Only record one in n line endings to save space.

src/objects.cc

Index: src/objects.cc
diff --git a/src/objects.cc b/src/objects.cc
index 126ca1bdc5dd5fa75880548d64ea0249f5e29e3a..44ef1a64cd49bd0dff394df134979092fb9607f7 100644
--- a/src/objects.cc
+++ b/src/objects.cc
@@ -8901,18 +8901,62 @@ static void CalculateLineEndsImpl(Isolate* isolate,
                                   Vector<const SourceChar> src,
                                   bool include_ending_line) {
   const int src_len = src.length();
+  bool exotic_newlines = false;
+  if (include_ending_line) {
+    // Initally assume reduction is 1, ie all line endings are in the array.
+    DCHECK_EQ(line_ends->length(), Script::kReductionIndex);
+    line_ends->Add(1);
+    // Write a placeholder for the number-of-lines indicator.
+    DCHECK_EQ(line_ends->length(), Script::kNumberOfLinesIndex);
+    line_ends->Add(0);
+    DCHECK_EQ(line_ends->length(), Script::kFirstLineEndIndex);
+    // There's a fictional newline just before the first character.  This
+    // simplifies a lot of things.
+    line_ends->Add(-1);
+  }
   UnicodeCache* cache = isolate->unicode_cache();
   for (int i = 0; i < src_len - 1; i++) {
     SourceChar current = src[i];
     SourceChar next = src[i + 1];
-    if (cache->IsLineTerminatorSequence(current, next)) line_ends->Add(i);
+    if (cache->IsLineTerminatorSequence(current, next)) {
+      if (current != '\n' && current != '\r') exotic_newlines = true;
+      line_ends->Add(i);
+    }
   }
 
-  if (src_len > 0 && cache->IsLineTerminatorSequence(src[src_len - 1], 0)) {
-    line_ends->Add(src_len - 1);
+  int last_posn = src_len - 1;
+  if (last_posn >= 0 && cache->IsLineTerminatorSequence(src[last_posn], 0)) {
+    if (src[last_posn] != '\n' && src[last_posn] != '\r')
+      exotic_newlines = true;
+    line_ends->Add(last_posn);
   } else if (include_ending_line) {
-    // Even if the last line misses a line end, it is counted.
-    line_ends->Add(src_len);
+    // Even if the last line misses a line end, it is counted. Because we
+    // sometimes use character positions that are one beyond the end of the
+    // source (see Rewriter::Rewrite) we set the newline one beyond that.
+    // This is used for substr calculations, which trims to string length,
+    // so it's harmless.
+    line_ends->Add(last_posn + 1);
+  }
+  if (include_ending_line) {
+    // Update number of lines in script.
+    int lines = line_ends->length() - (Script::kFirstLineEndIndex + 1);
+    line_ends->Set(Script::kNumberOfLinesIndex, lines);
+    // Abuse some flags. The bots will run with a good variety of these flags,
+    // giving better coverage for the reduction code.
+    bool always_reduce = FLAG_always_opt;
+    bool never_reduce = !FLAG_crankshaft;
+    if (!never_reduce && !exotic_newlines &&
+        (always_reduce ||
+         (line_ends->length() > 5 && line_ends->length() * 8 > src_len / 12))) {
+      // If the line-ends array (8 bytes per entry) is larger than about 8%
+      // of the source length, then we reduce it to save memory. This won't
+      // trigger if lines are > 100 characters on average. If it triggers, then
+      // the goal is for it to take only 3% of the source size.
+      int reduction =
+          always_reduce ? 2 : (line_ends->length() * 8 * 33 / src_len);
+      DCHECK(reduction > 1);
+      line_ends->Set(Script::kReductionIndex, reduction);
+    }
   }
 }
 
@@ -8941,12 +8985,30 @@ Handle<FixedArray> String::CalculateLineEnds(Handle<String> src,
                             include_ending_line);
     }
   }
-  int line_count = line_ends.length();
-  Handle<FixedArray> array = isolate->factory()->NewFixedArray(line_count);
-  for (int i = 0; i < line_count; i++) {
-    array->set(i, Smi::FromInt(line_ends[i]));
+  if (include_ending_line) {
+    const int kReductionIndex = Script::kReductionIndex;
+    const int kFirstLineEndIndex = Script::kFirstLineEndIndex;
+    int line_count = line_ends.length() - kFirstLineEndIndex;
+    int reduction = line_ends[kReductionIndex];
+    int reduced_lines = (line_count + reduction - 1) / reduction;
+    Handle<FixedArray> array =
+        isolate->factory()->NewFixedArray(kFirstLineEndIndex + reduced_lines);
+    for (int i = 0; i < kFirstLineEndIndex; i++) {
+      array->set(i, Smi::FromInt(line_ends[i]));
+    }
+    int j = kFirstLineEndIndex;
+    for (int i = 0; i < line_count; i += reduction, ++j) {
+      array->set(j, Smi::FromInt(line_ends[i + kFirstLineEndIndex]));
+    }
+    return array;
+  } else {
+    Handle<FixedArray> array =
+        isolate->factory()->NewFixedArray(line_ends.length());
+    for (int i = 0; i < line_ends.length(); i++) {
+      array->set(i, Smi::FromInt(line_ends[i]));
+    }
+    return array;
   }
-  return array;
 }
 
