[subset] sketch out dropping in-glyf instructions. Sometimes yields differnet size glyphs than fonttools, possibly due to padding not being corrected

src/hb-ot-glyf-table.hh

@@ -301,6 +301,51 @@ struct glyf
       return true;
     }
 
+    inline bool get_instruction_offsets(unsigned int start_offset,
+                                        unsigned int end_offset,
+                                        unsigned int *instruction_start /* OUT */,
+                                        unsigned int *instruction_end /* OUT */) const
+    {
+      if (end_offset - start_offset < GlyphHeader::static_size)
+      {
+        *instruction_start = 0;
+        *instruction_end = 0;
+        return true; /* Empty glyph; no instructions. */
+      }
+      const GlyphHeader &glyph_header = StructAtOffset<GlyphHeader> (glyf_table, start_offset);
+      int16_t num_contours = (int16_t) glyph_header.numberOfContours;
+      if (num_contours < 0)
+      {
+        CompositeGlyphHeader::Iterator *composite_it;
+        if (unlikely (!CompositeGlyphHeader::get_iterator (
+            (const char*) this->glyf_table + start_offset,
+             end_offset - start_offset, composite_it))) return false;
+        const CompositeGlyphHeader *last;
+        do {
+          last = composite_it->current;
+        } while (composite_it->move_to_next());
+
+        if ( (uint16_t) last->flags & CompositeGlyphHeader::WE_HAVE_INSTRUCTIONS)
+          *instruction_start = start_offset + ((char *) last - (char *) glyf_table->dataX) + last->get_size();
+        else
+          *instruction_start = end_offset;
+        *instruction_end = end_offset;
+        if (unlikely (*instruction_start > *instruction_end))
+        {
+          DEBUG_MSG(SUBSET, nullptr, "Invalid instruction offset, %d is outside [%d, %d]", *instruction_start, start_offset, end_offset);
+          return false;
+        }
+      }
+      else
+      {
+        unsigned int instruction_length_offset = start_offset + GlyphHeader::static_size + 2 * num_contours;
+        const HBUINT16 &instruction_length = StructAtOffset<HBUINT16> (glyf_table, instruction_length_offset);
+        *instruction_start = instruction_length_offset + 2;
+        *instruction_end = *instruction_start + (uint16_t) instruction_length;
+      }
+      return true;
+    }
+
     inline bool get_extents (hb_codepoint_t glyph,
 			     hb_glyph_extents_t *extents) const
     {

src/hb-subset-glyf.cc

@@ -21,7 +21,7 @@
  * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
  * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
  *
- * Google Author(s): Garret Rieger
+ * Google Author(s): Garret Rieger, Roderick Sheeter
  */
 
 #include "hb-open-type-private.hh"
@@ -33,26 +33,42 @@
 static bool
 _calculate_glyf_and_loca_prime_size (const OT::glyf::accelerator_t &glyf,
                                      hb_prealloced_array_t<hb_codepoint_t> &glyph_ids,
-                                     bool *use_short_loca, /* OUT */
+                                     hb_bool_t drop_hints,
-                                     unsigned int *glyf_size, /* OUT */
+                                     bool *use_short_loca /* OUT */,
-                                     unsigned int *loca_size /* OUT */)
+                                     unsigned int *glyf_size /* OUT */,
+                                     unsigned int *loca_size /* OUT */,
+                                     hb_prealloced_array_t<unsigned int> *instruction_ranges /* OUT */)
 {
   unsigned int total = 0;
-  unsigned int count = 0;
   for (unsigned int i = 0; i < glyph_ids.len; i++)
   {
     hb_codepoint_t next_glyph = glyph_ids[i];
     unsigned int start_offset, end_offset;
-    if (unlikely (!glyf.get_offsets (next_glyph, &start_offset, &end_offset)))
+    if (unlikely (!glyf.get_offsets(next_glyph, &start_offset, &end_offset)))
-      end_offset = start_offset = 0;
+    {
+      DEBUG_MSG(SUBSET, nullptr, "Invalid gid %d", next_glyph);
+      continue;
+    }
+    unsigned int instruction_start = 0, instruction_end = 0;
+    if (drop_hints)
+    {
+      if (unlikely (!glyf.get_instruction_offsets(start_offset, end_offset,
+                                                  &instruction_start, &instruction_end)))
+      {
+        DEBUG_MSG(SUBSET, nullptr, "Unable to get instruction offsets for %d", next_glyph);
+        return false;
+      }
+    }
+    *(instruction_ranges->push()) = instruction_start;
+    *(instruction_ranges->push()) = instruction_end;
 
