2011/12/01 00:59:35]

Done.

+  Long background comments

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

This line isn't necessary

[arthurhsu, 2011/12/01 00:59:35]

Done.

+
+EBLC structure:
+  header
+  bitmapSizeTable[]
+    one per strike
+    holds strike metrics - sbitLineMetrics
+    holds info about indexSubTableArray
+  indexSubTableArray[][]
+    one per strike and then one per indexSubTable for that strike
+    holds info about the indexSubTable
+    the indexSubTable entries pointed to can be of different formats
+  indexSubTable
+    one per indexSubTableArray entry
+    tells how to get the glyphs
+    may hold the glyph metrics if they are uniform for all the glyphs in range
+
+There is nothing that says that the indexSubTableArray entries and/or the
+indexSubTable items need to be unique. They may be shared between strikes.
+
+EBDT structure:
+  header
+  glyphs
+    amorphous blob of data
+    different glyphs that are only able to be figured out from the EBLC table
+    may hold metrics - depends on the EBLC entry that pointed to them
+
+Subsetting EBLC table:
+  Most pages use only a fraction (hundreds or less) glyphs out of a given font
+  (which can have >20K glyphs for CJK).  It's safe to assume that the subset
+  font will have sparse bitmap glyphs.  As a result, the EBLC table shall be

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: "As a result..." -> "So reconstruct it as a format 4 or 5 table."

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

Missed

[arthurhsu, 2011/12/02 00:06:47]

Done.

+  reconstructed to either format 4 or 5.
+*******************************************************************************/
+bool SetupBitmapBuilders(Font* font, Font::Builder* builder,

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: SetUp

+                         const IntegerSet& glyph_ids, bool use_ebdt) {
+  if (!font || !builder) {
+    return false;
+  }
+
+  EbdtTablePtr ebdt_table =

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

This may be the ebdt or bdat table, so calling it the ebdt table could be
confusion.  Is there a generic name you can use for either table?

[arthurhsu, 2011/12/01 00:59:35]

bdat is actually zoo'ed and most fonts use ebdt therefore ebdt is chosen in Java
code.  I use the same name as its class type to avoid confusion.

+      down_cast<EbdtTable*>(font->GetTable(use_ebdt ? Tag::EBDT : Tag::bdat));
+  EblcTablePtr eblc_table =
+      down_cast<EblcTable*>(font->GetTable(use_ebdt ? Tag::EBLC : Tag::bloc));
+
+  // If the bitmap table's size is too small, skip subsetting.
+  if (ebdt_table->DataLength() + eblc_table->DataLength() <
+      BITMAP_SIZE_THRESHOLD) {
+    return true;
+  }
+
+  // Get the builders.
+  FontBuilderPtr font_builder = builder;
+  EbdtTableBuilderPtr ebdt_table_builder = down_cast<EbdtTable::Builder*>(
+      font_builder->NewTableBuilder(use_ebdt ? Tag::EBDT : Tag::bdat,
+                                    ebdt_table->ReadFontData()));
+  EblcTableBuilderPtr eblc_table_builder = down_cast<EblcTable::Builder*>(
+      font_builder->NewTableBuilder(use_ebdt ? Tag::EBLC : Tag::bloc,
+                                    eblc_table->ReadFontData()));
+  if (ebdt_table_builder == NULL || eblc_table_builder == NULL) {
+    // Out of memory.
+    return false;
+  }
+
+  if (!ShallSubset(ebdt_table_builder, eblc_table_builder, glyph_ids)) {
+    // Bitmap tables do not cover the glyphs in our subset.
+    font_builder->RemoveTableBuilder(use_ebdt ? Tag::EBLC : Tag::bloc);
+    font_builder->RemoveTableBuilder(use_ebdt ? Tag::EBDT : Tag::bdat);
+    return false;
+  }
+
+  BitmapLocaList new_loca;
+  ebdt_table_builder->GenerateLocaList(&new_loca);
+  SubsetEBLC(eblc_table_builder, new_loca);
+
+  return true;
+}
+
+enum BitmapDetection {
+  kNotFound,
+  kEBDTFound,
+  kOnlyBDATFound
+};
+
+// Some fonts have both EBDT/EBLC and bdat/bloc, we need only one set of them.
+int DetectBitmapBuilders(Font* font) {
+  // Check if bitmap table exists.
+  EbdtTablePtr ebdt_table = down_cast<EbdtTable*>(font->GetTable(Tag::EBDT));
+  EblcTablePtr eblc_table = down_cast<EblcTable*>(font->GetTable(Tag::EBLC));
+  if (ebdt_table == NULL && eblc_table == NULL) {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

