Reland "Normalize Katakana to Hiragana in ScriptRunIterator"

This is a reland of 05672d8d

The original change was reverted as suspected to cause
Android CFI builder to fail:

[ RUN      ] RenderViewImplTest.PreferredSizeZoomed
[FATAL:data_pack.cc(444)] Check failed: !handle->HasResource(resource_id). Duplicate resource 25400 with scale 1

I cannot find any relationship to the failure, and
android_cfi_rel_ng trybot passes. Relanding to see if this
was really the cause.

Original change's description:
> Normalize Katakana to Hiragana in ScriptRunIterator
>
> This patch normalizes Katakana to Hiragana in ScriptData, and
> thus in ScriptRunIterator.
>
> There are 3 ICU script code for kana; USCRIPT_HIRAGANA,
> USCRIPT_KATAKANA, and USCRIPT_KATAKANA_OR_HIRAGANA. However,
> OpenType has only one 'kana' for all these 3 script codes.
>
> When shaping, HarfBuzz handles the normalization. However,
> Blink splits the incoming string by the ICU scripts.
> Splitting these 3 scripts is not useful but consumes CPU
> and memory.
>
> By normalizing in ScriptRunIterator, all types of Kana are
> in single run.
>
> timeToFirstContentfulPaint:layout shows ~3% improvements
> https://pinpoint-dot-chromeperf.appspot.com/job/10d9760fe40000
>
> though bink_perf.layout does not show much differences
> https://pinpoint-dot-chromeperf.appspot.com/job/14f19c00140000
>
> This is probably because Katakana characters are not used
> much in `japanese-kokoro-insert` (written in ~100 years ago,)
> while they are more used in modern pages.
>
> Bug: 636993
> Change-Id: If692bb575f2232b8fca10cd2ea87e3022debdf05
> Reviewed-on: https://chromium-review.googlesource.com/c/1337650
> Commit-Queue: Koji Ishii <kojii@chromium.org>
> Reviewed-by: Dominik Röttsches <drott@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#609119}

TBR=drott@chromium.org

Bug: 636993
Change-Id: I1d741fa47556c5e7bf47e3badcdcf60a81fbc22b
Reviewed-on: https://chromium-review.googlesource.com/c/1343583Reviewed-by: Koji Ishii <kojii@chromium.org>
Commit-Queue: Koji Ishii <kojii@chromium.org>
Cr-Commit-Position: refs/heads/master@{#609680}

third_party/blink/renderer/platform/fonts/script_run_iterator.cc

@@ -10,6 +10,26 @@
 
 namespace blink {
 
+namespace {
+
+// UScriptCode and OpenType script are not 1:1; specifically, both Hiragana and
+// Katakana map to 'kana' in OpenType. They will be mapped correctly in
+// HarfBuzz, but normalizing earlier helps to reduce splitting runs between
+// these scripts.
+// https://docs.microsoft.com/en-us/typography/opentype/spec/scripttags
+inline UScriptCode getScriptForOpenType(UChar32 ch, UErrorCode* status) {
+  UScriptCode script = uscript_getScript(ch, status);
+  if (UNLIKELY(U_FAILURE(*status)))
+    return script;
+  if (UNLIKELY(script == USCRIPT_KATAKANA ||
+               script == USCRIPT_KATAKANA_OR_HIRAGANA)) {
+    return USCRIPT_HIRAGANA;
+  }
+  return script;
+}
+
+}  // namespace
+
 typedef ScriptData::PairedBracketType PairedBracketType;
 
 constexpr int ScriptRunIterator::kMaxScriptCount;
@@ -27,6 +47,10 @@ void ICUScriptData::GetScripts(UChar32 ch, UScriptCodeList& dst) const {
   // regardless of the capacity passed to the call. So count can be greater
   // than dst->size(), if a later version of the unicode data has more
   // than kMaxScriptCount items.
+
+  // |uscript_getScriptExtensions| do not need to be collated to
+  // USCRIPT_HIRAGANA because when ScriptExtensions contains Kana, it contains
+  // Hira as well, and Hira is always before Kana.
   int count = uscript_getScriptExtensions(ch, &dst[0], dst.size(), &status);
   if (status == U_BUFFER_OVERFLOW_ERROR) {
     // Allow this, we'll just use what we have.
@@ -35,7 +59,7 @@ void ICUScriptData::GetScripts(UChar32 ch, UScriptCodeList& dst) const {
     count = dst.size();
     status = U_ZERO_ERROR;
   }
-  UScriptCode primary_script = uscript_getScript(ch, &status);
+  UScriptCode primary_script = getScriptForOpenType(ch, &status);
 
   if (U_FAILURE(status)) {
     DLOG(ERROR) << "Could not get icu script data: " << status << " for 0x"

third_party/blink/renderer/platform/fonts/script_run_iterator_test.cc

@@ -370,6 +370,50 @@ TEST_F(ScriptRunIteratorTest, Chinese) {
   CHECK_SCRIPT_RUNS({{"萬國碼", USCRIPT_HAN}});
 }
 
+struct JapaneseMixedScript {
+  const char* string;
+  // The expected primary_script when the string alone was evaluated.
+  UScriptCode script;
+} japanese_mixed_scripts[] = {{"あ", USCRIPT_HIRAGANA},
+                              // Katakana should be normalized to Hiragana
+                              {"ア", USCRIPT_HIRAGANA},
+                              // Script_Extensions=Hira Kana
+                              {"\u30FC", USCRIPT_HIRAGANA},
+                              // Script_Extensions=Hani Hira Kana
+                              {"\u303C", USCRIPT_HAN},
+                              // Script_Extensions=Bopo Hang Hani Hira Kana
+                              {"\u3003", USCRIPT_BOPOMOFO},
+                              // Script_Extensions=Bopo Hang Hani Hira Kana Yiii
+                              {"\u3001", USCRIPT_BOPOMOFO}};
+
+class JapaneseMixedScriptTest
+    : public ScriptRunIteratorTest,
+      public testing::WithParamInterface<JapaneseMixedScript> {};
+
+INSTANTIATE_TEST_CASE_P(ScriptRunIteratorTest,
+                        JapaneseMixedScriptTest,
+                        testing::ValuesIn(japanese_mixed_scripts));
+
+TEST_P(JapaneseMixedScriptTest, Data) {
+  const auto& data = GetParam();
+  std::string string(data.string);
+
+  CheckRuns({{string.data(), data.script}});
+
+  // If the string follows Hiragana or Katakana, or is followed by Hiragnaa or
+  // Katakana, it should be normalized as Hiragana.
+  std::string hiragana("か");
+  std::string katakana("カ");
+  CheckRuns({{(hiragana + string).data(), USCRIPT_HIRAGANA}});
+  CheckRuns({{(string + hiragana).data(), USCRIPT_HIRAGANA}});
+
+  CheckRuns({{(katakana + string).data(), USCRIPT_HIRAGANA}});
+  CheckRuns({{(string + katakana).data(), USCRIPT_HIRAGANA}});
+
+  CheckRuns({{(hiragana + string + katakana).data(), USCRIPT_HIRAGANA}});
+  CheckRuns({{(katakana + string + hiragana).data(), USCRIPT_HIRAGANA}});
+}
+
 // Close bracket without matching open is ignored
 TEST_F(ScriptRunIteratorTest, UnbalancedParens1) {
   CHECK_SCRIPT_RUNS(