cache ctable.dtype

Among others, this allow improving the performance of custom implementations of
the outflavor abstraction layer.

bcolz/ctable.py

@@ -139,9 +139,13 @@ def cparams(self):
         "The compression parameters for this object."
         return self._cparams
 
+    _dtype = None
     @property
     def dtype(self):
         "The data type of this object (numpy dtype)."
+        if self._dtype:
+            return self._dtype
+
         names, cols = self.names, self.cols
         l = []
         for name in names:
@@ -150,7 +154,8 @@ def dtype(self):
             t = (name, col.dtype) if col.ndim == 1 else \
                 (name, (col.dtype, col.shape[1:]))
             l.append(t)
-        return np.dtype(l)
+        self._dtype = np.dtype(l)
+        return self._dtype
 
     @property
     def names(self):
@@ -487,6 +492,8 @@ def addcol(self, newcol, name=None, pos=None, move=False, **kwargs):
 
         # Insert the column
         self.cols.insert(name, pos, newcol)
+        # Trigger update of dtype cache
+        self._dtype = None
         # Update _arr1
         self._arr1 = np.empty(shape=(1,), dtype=self.dtype)
 
@@ -540,6 +547,8 @@ def delcol(self, name=None, pos=None, keep=False):
         if not keep:
             col.purge()
 
+        # Trigger update of dtype cache
+        self._dtype = None
         # Update _arr1
         self._arr1 = np.empty(shape=(1,), dtype=self.dtype)