make scipy and matplotlib required dependencies (#1159) #1404

orbeckst · 2017-06-16T04:18:27Z

Changes made in this Pull Request:

make scipy and matplotlib full dependencies (see discussion in Handling Optional dependencies for the analysis module #1159 and the updated list of core dependencies)
scipy and matplotlib are imported at top in analysis
removed conditional skipping of tests when scipy or matplotlib are missing

PR Checklist

Tests?
n/a Docs?
CHANGELOG updated?
Issue raised/referenced?

kain88-de · 2017-06-16T07:52:08Z

.travis.yml

-    - PIP_DEPENDENCIES='griddataformats'
+    - CONDA_DEPENDENCIES="mmtf-python nose=1.3.7 mock six biopython networkx cython joblib nose-timer matplotlib scipy griddataformats"
+    - CONDA_ALL_DEPENDENCIES="mmtf-python nose=1.3.7 mock six biopython networkx cython joblib nose-timer matplotlib netcdf4 scikit-learn scipy griddataformats seaborn coveralls clustalw=2.1"
+    - PIP_DEPENDENCIES=""


you can remove this line completely.

kain88-de

Looks good besides some minor comments

kain88-de · 2017-06-16T07:52:46Z

package/MDAnalysis/analysis/distances.py

@@ -42,6 +42,7 @@
           'contact_matrix', 'dist', 'between']

 import numpy as np
+import scipy.sparse


why not from scipy import sparse?

I like the full imports much better if we only use a few things. Much easier to grep. And really not much more to write. Why not be explicit with where you get functions from? And it keeps the name space tidier.

(Sidenote: I particular dislike function imports – I think they are really bad because you have no idea where they come from when reading code. )

kain88-de · 2017-06-16T07:53:29Z

package/MDAnalysis/analysis/encore/similarity.py

-    del msg
+
+import numpy as np
+import scipy.stats


why not from scipy.stats import gaussian_kde?

Because that is evil.

Honestly, function imports at the module level are bad. See comments above. See PEP something or other IIRC...

PS: Please don't.

kain88-de · 2017-06-16T07:54:07Z

package/MDAnalysis/analysis/hbonds/hbond_autocorrel.py

@@ -155,14 +155,16 @@
 from __future__ import division, absolute_import
 from six.moves import zip
 import numpy as np
+import scipy.optimize


again why not import the one function we need?

See above. Don't.

kain88-de · 2017-06-16T07:54:28Z

package/MDAnalysis/analysis/hole.py

@@ -258,6 +257,10 @@
 import logging
 from itertools import cycle

+import numpy as np
+import matplotlib


a common idom is import matplotlib as mpl

yes, but we don't use it a lot and it's clearer to read and to grep for: if I am looking for matplotlib I use git grep matplotlib.

kain88-de · 2017-06-16T07:55:45Z

package/MDAnalysis/analysis/psa.py

+    :func:`scipy.spatial.distance.directed_hausdorff` is an optimized
+    implementation of the early break algorithm of [Taha2015]_; note that one
+    still has to calculate the *symmetric* Hausdorff distance as
+    `max(directed_hausdorff(P, Q)[0], directed_hausdorff(Q, P)[0])`.


are both still rendered under notes in the sphinx output?

kain88-de · 2017-06-16T07:57:40Z

package/setup.py

          # all standard requirements are available through PyPi and
          # typically can be installed without difficulties through setuptools
          setup_requires=[
-              'numpy>=1.9.3',
+              'numpy>=1.10.4',


did you add this bump to the CHANGELOG? I know we did it a while again and forgot to change setup.py

Yes we did add it to the CHANGELOG alread.

kain88-de · 2017-06-16T07:58:40Z

testsuite/MDAnalysisTests/analysis/test_encore.py

@@ -128,7 +128,7 @@ def test_triangular_matrix():

        multiplied_triangular_matrix_2 = triangular_matrix_2 * scalar
        assert_equal(multiplied_triangular_matrix_2[0,1], expected_value * scalar,
-			err_msg="Error in TriangularMatrix: multiplication by scalar gave\
+            err_msg="Error in TriangularMatrix: multiplication by scalar gave\


please remove the trailing \ and align with the opening brackets in the previous line

I did changes myself

kain88-de · 2017-06-16T20:34:15Z

I pushed more commits to the branch. I don't know why they dont show up for the PR

orbeckst

I really dislike importing functions from external modules because it makes it hard to understand where they come from (and they pollute the name space, but that's minor... until you define similar sounding things yourself).

orbeckst · 2017-06-16T21:01:46Z

package/MDAnalysis/analysis/distances.py

@@ -42,7 +42,7 @@
           'contact_matrix', 'dist', 'between']

 import numpy as np
-import scipy.sparse
+from scipy import sparse


That's ok, I can live with it.

