Add histograms to measure CLD2 language detection as well as the accuracy of the chosen detection.

components/translate/core/language_detection/language_detection_util.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "components/translate/core/language_detection/language_detection_util.h"
 
 #include <stddef.h>
 
 #include "base/logging.h"
 #include "base/macros.h"
+#include "base/metrics/histogram_macros.h"
 #include "base/strings/string_split.h"
 #include "base/strings/string_util.h"
 #include "base/strings/utf_string_conversions.h"
 #include "base/time/time.h"
 #include "components/translate/core/common/translate_constants.h"
 #include "components/translate/core/common/translate_metrics.h"
 #include "components/translate/core/common/translate_util.h"
 
 #if CLD_VERSION==1
 #include "third_party/cld/encodings/compact_lang_det/compact_lang_det.h"
@@ skipping 108 matching lines @@
         raw_utf8_bytes, num_utf8_bytes, is_plain_text, &cldhints, flags,
         language3, percent3, normalized_score3,
         nullptr /* No ResultChunkVector used */, &text_bytes, &is_reliable);
   }
   is_valid_language = cld_language != CLD2::NUM_LANGUAGES &&
                       cld_language != CLD2::UNKNOWN_LANGUAGE &&
                       cld_language != CLD2::TG_UNKNOWN_LANGUAGE;
 
   // Choose top language.
   cld_language = language3[0];
+  UMA_HISTOGRAM_ENUMERATION("Translate.CLD2LanguageDetected",
+                            cld_language, CLD2::NUM_LANGUAGES);

[Alexei Svitkine (slow), 2016/03/31 15:10:21]

Does CLD guarantee that it's language enums won't be renumbered in a new version
of the library?

[rkaplow, 2016/03/31 15:26:14]

I suspect only the unused space (i.e. x_nnn) would be replaced with actual
languages. It seems pretty unlikely to me they would do reordering (also doesn't
look like there's much more development)

Not sure what you would suggest - I could add a comment here to warn that a new
histogram would need to be added if the enum is reordered but that seems
overkill.

+  if (is_valid_language) {

[Alexei Svitkine (slow), 2016/03/31 15:10:22]

Nit: No {}'s

[rkaplow, 2016/03/31 15:26:14]

Done.

+    UMA_HISTOGRAM_PERCENTAGE("Translate.CLD2LanguageAccuracy", percent3[0]);
+  }
+
 #else
 # error "CLD_VERSION must be 1 or 2"
 #endif
 
   if (is_cld_reliable != NULL)
     *is_cld_reliable = is_reliable;
 
   // We don't trust the result if the CLD reports that the detection is not
   // reliable, or if the actual text used to detect the language was less than
   // 100 bytes (short texts can often lead to wrong results).
   // TODO(toyoshim): CLD provides |is_reliable| flag. But, it just says that
   // the determined language code is correct with 50% confidence. Chrome should
   // handle the real confidence value to judge.
   if (is_reliable && num_bytes_evaluated >= 100 && is_valid_language) {
     // We should not use LanguageCode_ISO_639_1 because it does not cover all
     // the languages CLD can detect. As a result, it'll return the invalid
-    // language code for tradtional Chinese among others.
+    // language code for traditional Chinese among others.
     // |LanguageCodeWithDialect| will go through ISO 639-1, ISO-639-2 and
     // 'other' tables to do the 'right' thing. In addition, it'll return zh-CN
     // for Simplified Chinese.
 #if CLD_VERSION==1
     language = LanguageCodeWithDialects(static_cast<Language>(cld_language));
 #elif CLD_VERSION==2
     // (1) CLD2's LanguageCode returns general Chinese 'zh' for
     // CLD2::CHINESE, but Translate server doesn't accept it. This is
     // converted to 'zh-CN' in the same way as CLD1's
     // LanguageCodeWithDialects.
@@ skipping 219 matching lines @@
   // distinguish from English, and the language is one of well-known languages
   // which often provide "en-*" meta information mistakenly.
   for (size_t i = 0; i < arraysize(kWellKnownCodesOnWrongConfiguration); ++i) {
     if (cld_language == kWellKnownCodesOnWrongConfiguration[i])
       return true;
   }
   return false;
 }
 
 }  // namespace translate