[Python] Defer import of h5py until time of use

This avoids potential conflicts between the versions of libhdf5 linked
to the Cantera library and h5py, which could occur when a C++ main
application made use of the Python ExtensibleRate class.

.github/workflows/main.yml

@@ -53,7 +53,7 @@ jobs:
       run: |
         python3 -m pip install ruamel.yaml scons==3.1.2 numpy cython pandas pytest \
         pytest-github-actions-annotate-failures
-        python3 -m pip install h5py || python3 -m pip install --no-binary=h5py h5py
+        python3 -m pip install h5py
     - name: Build Cantera
       run: |
         python3 `which scons` build env_vars=all -j2 debug=n --debug=time \
@@ -94,8 +94,7 @@ jobs:
       run: python3 -m pip install -U pip setuptools wheel
     - name: Install Python dependencies
       run: |
-        python3 -m pip install ruamel.yaml scons numpy cython pandas pytest pytest-github-actions-annotate-failures
-        python3 -m pip install --no-binary=h5py h5py
+        python3 -m pip install ruamel.yaml scons numpy cython pandas h5py pytest pytest-github-actions-annotate-failures
     - name: Build Cantera
       run: python3 `which scons` build env_vars=all
         CXX=clang++-12 CC=clang-12 f90_interface=n extra_lib_dirs=/usr/lib/llvm/lib
@@ -194,9 +193,8 @@ jobs:
       run: python3 -m pip install -U pip setuptools wheel
     - name: Install Python dependencies
       run: |
-        python3 -m pip install ruamel.yaml scons numpy cython pandas scipy pytest \
+        python3 -m pip install ruamel.yaml scons numpy cython pandas scipy pytest h5py \
         pytest-github-actions-annotate-failures pytest-cov gcovr
-        python3 -m pip install --no-binary=h5py h5py
     - name: Setup .NET Core SDK
       uses: actions/setup-dotnet@v2
       with:
@@ -366,8 +364,7 @@ jobs:
         run: python3 -m pip install -U pip setuptools wheel
       - name: Install Python dependencies
         run: |
-          python3 -m pip install ruamel.yaml scons numpy cython pandas matplotlib scipy
-          python3 -m pip install --no-binary=h5py h5py
+          python3 -m pip install ruamel.yaml scons numpy cython pandas matplotlib scipy h5py
       - name: Build Cantera
         # compile with GCC 9.4.0 on ubuntu-20.04 as an alternative to the default
         # (GCC 7.5.0 is both default and oldest supported version)
@@ -619,9 +616,8 @@ jobs:
       run: python3 -m pip install -U pip setuptools wheel
     - name: Install Python dependencies
       run: |
-        python3 -m pip install ruamel.yaml scons numpy cython pandas pytest \
+        python3 -m pip install ruamel.yaml scons numpy cython pandas h5py pytest \
         pytest-github-actions-annotate-failures
-        python3 -m pip install --no-binary=h5py h5py
     - name: Setup Intel oneAPI environment
       run: |
         source /opt/intel/oneapi/setvars.sh

interfaces/cython/cantera/composite.py

@@ -7,17 +7,20 @@ import numpy as np
 from collections import OrderedDict
 import csv as _csv
 
-import pkg_resources
-
-# avoid explicit dependence of cantera on h5py
-try:
-    pkg_resources.get_distribution('h5py')
-except pkg_resources.DistributionNotFound:
-    _h5py = ImportError('Method requires a working h5py installation.')
-else:
-    import h5py as _h5py
+def _import_h5py():
+    # avoid explicit dependence of cantera on h5py
+    import pkg_resources  # local import to reduce overall import time
+    try:
+        pkg_resources.get_distribution('h5py')
+    except pkg_resources.DistributionNotFound:
+        raise ImportError('Method requires a working h5py installation.')
+    else:
+        import h5py
+        return h5py
 
 # avoid explicit dependence of cantera on pandas
+import pkg_resources
+
 try:
     pkg_resources.get_distribution('pandas')
 except pkg_resources.DistributionNotFound:
@@ -1311,8 +1314,7 @@ class SolutionArray:
         requires a working installation of *h5py* (``h5py`` can be installed using
         pip or conda).
         """
-        if isinstance(_h5py, ImportError):
-            raise _h5py
+        h5py = _import_h5py()
 
         # collect data
         data = self.collect_data(*args, cols=cols, **kwargs)
@@ -1322,7 +1324,7 @@ class SolutionArray:
         hdf_kwargs = {k: v for k, v in hdf_kwargs.items() if v is not None}
 
         # save to container file
-        with _h5py.File(filename, mode) as hdf:
+        with h5py.File(filename, mode) as hdf:
 
             # check existence of tagged item
             if not group:
@@ -1395,10 +1397,9 @@ class SolutionArray:
         The method imports data using `restore_data` and requires a working
         installation of *h5py* (``h5py`` can be installed using pip or conda).
         """
-        if isinstance(_h5py, ImportError):
-            raise _h5py
+        h5py = _import_h5py()
 
