Optimize Char#to_s when it's ascii and the IO has no encoding.

This makes CSV and JSON parsing, and probably other parsers, significantly faster. - For a CSV test, it goes down from 1.67s to 1.25s - For a JSON test, it goes down from 3.93s to 3.38s
crystal-lang · Sep 10, 2016 · 6d867d9 · 6d867d9
1 parent 3aadab7
commit 6d867d9
Showing 1 changed file with 8 additions and 2 deletions.
diff --git a/src/char.cr b/src/char.cr
@@ -633,9 +633,15 @@ struct Char
   #
   # This appends this char's bytes as encoded by UTF-8 to the given `IO`.
   def to_s(io : IO)
-    if ord <= 0x7f
+    if ascii?
       byte = ord.to_u8
-      io.write_utf8 Slice.new(pointerof(byte), 1)
+
+      # Optimization: writing a slice is much slower than writing a byte
+      if io.@encoding
+        io.write_utf8 Slice.new(pointerof(byte), 1)
+      else
+        io.write_byte byte
+      end
     else
       chars = uninitialized UInt8[4]
       i = 0