-    total += end_offset - start_offset;
+    total += end_offset - start_offset - (instruction_end - instruction_start);
-    count++;
+    fprintf(stderr, "  %d ends at %d (was %d to %d, remove %d)\n", next_glyph, total, start_offset, end_offset, instruction_end - instruction_start);
   }
 
   *glyf_size = total;
   *use_short_loca = (total <= 131070);
-  *loca_size = (count + 1)
+  *loca_size = (glyph_ids.len + 1)
       * (*use_short_loca ? sizeof(OT::HBUINT16) : sizeof(OT::HBUINT32));
 
   DEBUG_MSG(SUBSET, nullptr, "preparing to subset glyf: final size %d, loca size %d, using %s loca",
@@ -118,6 +134,7 @@ _write_glyf_and_loca_prime (hb_subset_plan_t              *plan,
 			    const OT::glyf::accelerator_t &glyf,
                             const char                    *glyf_data,
                             bool                           use_short_loca,
+                            hb_prealloced_array_t<unsigned int> &instruction_ranges,
                             unsigned int                   glyf_prime_size,
                             char                          *glyf_prime_data /* OUT */,
                             unsigned int                   loca_prime_size,
@@ -132,25 +149,44 @@ _write_glyf_and_loca_prime (hb_subset_plan_t              *plan,
     unsigned int start_offset, end_offset;
     if (unlikely (!glyf.get_offsets (glyph_ids[i], &start_offset, &end_offset)))
       end_offset = start_offset = 0;
+    unsigned int instruction_start = instruction_ranges[i * 2];
+    unsigned int instruction_end = instruction_ranges[i * 2 + 1];
 
-    int length = end_offset - start_offset;
+    int length = end_offset - start_offset - (instruction_end - instruction_start);
 
     if (glyf_prime_data_next + length > glyf_prime_data + glyf_prime_size)
     {
       DEBUG_MSG (SUBSET,
                  nullptr,
-                 "WARNING: Attempted to write an out of bounds glyph entry for gid %d",
+                 "WARNING: Attempted to write an out of bounds glyph entry for gid %d (length %d)",
-                 i);
+                 i, length);
       return false;
     }
+
+    if (instruction_start == instruction_end)
+    {
+      fprintf(stderr, "i=%d copy %d bytes from %d to %ld\n", i, length, start_offset, glyf_prime_data_next - glyf_prime_data);
       memcpy (glyf_prime_data_next, glyf_data + start_offset, length);
+    }
+    else
+    {
+      fprintf(stderr, "i=%d copy %d bytes from %d to %ld\n", i, instruction_start - start_offset, start_offset, glyf_prime_data_next - glyf_prime_data);
+      fprintf(stderr, "i=%d copy %d bytes from %d to %ld\n", i, end_offset - instruction_end, instruction_end, glyf_prime_data_next + instruction_start - start_offset - glyf_prime_data);
+      memcpy (glyf_prime_data_next, glyf_data + start_offset, instruction_start - start_offset);
+      memcpy (glyf_prime_data_next + instruction_start - start_offset, glyf_data + instruction_end, end_offset - instruction_end);
+      /* if the instructions end at the end this was a composite glyph */
+      if (instruction_end == end_offset)
+        ; // TODO(rsheeter) remove WE_HAVE_INSTRUCTIONS from last flags
+      else
+        ; // TODO(rsheeter) zero instructionLength
+    }
 
     success = success && _write_loca_entry (i,
                                             glyf_prime_data_next - glyf_prime_data,
                                             use_short_loca,
                                             loca_prime_data,
                                             loca_prime_size);
-    _update_components (plan, glyf_prime_data_next, end_offset - start_offset);
+    _update_components (plan, glyf_prime_data_next, length);
 
     glyf_prime_data_next += length;
   }
@@ -176,12 +212,17 @@ _hb_subset_glyf_and_loca (const OT::glyf::accelerator_t  &glyf,
 
   unsigned int glyf_prime_size;
   unsigned int loca_prime_size;
+  hb_prealloced_array_t<unsigned int> instruction_ranges;
+  instruction_ranges.init();
 
   if (unlikely (!_calculate_glyf_and_loca_prime_size (glyf,
                                                       glyphs_to_retain,
+                                                      plan->drop_hints,
                                                       use_short_loca,
                                                       &glyf_prime_size,
-                                                      &loca_prime_size))) {
+                                                      &loca_prime_size,
+                                                      &instruction_ranges))) {
+    instruction_ranges.finish();
     return false;
   }
 