If I take only few things I like fully qualified imports but this one is fine if you like it better.

orbeckst · 2017-06-16T21:07:06Z

package/MDAnalysis/analysis/encore/similarity.py

@@ -176,7 +176,7 @@
 import logging

 import numpy as np
-import scipy.stats
+from scipy.stats import gaussian_kde


Please don't. This looses all the context and when browsing code it is not all clear where this function comes from. Is it defined locally? Is it imported from elsewhere?

I really hate it when used for things not in the standard library.

Please revert.

orbeckst · 2017-06-16T21:07:30Z

package/MDAnalysis/analysis/hbonds/hbond_autocorrel.py

@@ -155,7 +155,7 @@
 from __future__ import division, absolute_import
 from six.moves import zip
 import numpy as np
-import scipy.optimize
+from scipy.optimize leastsq


Please don't do this. See comment above.

orbeckst · 2017-06-16T21:07:49Z

package/MDAnalysis/analysis/pca.py

@@ -106,7 +106,7 @@
 import warnings

 import numpy as np
-import scipy.integrate
+from scipy.integrate import simps


Please don't. See above.

orbeckst · 2017-06-16T21:08:02Z

package/MDAnalysis/analysis/polymer.py

@@ -36,7 +36,7 @@
 from six.moves import range

 import numpy as np
-import scipy.optimize
+from scipy.optimize import curve_fit


Please don't. See above.

orbeckst · 2017-06-18T00:09:23Z

@kain88-de I do appreciate it that you humoured my personal pet-peeve with the imports. Apologies if my comments were not quite as civil as they could have been.

Thanks for working with me on these boring last minute PRs!

orbeckst · 2017-06-18T00:11:33Z

The travis failure is in full because it exceeded 50 mins ( #1394 ...).

orbeckst · 2017-06-18T00:12:24Z

I will clean up the history.

orbeckst · 2017-06-18T01:00:43Z

Compacted history and force-pushed.

Should be ready for review and merge.

kain88-de · 2017-06-19T08:58:27Z

Can we retarget this to 0.17.0 ? This takes consistently longer then 50 minutes to run our full test suite. I don't understand how this change suddenly added more then 10 minutes to our overall runtime.

orbeckst · 2017-06-19T18:45:04Z

It looks to me that the overall +10 minutes comes from the fact that "minimal" is not so minimal anymore – having scipy enables many more tests. Minimal ran for 41 min and full timed out at 50 mins but in other cases full ran for 48:44 min. In the other case, minimal took 25:33 min.

Running the setup (astropy helpers) took 441.98s for the time-out full, 328.76s for the minimal with scipy and 294.30s for the other full that passed but only 137.51s for the other minimal without scipy.

I think we still have a general problem with the run time of full. I didn't quite understand the data in #1409 but if different installation of netcdf4 or travis caching #1405 could help then we should investigate further.

I am not convinced that this PR in particular extends the installation time of full because these installation times have been fairly random as of late; the numbers that I quote above differ by almost 150s for full without any changes in the travis config that would affect the full build.

I would advocate for adding it to 0.16.2 because it is ready and it will make things a bit easier for downstream packaging #1361.

kain88-de · 2017-06-19T19:12:52Z

The runtime of full is too long even with the change in installation times. We used to be at ~30-35 minutes for a full test run. Our installation doesn't suddenly take more then 10 minutes. Also if full regularly runs into the time limit we won't know about our code coverage.

orbeckst · 2017-06-19T19:44:37Z

The runtime of full is too long even with the change in installation times. We used to be at ~30-35 minutes for a full test run.

Yes, true. But if understand you correctly then you're hypothesizing that it is due to this PR. The numbers above show that it is not due to this PR: the full one https://travis-ci.org/MDAnalysis/mdanalysis/builds/244102132 for PR #1403 also takes almost 50 minutes.

Our installation doesn't suddenly take more then 10 minutes.

Fair enough, ~150 s is only 2:30 min. Did you look at what takes more time in the tests?

