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

Index: components/translate/core/language_detection/language_detection_util.cc
diff --git a/components/translate/core/language_detection/language_detection_util.cc b/components/translate/core/language_detection/language_detection_util.cc
index 3b8bda76735e02b4023709e7775f60e0bd7432f3..66d8e65d7c40e08bd1f63f1dd2670a3fe1d0a7ab 100644
--- a/components/translate/core/language_detection/language_detection_util.cc
+++ b/components/translate/core/language_detection/language_detection_util.cc
@@ -8,6 +8,7 @@
 
 #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"
@@ -136,6 +137,12 @@ std::string DetermineTextLanguage(const base::string16& text,
 
   // 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
@@ -152,7 +159,7 @@ std::string DetermineTextLanguage(const base::string16& text,
   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.