Squashed commit of the following:

commit 08f7f74
Merge: 53c0f4e 278d2e6
Author: dcherian <[email protected]>
Date:   Tue Oct 29 09:36:58 2019 -0600

    Merge remote-tracking branch 'upstream/master' into fix/dask-computes

    * upstream/master:
      upgrade black verison to 19.10b0 (pydata#3456)
      Remove outdated code related to compatibility with netcdftime (pydata#3450)

commit 53c0f4e
Author: dcherian <[email protected]>
Date:   Tue Oct 29 09:25:27 2019 -0600

    Add identity check to lazy_array_equiv

commit 5e742e4
Author: dcherian <[email protected]>
Date:   Tue Oct 29 09:22:15 2019 -0600

    update whats new

commit ee0d422
Merge: e99148e 74ca69a
Author: dcherian <[email protected]>
Date:   Tue Oct 29 09:18:38 2019 -0600

    Merge remote-tracking branch 'upstream/master' into fix/dask-computes

    * upstream/master:
      Remove deprecated behavior from dataset.drop docstring (pydata#3451)
      jupyterlab dark theme (pydata#3443)
      Drop groups associated with nans in group variable (pydata#3406)
      Allow ellipsis (...) in transpose (pydata#3421)
      Another groupby.reduce bugfix. (pydata#3403)
      add icomoon license (pydata#3448)

commit e99148e
Author: dcherian <[email protected]>
Date:   Tue Oct 29 09:17:58 2019 -0600

    add concat test

commit 4a66e7c
Author: dcherian <[email protected]>
Date:   Mon Oct 28 10:19:32 2019 -0600

    review suggestions.

commit 8739ddd
Author: dcherian <[email protected]>
Date:   Mon Oct 28 08:32:15 2019 -0600

    better docstring

commit e84cc97
Author: dcherian <[email protected]>
Date:   Sun Oct 27 20:22:13 2019 -0600

    Optimize dask array equality checks.

    Dask arrays with the same graph have the same name. We can use this to quickly
    compare dask-backed variables without computing.

    Fixes pydata#3068 and pydata#3311

Author: dcherian

doc/whats-new.rst

@@ -59,6 +59,9 @@ Bug fixes
   but cloudpickle isn't (:issue:`3401`) by `Rhys Doyle <https://github.com/rdoyle45>`_
 - Fix grouping over variables with NaNs. (:issue:`2383`, :pull:`3406`).
   By `Deepak Cherian <https://github.com/dcherian>`_.
+- Use dask names to compare dask objects prior to comparing values after computation.
+  (:issue:`3068`, :issue:`3311`, :issue:`3454`, :pull:`3453`).
+  By `Deepak Cherian <https://github.com/dcherian>`_.
 - Sync with cftime by removing `dayofwk=-1` for cftime>=1.0.4.
   By `Anderson Banihirwe <https://github.com/andersy005>`_.
 - Fix :py:meth:`xarray.core.groupby.DataArrayGroupBy.reduce` and

xarray/core/concat.py

@@ -2,6 +2,7 @@
 
 from . import dtypes, utils
 from .alignment import align
+from .duck_array_ops import lazy_array_equiv
 from .merge import _VALID_COMPAT, unique_variable
 from .variable import IndexVariable, Variable, as_variable
 from .variable import concat as concat_vars
@@ -189,6 +190,21 @@ def process_subset_opt(opt, subset):
                 # all nonindexes that are not the same in each dataset
                 for k in getattr(datasets[0], subset):
                     if k not in concat_over:
+                        equals[k] = None
+                        variables = [ds.variables[k] for ds in datasets]
+                        # first check without comparing values i.e. no computes
+                        for var in variables[1:]:
+                            equals[k] = getattr(variables[0], compat)(
+                                var, equiv=lazy_array_equiv
+                            )
+                            if not equals[k]:
+                                break
+
+                        if equals[k] is not None:
+                            if equals[k] is False:
+                                concat_over.add(k)
+                            continue
+
                         # Compare the variable of all datasets vs. the one
                         # of the first dataset. Perform the minimum amount of
                         # loads in order to avoid multiple loads from disk

xarray/core/duck_array_ops.py

@@ -174,27 +174,53 @@ def as_shared_dtype(scalars_or_arrays):
     return [x.astype(out_type, copy=False) for x in arrays]
 
