cache ctable.dtype

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

bcolz/ctable.py

@@ -142,6 +142,9 @@ def cparams(self):
     @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 +153,8 @@ 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):
@@ -222,6 +226,8 @@ def __init__(self, columns=None, names=None, **kwargs):
         # Attach the attrs to this object
         self.attrs = attrs.attrs(self.rootdir, self.mode, _new=_new)
 
+        # Trigger update of dtype cache
+        self._dtype = None
         # Cache a structured array of len 1 for ctable[int] acceleration
         self._arr1 = np.empty(shape=(1,), dtype=self.dtype)
 
@@ -487,6 +493,8 @@ 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 +548,8 @@ 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)