@@ -189,12 +230,15 @@ _hb_subset_glyf_and_loca (const OT::glyf::accelerator_t  &glyf,
   char *loca_prime_data = (char *) malloc (loca_prime_size);
   if (unlikely (!_write_glyf_and_loca_prime (plan, glyf, glyf_data,
                                              *use_short_loca,
+                                             instruction_ranges,
                                              glyf_prime_size, glyf_prime_data,
                                              loca_prime_size, loca_prime_data))) {
     free (glyf_prime_data);
     free (loca_prime_data);
+    instruction_ranges.finish();
     return false;
   }
+  instruction_ranges.finish();
 
   *glyf_prime = hb_blob_create (glyf_prime_data,
                                 glyf_prime_size,

src/hb-subset-input.cc

@@ -105,3 +105,9 @@ hb_subset_input_glyph_set (hb_subset_input_t *subset_input)
 {
   return subset_input->glyphs;
 }
+
+HB_EXTERN hb_bool_t *
+hb_subset_input_drop_hints(hb_subset_input_t *subset_input)
+{
+  return &subset_input->drop_hints;
+}

src/hb-subset.h

@@ -68,6 +68,8 @@ hb_subset_input_unicode_set (hb_subset_input_t *subset_input);
 HB_EXTERN hb_set_t *
 hb_subset_input_glyph_set (hb_subset_input_t *subset_input);
 
+HB_EXTERN hb_bool_t *
+hb_subset_input_drop_hints(hb_subset_input_t *subset_input);
 
 /* hb_subset() */
 

test/api/hb-subset-test.h

@@ -95,17 +95,20 @@ hb_subset_test_open_font (const char *font_path)
   return NULL; /* Shut up, compiler! */
 }
 
+static inline hb_subset_input_t *
+hb_subset_test_create_input(const hb_set_t  *codepoints)
+{
+  hb_subset_input_t *input = hb_subset_input_create_or_fail ();
+  hb_set_t * input_codepoints = hb_subset_input_unicode_set (input);
+  hb_set_union (input_codepoints, codepoints);
+  return input;
+}
+
 static inline hb_face_t *
 hb_subset_test_create_subset (hb_face_t *source,
-                              const hb_set_t  *codepoints)
+                              hb_subset_input_t *input)
 {
   hb_subset_profile_t *profile = hb_subset_profile_create();
-  hb_subset_input_t *input = hb_subset_input_create_or_fail ();
-
-  hb_set_t * input_codepoints = hb_subset_input_unicode_set (input);
-
-  hb_set_union (input_codepoints, codepoints);
-
   hb_face_t *subset = hb_subset (source, profile, input);
   g_assert (subset);
 

test/api/test-subset-cmap.c

@@ -40,7 +40,7 @@ test_subset_cmap (void)
   hb_set_t *codepoints = hb_set_create ();
   hb_set_add (codepoints, 97);
   hb_set_add (codepoints, 99);
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   hb_subset_test_check (face_ac, face_abc_subset, HB_TAG ('c','m','a','p'));
@@ -59,7 +59,7 @@ test_subset_cmap_noop (void)
   hb_set_add (codepoints, 97);
   hb_set_add (codepoints, 98);
   hb_set_add (codepoints, 99);
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   hb_subset_test_check (face_abc, face_abc_subset, HB_TAG ('c','m','a','p'));

test/api/test-subset-glyf.c

@@ -40,7 +40,7 @@ test_subset_glyf (void)
   hb_set_t *codepoints = hb_set_create();
   hb_set_add (codepoints, 97);
   hb_set_add (codepoints, 99);
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   hb_subset_test_check (face_ac, face_abc_subset, HB_TAG ('g','l','y','f'));
@@ -60,7 +60,7 @@ test_subset_glyf_with_components (void)
 
   hb_set_t *codepoints = hb_set_create();
   hb_set_add (codepoints, 0x1fc);
-  hb_face_t *face_generated_subset = hb_subset_test_create_subset (face_components, codepoints);
+  hb_face_t *face_generated_subset = hb_subset_test_create_subset (face_components, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   hb_subset_test_check (face_subset, face_generated_subset, HB_TAG ('g','l','y','f'));
@@ -81,7 +81,7 @@ test_subset_glyf_noop (void)
   hb_set_add (codepoints, 97);
   hb_set_add (codepoints, 98);
   hb_set_add (codepoints, 99);
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   hb_subset_test_check (face_abc, face_abc_subset, HB_TAG ('g','l','y','f'));
@@ -91,6 +91,31 @@ test_subset_glyf_noop (void)
   hb_face_destroy (face_abc);
 }
 