 
-def allclose_or_equiv(arr1, arr2, rtol=1e-5, atol=1e-8):
-    """Like np.allclose, but also allows values to be NaN in both arrays
+def lazy_array_equiv(arr1, arr2):
+    """Like array_equal, but doesn't actually compare values.
+       Returns True or False when equality can be determined without computing.
+       Returns None when equality cannot determined (e.g. one or both of arr1, arr2 are numpy arrays)
     """
+    if arr1 is arr2:
+        return True
     arr1 = asarray(arr1)
     arr2 = asarray(arr2)
     if arr1.shape != arr2.shape:
         return False
-    return bool(isclose(arr1, arr2, rtol=rtol, atol=atol, equal_nan=True).all())
+    if (
+        dask_array
+        and isinstance(arr1, dask_array.Array)
+        and isinstance(arr2, dask_array.Array)
+    ):
+        # GH3068
+        if arr1.name == arr2.name:
+            return True
+    return None
+
+
+def allclose_or_equiv(arr1, arr2, rtol=1e-5, atol=1e-8):
+    """Like np.allclose, but also allows values to be NaN in both arrays
+    """
+    arr1 = asarray(arr1)
+    arr2 = asarray(arr2)
+    lazy_equiv = lazy_array_equiv(arr1, arr2)
+    if lazy_equiv is None:
+        return bool(isclose(arr1, arr2, rtol=rtol, atol=atol, equal_nan=True).all())
+    else:
+        return lazy_equiv
 
 
 def array_equiv(arr1, arr2):
     """Like np.array_equal, but also allows values to be NaN in both arrays
     """
     arr1 = asarray(arr1)
     arr2 = asarray(arr2)
-    if arr1.shape != arr2.shape:
-        return False
-    with warnings.catch_warnings():
-        warnings.filterwarnings("ignore", "In the future, 'NAT == x'")
-        flag_array = (arr1 == arr2) | (isnull(arr1) & isnull(arr2))
-        return bool(flag_array.all())
+    lazy_equiv = lazy_array_equiv(arr1, arr2)
+    if lazy_equiv is None:
+        with warnings.catch_warnings():
+            warnings.filterwarnings("ignore", "In the future, 'NAT == x'")
+            flag_array = (arr1 == arr2) | (isnull(arr1) & isnull(arr2))
+            return bool(flag_array.all())
+    else:
+        return lazy_equiv
 
 
 def array_notnull_equiv(arr1, arr2):
@@ -203,12 +229,14 @@ def array_notnull_equiv(arr1, arr2):
     """
     arr1 = asarray(arr1)
     arr2 = asarray(arr2)
-    if arr1.shape != arr2.shape:
-        return False
-    with warnings.catch_warnings():
-        warnings.filterwarnings("ignore", "In the future, 'NAT == x'")
-        flag_array = (arr1 == arr2) | isnull(arr1) | isnull(arr2)
-        return bool(flag_array.all())
+    lazy_equiv = lazy_array_equiv(arr1, arr2)
+    if lazy_equiv is None:
+        with warnings.catch_warnings():
+            warnings.filterwarnings("ignore", "In the future, 'NAT == x'")
+            flag_array = (arr1 == arr2) | isnull(arr1) | isnull(arr2)
+            return bool(flag_array.all())
+    else:
+        return lazy_equiv
 
 
 def count(data, axis=None):

xarray/core/merge.py

@@ -19,6 +19,7 @@
 
 from . import dtypes, pdcompat
 from .alignment import deep_align
+from .duck_array_ops import lazy_array_equiv
 from .utils import Frozen, dict_equiv
 from .variable import Variable, as_variable, assert_unique_multiindex_level_names
 
@@ -123,16 +124,24 @@ def unique_variable(
         combine_method = "fillna"
 
     if equals is None:
-        out = out.compute()
+        # first check without comparing values i.e. no computes
         for var in variables[1:]:
-            equals = getattr(out, compat)(var)
+            equals = getattr(out, compat)(var, equiv=lazy_array_equiv)
             if not equals:
                 break
 
+        # now compare values with minimum number of computes
+        if not equals:
+            out = out.compute()
+            for var in variables[1:]:
+                equals = getattr(out, compat)(var)
+                if not equals:
+                    break
+
     if not equals:
         raise MergeError(
-            "conflicting values for variable {!r} on objects to be combined. "
-            "You can skip this check by specifying compat='override'.".format(name)
+            f"conflicting values for variable {name!r} on objects to be combined. "
+            "You can skip this check by specifying compat='override'."
         )
 
     if combine_method:

xarray/core/variable.py

@@ -1231,7 +1231,9 @@ def transpose(self, *dims) -> "Variable":
             dims = self.dims[::-1]
         dims = tuple(infix_dims(dims, self.dims))
         axes = self.get_axis_num(dims)
-        if len(dims) < 2:  # no need to transpose if only one dimension
+        if len(dims) < 2 or dims == self.dims:
+            # no need to transpose if only one dimension
+            # or dims are in same order
             return self.copy(deep=False)
 
         data = as_indexable(self._data).transpose(axes)
@@ -1590,22 +1592,24 @@ def broadcast_equals(self, other, equiv=duck_array_ops.array_equiv):
             return False
         return self.equals(other, equiv=equiv)
 
-    def identical(self, other):
+    def identical(self, other, equiv=duck_array_ops.array_equiv):
         """Like equals, but also checks attributes.
         """
         try:
-            return utils.dict_equiv(self.attrs, other.attrs) and self.equals(other)
+            return utils.dict_equiv(self.attrs, other.attrs) and self.equals(
+                other, equiv=equiv
+            )
         except (TypeError, AttributeError):
             return False
 
-    def no_conflicts(self, other):
+    def no_conflicts(self, other, equiv=duck_array_ops.array_notnull_equiv):
         """True if the intersection of two Variable's non-null data is
         equal; otherwise false.
 