 
@@ -10304,41 +10366,113 @@ void Script::InitLineEnds(Handle<Script> script) {
 }
 
 
+static int CountForwardNNewlines(Handle<Script> script, int block_position,
+                                 int n) {
+  int position = block_position;
+  Handle<Object> source_object(script->source(), script->GetIsolate());
+  if (!source_object->IsString() || n == 0) return position;
+  Handle<String> source(Handle<String>::cast(source_object));
+  int length = source->length();
+  for (int i = position; i < length; i++) {
+    uc16 current = source->Get(i);
+    if (current == '\r') {
+      n--;
+      if (i + 1 < length && source->Get(i + 1) == '\n') i++;
+    } else if (current == '\n') {
+      n--;
+    }
+    if (n == 0) return i + 1;
+  }
+  if (n == 1 && length > 0) {
+    uc16 last = source->Get(length - 1);
+    if (last != '\n' && last != '\r') return length;
+  }
+  return -1;
+}
+
+
 int Script::GetColumnNumber(Handle<Script> script, int code_pos) {
+  // Get zero-based line number.
   int line_number = GetLineNumber(script, code_pos);
   if (line_number == -1) return -1;
 
   DisallowHeapAllocation no_allocation;
   FixedArray* line_ends_array = FixedArray::cast(script->line_ends());
   line_number = line_number - script->line_offset()->value();
-  if (line_number == 0) return code_pos + script->column_offset()->value();
-  int prev_line_end_pos =
-      Smi::cast(line_ends_array->get(line_number - 1))->value();
-  return code_pos - (prev_line_end_pos + 1);
+  int reduction = Smi::cast(line_ends_array->get(kReductionIndex))->value();
+
+  int line_block_position =
+      Smi::cast(line_ends_array->get(line_number / reduction +
+                                     kFirstLineEndIndex))->value() +
+      1;
+
+  int line_position = CountForwardNNewlines(script, line_block_position,
+                                            line_number % reduction);
+  if (line_number == 0) line_position = -script->column_offset()->value();
+  return code_pos - line_position;
 }
 
 
+// Zero-based line number, calculated from UTF16 character position.
 int Script::GetLineNumberWithArray(int code_pos) {
   DisallowHeapAllocation no_allocation;
   DCHECK(line_ends()->IsFixedArray());
   FixedArray* line_ends_array = FixedArray::cast(line_ends());
   int line_ends_len = line_ends_array->length();
-  if (line_ends_len == 0) return -1;
+  if (line_ends_len == 0) return -1;  // This happens if there is no source.
+  // There's always at least one line ending: A fictional newline just before
+  // the start.
+  DCHECK_GE(line_ends_len, kFirstLineEndIndex + 1);
+  int lower = kFirstLineEndIndex;
+  int upper = line_ends_len - 1;
+
+  if (code_pos < 0) return -1;
+  int index = 0;
 
-  if ((Smi::cast(line_ends_array->get(0)))->value() >= code_pos) {
-    return line_offset()->value();
+  if (code_pos > Smi::cast(line_ends_array->get(upper))->value()) {
+    index = upper;
+  } else {
+    while (lower + 1 < upper) {
+      DCHECK_LE(Smi::cast(line_ends_array->get(lower))->value(), code_pos);
+      DCHECK_LE(code_pos, Smi::cast(line_ends_array->get(upper))->value());
+      int i = (lower + upper) >> 1;
+      DCHECK(lower != i && upper != i);
+      if ((Smi::cast(line_ends_array->get(i)))->value() >= code_pos) {
+        upper = i;
+      } else {
+        lower = i;
+      }
+    }
+    index = lower;
   }
 
-  int left = 0;
-  int right = line_ends_len;
-  while (int half = (right - left) / 2) {
-    if ((Smi::cast(line_ends_array->get(left + half)))->value() > code_pos) {
-      right -= half;
-    } else {
-      left += half;
+  int reduction = Smi::cast(line_ends_array->get(kReductionIndex))->value();
+  int line_number = (index - kFirstLineEndIndex) * reduction;
+
+  // We only saved an nth of the line ends in the array, because there were so
+  // many.
+  int start_of_earlier_line =
+      Smi::cast(line_ends_array->get(index))->value() + 1;
+
+  if (reduction == 1 || !source()->IsString()) {
+    return line_number + line_offset()->value();
+  }
+  String* src = String::cast(source());
+  // This '>' would normally be a '>=', but due to {}-less 'with' statements in
+  // top-level code we sometimes encounter code positions that are one character
+  // after the end of the source. See comment in Rewriter::Rewrite.
+  if (code_pos > src->length()) return -1;
+  for (int i = start_of_earlier_line; i < src->length() && i < code_pos; i++) {
+    uc16 current = src->Get(i);
+    if (current == '\r') {
+      if (i < code_pos - 1 && i < src->length() - 1 && src->Get(i + 1) == '\n')
+        i++;
+      line_number++;
+    } else if (current == '\n') {
+      line_number++;
     }
   }
-  return right + line_offset()->value();
+  return line_number + line_offset()->value();
 }