What happens if we have just one of these two tables?

[arthurhsu, 2011/12/01 00:59:35]

It's a rogue font.  If the rogue font reaches here, OTS has a bug to fix since
it failed the checking.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

I think an easier and more fault tolerant structure would then be:

if (ebdt_table != NULL && eblc_table != NULL)
  return kEBDTFound;
...
if (bdat_table != NULL && bloc_table != NULL)
  return kEBDTFound;
return kNotFound;

[arthurhsu, 2011/12/02 00:06:47]

Done.

+    // Check BDAT variants.
+    ebdt_table = down_cast<EbdtTable*>(font->GetTable(Tag::bdat));
+    eblc_table = down_cast<EblcTable*>(font->GetTable(Tag::bloc));
+    if (ebdt_table == NULL || eblc_table == NULL) {
+      // There's no bitmap tables.
+      return kNotFound;
+    }
+    return kOnlyBDATFound;
+  }
+  return kEBDTFound;
+}
+
+SubsetterImpl::SubsetterImpl() {
+}
+
+SubsetterImpl::~SubsetterImpl() {
+}
+
+bool SubsetterImpl::LoadFont(const char* font_name,
+                             const unsigned char* original_font,
+                             size_t font_size) {
+  MemoryInputStream mis;
+  mis.Attach(original_font, font_size);
+  if (factory_ == NULL) {
+    factory_.Attach(FontFactory::GetInstance());
+  }
+
+  FontArray font_array;
+  factory_->LoadFonts(&mis, &font_array);
+  font_ = FindFont(font_name, font_array);
+  if (font_ == NULL) {
+    return false;
+  }
+
+  return true;
+}
+
+int SubsetterImpl::SubsetFont(const unsigned int* glyph_ids,
+                              size_t glyph_count,
+                              unsigned char** output_buffer) {
+  if (factory_ == NULL || font_ == NULL) {
+    return -1;
+  }
+
+  // Find glyf and loca table.
+  GlyphTablePtr glyph_table =
+      down_cast<GlyphTable*>(font_->GetTable(Tag::glyf));
+  LocaTablePtr loca_table = down_cast<LocaTable*>(font_->GetTable(Tag::loca));
+  if (glyph_table == NULL || loca_table == NULL) {
+    // We are not able to subset the font.
+    return 0;
+  }
+
+  IntegerSet glyph_id_processed;
+  if (!ResolveCompositeGlyphs(glyph_table, loca_table,
+                              glyph_ids, glyph_count, &glyph_id_processed) ||
+      glyph_id_processed.empty()) {
+    return 0;
+  }
+
+  FontPtr new_font;
+  new_font.Attach(Subset(glyph_id_processed));
+  if (new_font == NULL) {
+    return 0;
+  }
+
+  MemoryOutputStream output_stream;
+  factory_->SerializeFont(new_font, &output_stream);
+  int length = static_cast<int>(output_stream.Size());
+  if (length > 0) {
+    *output_buffer = new unsigned char[length];
+    memcpy(*output_buffer, output_stream.Get(), length);
+  }
+
+  return length;
+}
+
+/*******************************************************************************
+  Long comments regarding TTF tables and PDF
+
+According to PDF spec (section 9.9), the following tables must present:

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: must be present

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

9.9 in which version of the spec?

