Merge 922c1c5 into 837b1e4

etc/requirements_locked.txt

@@ -35,7 +35,7 @@ flask-cors==2.1.3
 flask==1.1.1              # via flask-cors
 funcsigs==1.0.2           # via mock, python-interface
 futures==3.2.0            # via tornado
-h5py==2.7.1
+h5py==2.10.0
 idna==2.7                 # via requests
 intervaltree==2.1.0
 iso3166==0.9

zipline/data/minute_bars.py

@@ -21,6 +21,7 @@
 from lru import LRU
 import bcolz
 from bcolz import ctable
+import h5py
 from intervaltree import IntervalTree
 import logbook
 import numpy as np
@@ -1373,7 +1374,31 @@ class H5MinuteBarUpdateReader(MinuteBarUpdateReader):
         The path of the HDF5 file from which to source data.
     """
     def __init__(self, path):
-        self._panel = pd.read_hdf(path)
+        # There is a bug in `pandas.read_hdf` whereby in Python 3 it fails to
+        # read the timezone attr of an h5 file if that file was written in
+        # Python 2. Until zipline has dropped Python 2 entirely we are at risk
+        # of hitting this issue. For now, use h5py to read the file instead.
+        # The downside of using h5py directly is that we need to interpret the
+        # attrs manually when creating our panel (specifically the tz attr),
+        # but since we know exactly how the file was written this should be
+        # pretty straightforward.
+        with h5py.File(path, 'r') as f:
+            updates = f['updates']
+            values = updates['block0_values']
+            items = updates['axis0']
+            major = updates['axis1']
+            minor = updates['axis2']
+
+            self._panel = pd.Panel(
+                data=np.array(values).T,
+                items=np.array(items),
+                major_axis=pd.DatetimeIndex(
+                    major,
+                    tz=major.attrs['tz'].decode(),
+                    freq='T',
+                ),
+                minor_axis=np.array(minor).astype('U'),
+            )
 
     def read(self, dts, sids):
         panel = self._panel[sids, dts, :]