Shorter repr for attributes

NetCDF files often have tens of attributes, including multi-paragraph
summaries or the full modification history of the file.  It's great to
have this available in the .attrs, but we can truncate it substantially
in the repr!  Hopefully this will stop people writing `data.attrs = {}`
and discarding metadata in interactive workflows for the sake of cleaner
output.

Author: Zac-HD

doc/whats-new.rst

@@ -41,6 +41,13 @@ By `Henry S. Harrison <https://hsharrison.github.io>`_.
   Note, the default is ``autoclose=False``, which is consistent with previous
   xarray behavior.  By `Phillip J. Wolfram <https://github.com/pwolfram>`_.
 
+- The ``repr()`` of ``Dataset`` and ``DataArray`` attributes uses a similar
+  format to coordinates and variables, with vertically aligned entries
+  truncated to fit on a single line.  Hopefully this will stop people writing
+  ``data.attrs = {}`` and discarding metadata in notebooks for the sake of
+  cleaner output.  The full metadata is still available as ``data.attrs``.
+  By `Zac Hatfield-Dodds <https://github.com/Zac-HD>`_.
+
 Bug fixes
 ~~~~~~~~~
 - ``rolling`` now keeps its original dimension order (:issue:`1125`).

xarray/core/formatting.py

@@ -253,8 +253,15 @@ def summarize_coord(name, var, col_width):
 
 
 def summarize_attr(key, value, col_width=None):
-    # ignore col_width for now to more clearly distinguish attributes
-    return u'    %s: %s' % (key, maybe_truncate(value))
+    """Summary for __repr__ - use ``X.attrs[key]`` for full value."""
+    # Indent key and add ':', then right-pad if col_width is not None
+    k_str = u'    %s:' % key
+    if col_width is not None:
+        k_str = pretty_print(k_str, col_width)
+    # Replace tabs and newlines, so we print on one line in known width
+    v_str = unicode_type(value).replace(u'\t', u'\\t').replace(u'\n', u'\\n')
+    # Finally, truncate to the desired display width
+    return maybe_truncate(u'%s %s' % (k_str, v_str), OPTIONS['display_width'])
 
 
 EMPTY_REPR = u'    *empty*'

xarray/tests/test_dataarray.py

@@ -49,7 +49,7 @@ def test_repr(self):
             other    int64 0
         Dimensions without coordinates: time
         Attributes:
-            foo: bar""")
+            foo:      bar""")
         self.assertEqual(expected, repr(data_array))
 
     def test_repr_multiindex(self):

xarray/tests/test_dataset.py

@@ -97,7 +97,7 @@ def test_repr(self):
             var2     (dim1, dim2) float64 1.162 -1.097 -2.123 1.04 -0.4034 -0.126 ...
             var3     (dim3, dim1) float64 0.5565 -0.2121 0.4563 1.545 -0.2397 0.1433 ...
         Attributes:
-            foo: bar""") % data['dim3'].dtype
+            foo:      bar""") % data['dim3'].dtype
         actual = '\n'.join(x.rstrip() for x in repr(data).split('\n'))
         print(actual)
         self.assertEqual(expected, actual)
@@ -183,7 +183,7 @@ def test_unicode_data(self):
         Data variables:
             *empty*
         Attributes:
-            å: ∑""" % u'ba®')
+            å:        ∑""" % u'ba®')
         actual = unicode_type(data)
         self.assertEqual(expected, actual)
 

xarray/tests/test_variable.py

@@ -644,7 +644,7 @@ def test_repr(self):
         array([[1, 2, 3],
                [4, 5, 6]])
         Attributes:
-            foo: bar
+            foo:      bar
         """).strip()
         self.assertEqual(expected, repr(v))