[arthurhsu, 2011/12/01 00:59:35]

This comment is done by stuartg and thus I'd rather keep it as-is.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

My comments only request clarification or fix grammar issues, so please address
them.

[arthurhsu, 2011/12/02 00:06:47]

Done.

+  head, hhea, loca, maxp, cvt, prep, glyf, hmtx, fpgm
+  cmap if font is used as a TTF and not a CIDFont dict

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: TTF -> "simple font"

[arthurhsu, 2011/12/02 00:06:47]

Done.

+
+Other tables we need to keep for PDF rendering to support zoom in/out:
+  bdat, bloc, ebdt, eblc, ebsc, gasp
+
+Special table:
+  CFF - if you have this table then you shouldn't have a glyf table and this is
+        the table with all the glyphs.  Shall skip subsetting completely since
+        sfntly is not capable of subsetting it for now.
+  post - extra info here for printing on PostScript printers but maybe not
+         enough to outweigh the space taken by the names
+
+Tables to break apart:
+  name - could throw away all but one language and one platform strings / might
+         throw away some of the name entries
+  cmap - could strip out non-needed cmap subtables

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

We always use TTF fonts as CIDFonts and never as simple fonts in Chromium, so do
we need the cmap at all?

[arthurhsu, 2011/12/01 00:59:35]

Theoretically, we don't need it until implementing glyph renumbering (if it's
going to be done).  However, cmap is a core table for TTF font and relatively
small.  I don't think we want to risk taking it out giving the limited benefit.

+       - format 4 subtable can be subsetted as well using sfntly
+
+Graphite tables:
+  silf, glat, gloc, feat - shall be okay to strip out

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: shall -> should

[arthurhsu, 2011/12/02 00:06:47]

Done.

+
+Tables that can be discarded:
+  OS/2 - everything here is for layout and description of the font that is
+         elsewhere (some in the PDF objects)
+  BASE, GDEF, GSUB, GPOS, JSTF - all used for layout
+  kern - old style layout
+  DSIG - this will be invalid after subsetting
+  hdmx - layout
+  PCLT - metadata that's not needed
+  vmtx - layout
+  vhea - layout
+  VDMX
+  VORG - not used by TT/OT - used by CFF
+  hsty - would be surprised if you saw one of these - used on the Newton

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: "if you saw" -> "to see"

[arthurhsu, 2011/12/02 00:06:47]

Done.

+  AAT tables - mort, morx, feat, acnt, bsin, just, lcar, fdsc, fmtx, prop,
+               Zapf, opbd, trak, fvar, gvar, avar, cvar
+             - these are all layout tables and once layout happens are not
+               needed anymore
+  LTSH - layout
+*******************************************************************************/
+CALLER_ATTACH Font* SubsetterImpl::Subset(const IntegerSet& glyph_ids) {
+  // The const is initialized here to workaround VC bug of rendering all Tag::*

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

hmm?

[arthurhsu, 2011/12/01 00:59:35]

If someone attempt to move this const array outside this function, he/she'll be
screwed in VC.  gcc does not have the problem.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

See if you can get stuart to accept a preprocessor macro definition of
GenerateTag

[arthurhsu, 2011/12/02 00:06:47]

That will be an upstream change and won't be in this roll.

+  // as 0.  These tags represents the TTF tables that we will embed in subset
+  // font.
+  const int32_t VALID_TABLE_TAG[] = {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit: TABLES_IN_SUBSET

[arthurhsu, 2011/12/01 00:59:35]

Done.

+    Tag::head, Tag::hhea, Tag::loca, Tag::maxp, Tag::cvt,
+    Tag::prep, Tag::glyf, Tag::hmtx, Tag::fpgm, Tag::EBDT,
+    Tag::EBLC, Tag::EBSC, Tag::bdat, Tag::bloc, Tag::bhed,
+    Tag::cmap,  // Keep here for future tagged PDF development.
+    Tag::name,  // Keep here due to legal concerns: copyright info inside.
+  };
+
+  // Setup font builders we need.
+  FontBuilderPtr font_builder;
+  font_builder.Attach(factory_->NewFontBuilder());
+  IntegerSet remove_tags;
+
+  GlyphTablePtr glyph_table =

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

You got these in the caller, maybe just pass them down?