From the 48 min full:

MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_data.test_Writer_atoms: 161.1134s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_closeContactGNMAnalysis: 145.9985s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_closeContactGNMAnalysis_weights_None: 138.4231s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_conect_bonds_all: 37.0044s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_conect_bonds_conect: 34.5182s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_writer_large: 32.3000s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes_custom_weights: 30.8317s
MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_cnt.test_Writer_atoms: 23.0828s
MDAnalysisTests.analysis.test_pca.TestPCA.test_cov: 22.8741s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_slice_iteration: 20.3056s
MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_data.test_Writer_dimensions: 19.9035s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_n_atoms_frame: 19.4444s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_n_frames: 19.3878s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_numconnections: 19.3644s
MDAnalysisTests.analysis.test_pca.TestPCA.test_transform_universe: 18.7797s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_rewind: 18.7055s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_iteration: 18.2078s
MDAnalysisTests.analysis.test_pca.TestPCA.test_transform: 18.1015s
MDAnalysisTests.analysis.test_pca.TestPCA.test_cosine_content: 17.9944s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes: 17.5463s
MDAnalysisTests.analysis.test_pca.TestPCA.test_different_steps: 17.3986s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_generate_kirchoff: 17.1699s
MDAnalysisTests.analysis.test_encore.TestEncore.test_ces_error_estimation_ensemble_bootstrap: 16.8289s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_write_trajectory_universe: 16.7904s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_write_trajectory_atomgroup: 16.7405s
MDAnalysisTests.analysis.test_pca.TestPCA.test_cum_var: 16.4954s
MDAnalysisTests.analysis.test_pca.TestPCA.test_transform_mismatch: 15.7796s
MDAnalysisTests.analysis.test_pca.TestPCA.test_pcs: 15.6358s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes_align: 14.8079s
MDAnalysisTests.analysis.test_hbonds.TestHydrogenBondAnalysis.test_true_traj: 14.3645s
MDAnalysisTests.analysis.test_hbonds.TestHydrogenBondAnalysisHeuristic.test_true_traj: 13.9034s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_writer_large_residue_count: 13.4650s
MDAnalysisTests.analysis.test_hbonds.TestHydrogenBondAnalysisHeavy.test_true_traj: 13.0224s
MDAnalysisTests.analysis.test_hbonds.TestHydrogenBondAnalysisHeavyFail.test_true_traj: 12.7882s
MDAnalysisTests.core.test_atomselections.TestSelectionsXTC.test_same_fragment: 12.6707s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_gnm: 12.6454s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes_error_estimation: 12.2850s

If you compare this to the minimal with scipy

MDAnalysisTests.analysis.test_gnm.TestGNM.test_closeContactGNMAnalysis: 153.0787s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_closeContactGNMAnalysis_weights_None: 102.6995s
MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_data.test_Writer_atoms: 95.6346s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes_custom_weights: 60.4777s
MDAnalysisTests.analysis.test_pca.TestPCA.test_cov: 29.9189s
MDAnalysisTests.analysis.test_pca.TestPCA.test_cosine_content: 26.2424s
MDAnalysisTests.analysis.test_pca.TestPCA.test_pcs: 25.6768s
MDAnalysisTests.analysis.test_pca.TestPCA.test_transform: 24.8249s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_writer_large: 24.0293s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes_align: 23.4301s
MDAnalysisTests.analysis.test_pca.TestPCA.test_transform_universe: 23.1988s
MDAnalysisTests.analysis.test_pca.TestPCA.test_different_steps: 23.1815s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes: 22.9368s
MDAnalysisTests.analysis.test_encore.TestEncore.test_hes_to_self: 22.7151s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_conect_bonds_all: 20.8662s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_conect_bonds_conect: 20.3900s
MDAnalysisTests.analysis.test_pca.TestPCA.test_transform_mismatch: 18.8019s
MDAnalysisTests.analysis.test_pca.TestPCA.test_cum_var: 18.0902s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_generate_kirchoff: 14.8774s
MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_cnt.test_Writer_atoms: 13.6212s
MDAnalysisTests.analysis.test_gnm.TestGNM.test_gnm: 13.1033s
MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_data.test_Writer_dimensions: 12.9685s
MDAnalysisTests.analysis.test_encore.TestEncoreDimensionalityReduction.test_dimensionality_reduction_three_ensembles_two_identical: 12.6368s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_write_trajectory_atomgroup: 12.1999s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_write_trajectory_universe: 12.0097s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_n_frames: 11.8778s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_numconnections: 11.1740s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_iteration: 11.1318s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_n_atoms_frame: 11.1067s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_slice_iteration: 11.1007s
MDAnalysisTests.coordinates.test_pdb.TestMultiPDBReader.test_rewind: 10.9571s
MDAnalysisTests.analysis.test_hbonds.TestHydrogenBondAnalysisHeavyFail.test_true_traj: 10.2214s
MDAnalysisTests.coordinates.test_gro.TestGROLargeWriter.test_writer_large_residue_count: 10.0398s

There is a lot of variation: for instance MDAnalysisTests.coordinates.test_lammps.TestLAMMPSDATAWriter_data.test_Writer_atoms took either 161.1134s or 95.6346s. One GNM test differs by 30s. This is admittedly not an exhaustive comparison but it hints to variation in the run time on travis.