+static void
+test_subset_glyf_strip_hints (void)
+{
+  hb_face_t *face_abc = hb_subset_test_open_font ("fonts/Roboto-Regular.abc.ttf");
+  hb_face_t *face_ac = hb_subset_test_open_font ("fonts/Roboto-Regular.ac.nohints.ttf");
+
+  hb_set_t *codepoints = hb_set_create();
+  hb_set_add (codepoints, 'a');
+  hb_set_add (codepoints, 'c');
+  hb_subset_input_t *input = hb_subset_test_create_input (codepoints);
+  *hb_subset_input_drop_hints(input) = true;
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, input);
+  hb_set_destroy (codepoints);
+
+  hb_subset_test_check (face_ac, face_abc_subset, HB_TAG ('g','l','y','f'));
+  hb_subset_test_check (face_ac, face_abc_subset, HB_TAG ('l','o','c', 'a'));
+  hb_subset_test_check (face_ac, face_abc_subset, HB_TAG ('m','a','x', 'p'));
+
+  hb_face_destroy (face_abc_subset);
+  hb_face_destroy (face_abc);
+  hb_face_destroy (face_ac);
+}
+
+// TODO(rsheeter): test strip hints from composite
+
 // TODO(grieger): test for long loca generation.
 
 int
@@ -101,6 +126,7 @@ main (int argc, char **argv)
   hb_test_add (test_subset_glyf);
   hb_test_add (test_subset_glyf_with_components);
   hb_test_add (test_subset_glyf_noop);
+  hb_test_add (test_subset_glyf_strip_hints);
 
   return hb_test_run();
 }

test/api/test-subset-hmtx.c

@@ -55,7 +55,7 @@ test_subset_hmtx_simple_subset (void)
   hb_set_t *codepoints = hb_set_create ();
   hb_set_add (codepoints, 'a');
   hb_set_add (codepoints, 'c');
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   check_num_hmetrics(face_abc_subset, 3); /* nothing has same width */
@@ -76,7 +76,7 @@ test_subset_hmtx_monospace (void)
   hb_set_t *codepoints = hb_set_create ();
   hb_set_add (codepoints, 'a');
   hb_set_add (codepoints, 'c');
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   check_num_hmetrics(face_abc_subset, 1); /* everything has same width */
@@ -97,7 +97,7 @@ test_subset_hmtx_keep_num_metrics (void)
   hb_set_t *codepoints = hb_set_create ();
   hb_set_add (codepoints, 'a');
   hb_set_add (codepoints, 'c');
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   check_num_hmetrics(face_abc_subset, 3); /* c is wider */
@@ -117,7 +117,7 @@ test_subset_hmtx_decrease_num_metrics (void)
   hb_set_t *codepoints = hb_set_create ();
   hb_set_add (codepoints, 'a');
   hb_set_add (codepoints, 'b');
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   check_num_hmetrics(face_abc_subset, 1); /* everything left has same width */
@@ -137,7 +137,7 @@ test_subset_hmtx_noop (void)
   hb_set_add (codepoints, 'a');
   hb_set_add (codepoints, 'b');
   hb_set_add (codepoints, 'c');
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   check_num_hmetrics(face_abc_subset, 4); /* nothing has same width */

test/api/test-subset-os2.c

@@ -37,7 +37,7 @@ test_subset_os2 (void)
 
   hb_set_t *codepoints = hb_set_create();
   hb_set_add (codepoints, 98);
-  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, codepoints);
+  hb_face_t *face_abc_subset = hb_subset_test_create_subset (face_abc, hb_subset_test_create_input (codepoints));
   hb_set_destroy (codepoints);
 
   hb_subset_test_check (face_b, face_abc_subset, HB_TAG ('O','S','/','2'));