clang-format produced code that (choose all that apply): 
- Riles my finely honed stylistic dander
- No sane human would ever choose

Here's the code before formatting:
struct {
    const wchar_t* text;
    const wchar_t* display_text;
    const bool elision_expected;
  } cases[] = {
    // Strings shorter than the elision width should be laid out in full.
    { L"",        L""       , false },
    { L"M",       L""       , false },
    { L" . ",     L" . "    , false },
    { kWeak,      kWeak     , false },
    { kLtr,       kLtr      , false },
    { kLtrRtl,    kLtrRtl   , false },
    { kLtrRtlLtr, kLtrRtlLtr, false },
    { kRtl,       kRtl      , false },
    { kRtlLtr,    kRtlLtr   , false },
    { kRtlLtrRtl, kRtlLtrRtl, false },
    // Strings as long as the elision width should be laid out in full.
    { L"012ab",   L"012ab"  , false },
    // Long strings should be elided with an ellipsis appended at the end.
    { L"012abc",              L"012a\x2026", true },
    { L"012ab" L"\x5d0\x5d1", L"012a\x2026", true },
    { L"012a" L"\x5d1" L"b",  L"012a\x2026", true },
    // No RLM marker added as digits (012) have weak directionality.
    { L"01" L"\x5d0\x5d1\x5d2", L"01\x5d0\x5d1\x2026", true },
    // RLM marker added as "ab" have strong LTR directionality.
    { L"ab" L"\x5d0\x5d1\x5d2", L"ab\x5d0\x5d1\x2026\x200f", true },
    // Test surrogate pairs. \xd834\xdd1e forms a single code point U+1D11E;
    // \xd834\xdd22 forms a second code point U+1D122. The first should be kept;
    // the second removed (no surrogate pair should be partially elided).
    { L"0123\xd834\xdd1e\xd834\xdd22", L"0123\xd834\xdd1e\x2026", true },
    // Test combining character sequences. U+0915 U+093F forms a compound glyph;
    // U+0915 U+0942 forms a second compound glyph. The first should be kept;
    // the second removed (no combining sequence should be partially elided).
    { L"0123\x0915\x093f\x0915\x0942", L"0123\x0915\x093f\x2026", true },
    // U+05E9 U+05BC U+05C1 U+05B8 forms a four-character compound glyph. Again,
    // it should be either fully elided, or not elided at all. If completely
    // elided, an LTR Mark (U+200E) should be added.
    { L"0\x05e9\x05bc\x05c1\x05b8",   L"0\x05e9\x05bc\x05c1\x05b8", false },
    { L"0\x05e9\x05bc\x05c1\x05b8",   L"0\x2026\x200E"            , true  },
    { L"01\x05e9\x05bc\x05c1\x05b8",  L"01\x2026\x200E"           , true  },
    { L"012\x05e9\x05bc\x05c1\x05b8", L"012\x2026\x200E"          , true  },
    { L"012\xF0\x9D\x84\x9E",         L"012\xF0\x2026"            , true  },
    { L"012\xF0\x9D\x84\x9E", L"012\x2026", true},
  };

Here's the code after formatting:
  struct {
    const wchar_t* text;
    const wchar_t* display_text;
    const bool elision_expected;
  } cases[] = {
      // Strings shorter than the elision width should be laid out in full.
      {L"", L"", false},
      {L"M", L"", false},
      {L" . ", L" . ", false},
      {kWeak, kWeak, false},
      {kLtr, kLtr, false},
      {kLtrRtl, kLtrRtl, false},
      {kLtrRtlLtr, kLtrRtlLtr, false},
      {kRtl, kRtl, false},
      {kRtlLtr, kRtlLtr, false},
      {kRtlLtrRtl, kRtlLtrRtl, false},
      // Strings as long as the elision width should be laid out in full.
      {L"012ab", L"012ab", false},
      // Long strings should be elided with an ellipsis appended at the end.
      {L"012abc", L"012a\x2026", true},
      {L"012ab"
       L"\x5d0\x5d1",
       L"012a\x2026", true},
      {L"012a"
       L"\x5d1"
       L"b",
       L"012a\x2026", true},
      // No RLM marker added as digits (012) have weak directionality.
      {L"01"
       L"\x5d0\x5d1\x5d2",
       L"01\x5d0\x5d1\x2026", true},
      // RLM marker added as "ab" have strong LTR directionality.
      {L"ab"
       L"\x5d0\x5d1\x5d2",
       L"ab\x5d0\x5d1\x2026\x200f", true},
      // Test surrogate pairs. \xd834\xdd1e forms a single code point U+1D11E;
      // \xd834\xdd22 forms a second code point U+1D122. The first should be
      // kept;
      // the second removed (no surrogate pair should be partially elided).
      {L"0123\xd834\xdd1e\xd834\xdd22", L"0123\xd834\xdd1e\x2026", true},
      // Test combining character sequences. U+0915 U+093F forms a compound
      // glyph;
      // U+0915 U+0942 forms a second compound glyph. The first should be kept;
      // the second removed (no combining sequence should be partially elided).
      {L"0123\x0915\x093f\x0915\x0942", L"0123\x0915\x093f\x2026", true},
      // U+05E9 U+05BC U+05C1 U+05B8 forms a four-character compound glyph.
      // Again,
      // it should be either fully elided, or not elided at all. If completely
      // elided, an LTR Mark (U+200E) should be added.
      {L"0\x05e9\x05bc\x05c1\x05b8", L"0\x05e9\x05bc\x05c1\x05b8", false},
      {L"0\x05e9\x05bc\x05c1\x05b8", L"0\x2026\x200E", true},
      {L"01\x05e9\x05bc\x05c1\x05b8", L"01\x2026\x200E", true},
      {L"012\x05e9\x05bc\x05c1\x05b8", L"012\x2026\x200E", true},
      // \xF0\x9D\x84\x9E is the UTF-8 bytestring for MUSICAL SYMBOL G CLEF.  It
      // should not have any internal grapheme boundaries; it should be
      // completely removed when eliding.
      {L"012\xF0\x9D\x84\x9E", L"012\x2026", true},
  };

Here's how it ought to look:
It sounds remain unchanged.

Code review link for full files/context:
https://chromium-review.googlesource.com/c/611789/6/ui/gfx/render_text_unittest.cc#979
[ note: left side in that line is missing the file line.  I added the final line and then ran git cl format and it revised the whole array in this ugly manner that makes it harder to read.]