Also if full regularly runs into the time limit we won't know about our code coverage.

Yes, and that is definitely bad. Perhaps we need to split the build matrix into core and analysis and report coverage separately (or find a way to merge across builds).

orbeckst · 2017-06-20T23:30:49Z

@MDAnalysis/coredevs this needs a review, thanks.

jbarnoud

Except for the scipy version, everything looks OK.

jbarnoud · 2017-06-21T07:56:09Z

package/setup.py

@@ -494,11 +494,12 @@ def dynamic_author_list():
          classifiers=CLASSIFIERS,
          cmdclass=cmdclass,
          requires=['numpy (>=1.10.4)', 'biopython', 'mmtf (>=1.0.0)',
-                    'networkx (>=1.0)', 'GridDataFormats (>=0.3.2)', 'joblib'],
+                    'networkx (>=1.0)', 'GridDataFormats (>=0.3.2)', 'joblib',
+                    'scipy', 'matplotlib (>=1.5.1)'],


Hausdorff distance was introduced in scipy 0.19.0, maybe we should specify the minimum version.

We still have our own hausdorff distance I think. You can open an issue for switching to the scipy implementation with a fallback to our own.

I read a comment to fast; I thought we switched to the scipy version. My bad.

mnmelo · 2017-06-21T09:36:31Z

@orbeckst, you didn't add yourself to the version's author list in the CHANGELOG.

kain88-de · 2017-06-21T09:43:04Z

@orbeckst, you didn't add yourself to the version's author list in the CHANGELOG.

I added Oliver

richardjgowers · 2017-06-21T09:52:30Z

testsuite/MDAnalysisTests/analysis/test_distances.py

-    def setUp(self):
-        sys.modules.pop('MDAnalysis.analysis.distances', None)
-
-    @block_import('scipy')


I was so proud of this decorator, and now we hardly need it 😆

Ah, worry not, I'm opening a PR in 5 minutes where it is very useful!

richardjgowers · 2017-06-21T09:54:16Z

Is adding dependencies allowed in a 16.x release or does it have to be 0.17? Otherwise looks good

kain88-de · 2017-06-21T09:59:28Z

Is adding dependencies allowed in a 16.x release or does it have to be 0.17? Otherwise looks good

Well with strict following of SemVer it is not. Even deprecating publib API would have to be in 0.17.0 and not 0.16.2. http://semver.org/

I personally think this is less of an issue then the deprecations.

kain88-de · 2017-06-21T12:35:22Z

Tests are passing now. the full build just does a timeout like usual.

orbeckst · 2017-06-21T15:06:25Z

The FAQ says that changes in dependencies can be a minor or a patch level release; the latter if it fixes a bug. Good to know about the depredations supposed to be in a new minor. The most important thing is that there is another minor release between deprecation and removal. But we cannot afford that much time for Timeseries removal, the other deprecations are for 1.0 and users will have time to adjust. Nevertheless, because we are still in 0.x we are, at least technically, allowed to do whatever we want within semver. However, we are trying to conform as much as possible/feasible to help users.

…

-- Oliver Beckstein email: [email protected]

Am Jun 21, 2017 um 02:59 schrieb Max Linke ***@***.***>: Is adding dependencies allowed in a 16.x release or does it have to be 0.17? Otherwise looks good Well with strict following of SemVer it is not. Even deprecating publib API would have to be in 0.17.0 and not 0.16.2. http://semver.org/ I personally think this is less of an issue then the deprecations.

