Faster case-insensitive comparisons. (#837)

* Faster case-insensitive comparisons.

include/simdjson/document.h

@@ -313,6 +313,7 @@ public:
 
   /**
    * Get the value associated with the given key in a case-insensitive manner.
+   * It is only guaranteed to work over ASCII inputs.
    *
    * Note: The key will be matched against **unescaped** JSON.
    *

include/simdjson/inline/document.h

@@ -686,10 +686,10 @@ inline simdjson_result<element> object::at_key_case_insensitive(const std::strin
   for (iterator field = begin(); field != end_field; ++field) {
     auto field_key = field.key();
     if (key.length() == field_key.length()) {
-      bool equal = true;
+      // See For case-insensitive string comparisons, avoid char-by-char functions
-      for (size_t i=0; i<field_key.length(); i++) {
+      // https://lemire.me/blog/2020/04/30/for-case-insensitive-string-comparisons-avoid-char-by-char-functions/
-        equal = equal && std::tolower(key[i]) == std::tolower(field_key[i]);
+      // Note that it might be worth rolling our own strncasecmp function, with vectorization.
-      }
+      const bool equal = (simdjson_strncasecmp(key.data(), field_key.data(), key.length()) == 0);
       if (equal) { return field.value(); }
     }
   }

include/simdjson/portability.h

@@ -147,8 +147,13 @@ compiling for a known 64-bit platform."
 // regular visual studio and clang under visual studio.
 // clang under Windows has _stricmp (like visual studio) but not strcasecmp (as clang normally has)
 #define simdjson_strcasecmp _stricmp
+#define simdjson_strncasecmp _strnicmp
 #else
+// The strcasecmp, strncasecmp, and strcasestr functions do not work with multibyte strings (e.g. UTF-8).
+// So they are only useful for ASCII in our context.
+// https://www.gnu.org/software/libunistring/manual/libunistring.html#char-_002a-strings
 #define simdjson_strcasecmp strcasecmp
+#define simdjson_strncasecmp strncasecmp
 #endif
 
 namespace simdjson {