         Variables can thus still be equal if there are locations where either,
         or both, contain NaN values.
         """
-        return self.broadcast_equals(other, equiv=duck_array_ops.array_notnull_equiv)
+        return self.broadcast_equals(other, equiv=equiv)
 
     def quantile(self, q, dim=None, interpolation="linear", keep_attrs=None):
         """Compute the qth quantile of the data along the specified dimension.

xarray/tests/test_dask.py

@@ -22,6 +22,7 @@
     assert_identical,
     raises_regex,
 )
+from ..core.duck_array_ops import lazy_array_equiv
 
 dask = pytest.importorskip("dask")
 da = pytest.importorskip("dask.array")
@@ -423,7 +424,53 @@ def test_concat_loads_variables(self):
         out.compute()
         assert kernel_call_count == 24
 
-        # Finally, test that riginals are unaltered
+        # Finally, test that originals are unaltered
+        assert ds1["d"].data is d1
+        assert ds1["c"].data is c1
+        assert ds2["d"].data is d2
+        assert ds2["c"].data is c2
+        assert ds3["d"].data is d3
+        assert ds3["c"].data is c3
+
+        # now check that concat() is correctly using dask name equality to skip loads
+        out = xr.concat(
+            [ds1, ds1, ds1], dim="n", data_vars="different", coords="different"
+        )
+        assert kernel_call_count == 24
+        # variables are not loaded in the output
+        assert isinstance(out["d"].data, dask.array.Array)
+        assert isinstance(out["c"].data, dask.array.Array)
+
+        out = xr.concat(
+            [ds1, ds1, ds1], dim="n", data_vars=[], coords=[], compat="identical"
+        )
+        assert kernel_call_count == 24
+        # variables are not loaded in the output
+        assert isinstance(out["d"].data, dask.array.Array)
+        assert isinstance(out["c"].data, dask.array.Array)
+
+        out = xr.concat(
+            [ds1, ds2.compute(), ds3],
+            dim="n",
+            data_vars="all",
+            coords="different",
+            compat="identical",
+        )
+        # c1,c3 must be computed for comparison since c2 is numpy;
+        # d2 is computed too
+        assert kernel_call_count == 28
+
+        out = xr.concat(
+            [ds1, ds2.compute(), ds3],
+            dim="n",
+            data_vars="all",
+            coords="all",
+            compat="identical",
+        )
+        # no extra computes
+        assert kernel_call_count == 30
+
+        # Finally, test that originals are unaltered
         assert ds1["d"].data is d1
         assert ds1["c"].data is c1
         assert ds2["d"].data is d2
@@ -1135,3 +1182,42 @@ def test_make_meta(map_ds):
     for variable in map_ds.data_vars:
         assert variable in meta.data_vars
         assert meta.data_vars[variable].shape == (0,) * meta.data_vars[variable].ndim
+
+
+def test_identical_coords_no_computes():
+    lons2 = xr.DataArray(da.zeros((10, 10), chunks=2), dims=("y", "x"))
+    a = xr.DataArray(
+        da.zeros((10, 10), chunks=2), dims=("y", "x"), coords={"lons": lons2}
+    )
+    b = xr.DataArray(
+        da.zeros((10, 10), chunks=2), dims=("y", "x"), coords={"lons": lons2}
+    )
+    with raise_if_dask_computes():
+        c = a + b
+    assert_identical(c, a)
+
+
+def test_lazy_array_equiv():
+    lons1 = xr.DataArray(da.zeros((10, 10), chunks=2), dims=("y", "x"))
+    lons2 = xr.DataArray(da.zeros((10, 10), chunks=2), dims=("y", "x"))
+    var1 = lons1.variable
+    var2 = lons2.variable
+    with raise_if_dask_computes():
+        lons1.equals(lons2)
+    with raise_if_dask_computes():
+        var1.equals(var2 / 2, equiv=lazy_array_equiv)
+    assert var1.equals(var2.compute(), equiv=lazy_array_equiv) is None
+    assert var1.compute().equals(var2.compute(), equiv=lazy_array_equiv) is None
+
+    with raise_if_dask_computes():
+        assert lons1.equals(lons1.transpose("y", "x"))
+
+    with raise_if_dask_computes():
+        for compat in [
+            "broadcast_equals",
+            "equals",
+            "override",
+            "identical",
+            "no_conflicts",
+        ]:
+            xr.merge([lons1, lons2], compat=compat)