- updated all modules - removed any code that guards against scipy or matplotlib import - conforms to style guide https://github.com/MDAnalysis/mdanalysis/wiki/Style-Guide#module-imports-in-mdanalysisanalysis - fixes #1159 - fixes #1361

…ssing

kain88-de · 2017-06-22T07:10:37Z

If there are now more issues with this PR I will merge it in the afternoon

tylerjereddy · 2017-06-30T23:30:46Z

I note that my normal development workflow has been disrupted by alterations to the installation process, with development branch python setup.py install --user resulting in (truncated):

                        * The following required packages can not be built:
                        * freetype, png
error: Setup script exited with 1

The issue is resolved by doing this first: conda install matplotlib

Perhaps another dev can try a fresh conda environment without matplotlib to see if they can reproduce this. Granted, it is on WSL, but that's basically just ubuntu. If you can reproduce this, maybe open an issue if it seems annoying enough, but I guess the conda install is a simple enough fix.

orbeckst · 2017-07-01T06:09:25Z

When I pip install anything it nowadays comes as a whl and I don't notice the pain of compiling matplotlib. I think I never managed to build matplotlib from source with pip. That had been one of the main reasons why I used to advocate for light dependencies. But at least my experience (on OSX and Linux) has become that installation of standard packages has become easy.

Have you tried pip install --user matplotlib in a fresh virtualenv?

orbeckst added this to the 0.16.x milestone Jun 16, 2017

orbeckst requested review from mtiberti, kain88-de, sseyler, tylerjereddy and richardjgowers June 16, 2017 04:18

orbeckst changed the title ~~Issue 1159 analysis deps~~ make scipy and matplotlib required dependencies (#1159) Jun 16, 2017

orbeckst force-pushed the issue-1159-analysis-deps branch from 7552db8 to cf0e31a Compare June 16, 2017 06:24

kain88-de reviewed Jun 16, 2017

View reviewed changes

kain88-de previously requested changes Jun 16, 2017

View reviewed changes

orbeckst commented Jun 16, 2017

View reviewed changes

kain88-de force-pushed the issue-1159-analysis-deps branch 2 times, most recently from a40dbf8 to 259cdbd Compare June 17, 2017 18:38

orbeckst force-pushed the issue-1159-analysis-deps branch from 259cdbd to f8d5c5e Compare June 18, 2017 00:59

orbeckst added the needs review label Jun 18, 2017

orbeckst requested a review from jbarnoud June 20, 2017 23:29

orbeckst mentioned this pull request Jun 20, 2017

release 0.16.2 #1401

Closed

11 tasks

jbarnoud approved these changes Jun 21, 2017

View reviewed changes

kain88-de force-pushed the issue-1159-analysis-deps branch from f8d5c5e to a07d665 Compare June 21, 2017 09:05

kain88-de force-pushed the issue-1159-analysis-deps branch 2 times, most recently from d378937 to 85c0c56 Compare June 21, 2017 09:49

richardjgowers self-assigned this Jun 21, 2017

richardjgowers reviewed Jun 21, 2017

View reviewed changes

mnmelo mentioned this pull request Jun 21, 2017

WIP: Adds support for lazy and optional importing of dependencies #1415

Closed

kain88-de mentioned this pull request Jun 21, 2017

Release 0.16.2 #1421

Merged

orbeckst and others added 4 commits June 22, 2017 07:06

make scipy and matplotlib full dependencies (#1159)

4eab969

removed conditional skipping of tests when scipy or matplotlib are mi…

6779240

…ssing

minor clean ups

f2f45f8

kain88-de force-pushed the issue-1159-analysis-deps branch from 85c0c56 to f2f45f8 Compare June 22, 2017 05:07

kain88-de merged commit 853447c into develop Jun 22, 2017

kain88-de deleted the issue-1159-analysis-deps branch January 24, 2018 19:19

make scipy and matplotlib required dependencies (#1159) #1404

make scipy and matplotlib required dependencies (#1159) #1404

Conversation

orbeckst commented Jun 16, 2017

PR Checklist

Choose a reason for hiding this comment

kain88-de left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kain88-de commented Jun 16, 2017

orbeckst left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

orbeckst commented Jun 18, 2017

orbeckst commented Jun 18, 2017

orbeckst commented Jun 18, 2017

orbeckst commented Jun 18, 2017

kain88-de commented Jun 19, 2017

orbeckst commented Jun 19, 2017

kain88-de commented Jun 19, 2017

orbeckst commented Jun 19, 2017

orbeckst commented Jun 20, 2017

jbarnoud left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mnmelo commented Jun 21, 2017

kain88-de commented Jun 21, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

richardjgowers commented Jun 21, 2017

kain88-de commented Jun 21, 2017

kain88-de commented Jun 21, 2017

orbeckst commented Jun 21, 2017 via email

kain88-de commented Jun 22, 2017

tylerjereddy commented Jun 30, 2017

orbeckst commented Jul 1, 2017