[arthurhsu, 2011/12/01 00:59:35]

It's designed for upstream so that other callers are not required to do the
casting.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

I didn't mean just the cast, I meant the entire table lookup.

As I understand it, the only upstream callers are tests - they can do the extra
work instead of looking up the tables twice?

[arthurhsu, 2011/12/02 00:06:47]

Done.

+      down_cast<GlyphTable*>(font_->GetTable(Tag::glyf));
+  LocaTablePtr loca_table = down_cast<LocaTable*>(font_->GetTable(Tag::loca));
+
+  if (SetupGlyfBuilders(font_builder, glyph_table, loca_table, glyph_ids)) {
+    remove_tags.insert(Tag::glyf);
+    remove_tags.insert(Tag::loca);
+  }
+
+  int flag = DetectBitmapBuilders(font_);

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

nit flag -> bitmap_table_type

[arthurhsu, 2011/12/01 00:59:35]

Done.

+  if (flag != kNotFound) {
+    bool use_ebdt = (flag == kEBDTFound);
+    bool subset_success =
+        SetupBitmapBuilders(font_, font_builder, glyph_ids, use_ebdt);
+
+    if (use_ebdt || !subset_success) {

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

I don't understand the lack of symmetry between this line and 799? 

[arthurhsu, 2011/12/01 00:59:35]

bdat shall be gone if ebdt exists, or the bdat contains no glyph of our subset

ebdt shall be gone if ebdt exists, and the ebdt contains no glyph of our subset

+      remove_tags.insert(Tag::bdat);
+      remove_tags.insert(Tag::bloc);
+      remove_tags.insert(Tag::bhed);
+    }
+    if (use_ebdt && !subset_success) {
+      remove_tags.insert(Tag::EBDT);
+      remove_tags.insert(Tag::EBLC);
+      remove_tags.insert(Tag::EBSC);
+    }
+  }
+
+  IntegerSet allowed_tags;

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

It seems like this should be a static lazy instance.

[arthurhsu, 2011/12/01 00:59:35]

There will be one more global variable buying little performance.

[vandebo (ex-Chrome), 2011/12/01 21:50:28]

What's wrong with a class static? It seems wasteful to recompute it each time.

[arthurhsu, 2011/12/02 00:06:47]

The set is static initialized than it will stay for the process's lifetime. 
Since most people don't print, I'd opt to use more CPU's during printing than
use more memory for the process' lifetime.

+  for (size_t i = 0; i < sizeof(VALID_TABLE_TAG) / sizeof(int32_t); ++i) {
+    allowed_tags.insert(VALID_TABLE_TAG[i]);
+  }
+  for (IntegerSet::iterator i = remove_tags.begin(), e = remove_tags.end();

[vandebo (ex-Chrome), 2011/11/30 22:52:58]

Should this use set_difference?

[arthurhsu, 2011/12/01 00:59:35]

Done.

+                            i != e; i++) {
+    IntegerSet::iterator it = allowed_tags.find(*i);
+    if (it != allowed_tags.end()) {
+      allowed_tags.erase(it);
+    }
+  }
+
   // Setup remaining builders.
-  for (TableMap::const_iterator i = font_->GetTableMap()->begin(),
-                                e = font_->GetTableMap()->end(); i != e; ++i) {
-    // We already build the builder for glyph and loca.
-    if (i->first != Tag::glyf && i->first != Tag::loca) {
-      font_builder->NewTableBuilder(i->first, i->second->ReadFontData());
+  for (IntegerSet::iterator i = allowed_tags.begin(), e = allowed_tags.end();
+                            i != e; ++i) {
+    Table* table = font_->GetTable(*i);
+    if (table) {
+      font_builder->NewTableBuilder(*i, table->ReadFontData());
     }
   }