Use np.strings in the string accessor

asv_bench/benchmarks/accessors.py

@@ -22,3 +22,42 @@ def time_year(self, calendar):
 
     def time_floor(self, calendar):
         _ = self.da.time.dt.floor("D")
+
+
+NSTRINGS = 100_000
+
+
+@parameterize({"dtype": ("U", "S", "O")})
+class StringAccessor:
+    def setup(self, dtype):
+        rng = np.random.default_rng(0)
+        chars = np.array(list("abcdefgh XYZ"))
+        strings = rng.choice(chars, size=(NSTRINGS, 12)).view("U12").ravel()
+        self.da = xr.DataArray(strings.astype(dtype), dims="x")
+
+    def time_len(self, dtype):
+        self.da.str.len()
+
+    def time_isalpha(self, dtype):
+        self.da.str.isalpha()
+
+    def time_startswith(self, dtype):
+        self.da.str.startswith("ab")
+
+    def time_find(self, dtype):
+        self.da.str.find("ab")
+
+    def time_contains(self, dtype):
+        self.da.str.contains("ab", regex=False)
+
+    def time_strip(self, dtype):
+        self.da.str.strip()
+
+    def time_center(self, dtype):
+        self.da.str.center(16)
+
+    def time_replace(self, dtype):
+        self.da.str.replace("ab", "-", regex=False)
+
+    def time_repeat(self, dtype):
+        self.da.str.repeat(2)

doc/whats-new.rst

@@ -147,6 +147,11 @@ Deprecations
 
 Bug Fixes
 ~~~~~~~~~
+- Results of the string accessor methods ``strip``, ``pad``, ``center``,
+  ``ljust``, ``rjust``, ``zfill`` and ``replace`` on dask arrays are no longer
+  truncated to a single character, and ``repeat`` no longer fails on dask arrays
+  (:pull:`11746`).
+  By `Michael Niklas <https://github.com/headtr1ck>`_.
 - Encoding a string variable whose ``char_dim_name`` has characters after its
   last number (e.g. ``"n4_chars"``) no longer drops them. When the length does
   not match, only the number is replaced (``"n3_chars"``) (:pull:`11757`).
@@ -386,6 +391,14 @@ Documentation
 
 Performance
 ~~~~~~~~~~~
+- The string accessor uses the vectorized functions of ``np.strings`` for
+  ``len``, ``startswith``, ``endswith``, ``find``, ``rfind``, ``index``,
+  ``rindex``, ``strip``, ``lstrip``, ``rstrip``, ``pad``, ``center``,
+  ``ljust``, ``rjust``, ``zfill``, ``repeat``, the ``is*`` methods and
+  ``contains`` and ``replace`` with ``regex=False`` and case-sensitive matching,
+  which makes them 5-80 times faster for arrays with a string dtype
+  (:pull:`11746`).
+  By `Michael Niklas <https://github.com/headtr1ck>`_.
 - Speed up rolling reductions of dask arrays without bottleneck, e.g. ``sum``,
   ``mean`` and ``count``, by about 7x with about 12x fewer tasks for a window of 20.
   Counting the values in each window no longer splits the chunks of the other