-        with _h5py.File(filename, 'r') as hdf:
+        with h5py.File(filename, 'r') as hdf:
 
             groups = list(hdf.keys())
             if not len(groups):
@@ -1417,7 +1418,7 @@ class SolutionArray:
             # identify subgroup
             if subgroup is not None:
                 sub_names = [key for key, value in root.items()
-                             if isinstance(value, _h5py.Group)]
+                             if isinstance(value, h5py.Group)]
                 if not len(sub_names):
                     msg = "HDF group '{}' does not contain valid data"
                     raise IOError(msg.format(group))
@@ -1454,13 +1455,13 @@ class SolutionArray:
             self._meta = dict(dgroup.attrs.items())
             for name, value in dgroup.items():
                 # support one level of recursion
-                if isinstance(value, _h5py.Group):
+                if isinstance(value, h5py.Group):
                     self._meta[name] = dict(value.attrs.items())
 
             # load data
             data = OrderedDict()
             for name, value in dgroup.items():
-                if isinstance(value, _h5py.Group):
+                if isinstance(value, h5py.Group):
                     continue
                 elif value.dtype.type == np.bytes_:
                     data[name] = np.array(value).astype('U')

test/kinetics/kineticsFromYaml.cpp

@@ -518,10 +518,6 @@ TEST(Reaction, PythonExtensibleRate)
     #ifndef CT_HAS_PYTHON
     GTEST_SKIP();
     #endif
-    #ifdef CT_USE_HDF5
-    // potential mismatch of HDF libraries between h5py and HighFive
-    GTEST_SKIP();
-    #endif
     auto sol = newSolution("extensible-reactions.yaml");
     auto R = sol->kinetics()->reaction(0);
     EXPECT_EQ(R->type(), "square-rate");

test/python/test_composite.py

@@ -5,7 +5,14 @@ from collections import OrderedDict
 import pickle
 
 import cantera as ct
-from cantera.composite import _h5py, _pandas
+
+try:
+    h5py = ct.composite._import_h5py()
+    have_h5py = True
+except ImportError:
+    have_h5py = False
+
+from cantera.composite import _pandas
 from . import utilities
 
 
@@ -255,7 +262,7 @@ class TestSolutionArrayIO(utilities.CanteraTest):
         self.assertEqual(states[0].P, gas.P)
         self.assertArrayNear(states[0].Y, gas.Y)
 
-    @utilities.unittest.skipIf(isinstance(_h5py, ImportError), "h5py is not installed")
+    @utilities.unittest.skipIf(not have_h5py, "h5py is not installed")
     def test_import_no_norm_data(self):
         outfile = self.test_work_path / "solutionarray.h5"
         # In Python >= 3.8, this can be replaced by the missing_ok argument
@@ -336,7 +343,7 @@ class TestSolutionArrayIO(utilities.CanteraTest):
         with self.assertRaisesRegex(NotImplementedError, 'not supported'):
             states.to_pandas()
 
-    @utilities.unittest.skipIf(isinstance(_h5py, ImportError), "h5py is not installed")
+    @utilities.unittest.skipIf(not have_h5py, "h5py is not installed")
     def test_write_hdf(self):
         outfile = self.test_work_path / "solutionarray.h5"
         # In Python >= 3.8, this can be replaced by the missing_ok argument
@@ -365,7 +372,7 @@ class TestSolutionArrayIO(utilities.CanteraTest):
         gas = ct.Solution('gri30.yaml', transport_model=None)
         ct.SolutionArray(gas, 10).write_hdf(outfile)
 
-        with _h5py.File(outfile, 'a') as hdf:
+        with h5py.File(outfile, 'a') as hdf:
             hdf.create_group('spam')
 
         c = ct.SolutionArray(self.gas)
@@ -382,7 +389,7 @@ class TestSolutionArrayIO(utilities.CanteraTest):
         c.read_hdf(outfile, group='foo/bar/baz')
         self.assertArrayNear(states.T, c.T)
 
-    @utilities.unittest.skipIf(isinstance(_h5py, ImportError), "h5py is not installed")
+    @utilities.unittest.skipIf(not have_h5py, "h5py is not installed")
     def test_write_hdf_str_column(self):
         outfile = self.test_work_path / "solutionarray.h5"
         # In Python >= 3.8, this can be replaced by the missing_ok argument
@@ -396,7 +403,7 @@ class TestSolutionArrayIO(utilities.CanteraTest):
         b.read_hdf(outfile)
         self.assertEqual(list(states.spam), list(b.spam))
 
-    @utilities.unittest.skipIf(isinstance(_h5py, ImportError), "h5py is not installed")
+    @utilities.unittest.skipIf(not have_h5py, "h5py is not installed")
     def test_write_hdf_multidim_column(self):
         outfile = self.test_work_path / "solutionarray.h5"
         # In Python >= 3.8, this can be replaced by the missing_ok argument
@@ -571,7 +578,7 @@ class TestRestorePureFluid(utilities.CanteraTest):
         b.restore_data(data)
         check(a, b)
 
-    @utilities.unittest.skipIf(isinstance(_h5py, ImportError), "h5py is not installed")
+    @utilities.unittest.skipIf(not have_h5py, "h5py is not installed")
     def test_import_no_norm_water(self):
         outfile = self.test_work_path / "solutionarray.h5"
         # In Python >= 3.8, this can be replaced by the missing_ok argument

test/python/test_onedim.py

@@ -4,8 +4,6 @@ import numpy as np
 from .utilities import allow_deprecated
 import pytest
 
-from cantera.composite import _h5py
-
 
 class TestOnedim(utilities.CanteraTest):
     def test_instantiate(self):