Say you have a test for HarfBuzz run lists:

  const base::string16 mixed = WideToUTF16(
      L"\x05D0\x05D1"
      L"1234"
      L"\x05D2\x05D3"
      L"abc");
  render_text->SetText(mixed);
  render_text->SetDirectionalityMode(DIRECTIONALITY_FORCE_LTR);
  test_api()->EnsureLayout();
  internal::TextRunList* run_list = GetHarfBuzzRunList();

What if I told you we could take expectations like this:

  ASSERT_EQ(4U, run_list->size());
  EXPECT_TRUE(run_list->runs()[0]->is_rtl);
  EXPECT_FALSE(run_list->runs()[1]->is_rtl);
  EXPECT_TRUE(run_list->runs()[2]->is_rtl);
  EXPECT_FALSE(run_list->runs()[3]->is_rtl);
  EXPECT_EQ(2U, run_list->logical_to_visual(0));
  EXPECT_EQ(1U, run_list->logical_to_visual(1));
  EXPECT_EQ(0U, run_list->logical_to_visual(2));
  EXPECT_EQ(3U, run_list->logical_to_visual(3));

and rewrite them like this:

  EXPECT_EQ("[7<-6][2->5][1<-0][8->10]", RunListDebugString(*run_list));

By making a TextRunList into a human-readable string then comparing it in the test, we get:

- More compact tests: Instead of two lines of code for each run, one line for the whole runlist.
- More readable tests: You can actually see the author's intent.
- More readable failure messages: Shows the expected and actual runlist string, rather than individual numerical failures.
- Tests more: Some of these tests *only* test the order of runs (not their ranges). Other tests *only* test the ranges (not the order). The proposed style tests both in one.

The debug string shows the runs in visual order. Each run is enclosed in square brackets, and shows the begin and end inclusive logical character position, with an arrow indicating the direction of the run. Single-character runs just show the character position.

For example, the corresponding run-list for the string "abc+אבג" is expressed as "[0->2][3][6<-4]".

So, that's the plan.