#general
- e
enough-analyst-54434
07/21/2015, 6:48 PMie:
to both turn on junit xml reporting and control where the reports landed./pants test.junit --xml-dir=dist/test/ - e
enough-analyst-54434
07/21/2015, 6:51 PMHey @early-mechanic-59271 - what languages are folks using pants for at Medium? - e
early-mechanic-59271
07/21/2015, 6:53 PMRight now, java/scala/python - e
early-mechanic-59271
07/21/2015, 6:53 PMPython is primarily internal tools and ML apps - e
enough-analyst-54434
07/21/2015, 6:53 PMAha - ok. - e
early-mechanic-59271
07/21/2015, 6:53 PMScala for Spark - e
early-mechanic-59271
07/21/2015, 6:54 PMNot much java - e
early-mechanic-59271
07/21/2015, 6:54 PMOverall code quantity wise, most code is javascript and go. - e
enough-analyst-54434
07/21/2015, 6:54 PMHave you had issues with the numpy family in the ML apps or is that family not used? - e
early-mechanic-59271
07/21/2015, 6:54 PMYes! numpy/scikit-learn/matplotlib and friends. - e
early-mechanic-59271
07/21/2015, 6:54 PMThey’re… interesting simple smile - e
enough-analyst-54434
07/21/2015, 6:54 PMYeah - not fun - e
enough-analyst-54434
07/21/2015, 6:55 PMOK - I've done some work with this and I'm in general interested in making that ecosystem work better with pants/pex, even if its just recipes - e
enough-analyst-54434
07/21/2015, 6:55 PMI had one partner / bug-filer, but he gave up after ~3 days - r
rough-minister-58256
07/21/2015, 6:55 PMI’ve never had much issue pex’ing numpy et al with the standalone pex tool - is it just that they don’t follow standard setup.py conventions? e.g. they implement their own dependency fetching out of band etc? - e
early-mechanic-59271
07/21/2015, 6:55 PMAn idea we’ve tossed around is how pex stops at the pyhton interperter. It would be interesting to build a container with all the system libraries too. - e
enough-analyst-54434
07/21/2015, 6:56 PMYeah - actually do pars - e
enough-analyst-54434
07/21/2015, 6:56 PMMy work ended here: https://github.com/pantsbuild/issues-1751 - e
enough-analyst-54434
07/21/2015, 6:56 PM@rough-minister-58256: yeah - that's exactly the problem though - e
enough-analyst-54434
07/21/2015, 6:56 PMpeople just expect the distribution to build, and it doesn't for several reasons - e
enough-analyst-54434
07/21/2015, 6:57 PMSo its totally tractable - if you know what you're doing - e
enough-analyst-54434
07/21/2015, 6:58 PMAnd it seems to be the truth of tools - at least build tools - that people don't really want to know what they're doing. Can't blame them - e
early-mechanic-59271
07/21/2015, 6:59 PMSeems like these ML python libraries have a large number of native code dependencies, so the python folks want to stay away from the build and leave in the hands of distro providers. - e
enough-analyst-54434
07/21/2015, 7:00 PMagreed - e
early-mechanic-59271
07/21/2015, 7:00 PMTo get our setup working, I installed numpy and friends to the local system to get all the native libraries installed. - e
early-mechanic-59271
07/21/2015, 7:00 PMThen in a virtualenv I `pip wheel numpy`’d and installed that, then did the others to build the wheels. - r
rough-minister-58256
07/21/2015, 7:01 PMmaybe the problem is that by default most laptops just don’t build c-libs correctly ootb - e
enough-analyst-54434
07/21/2015, 7:01 PMFor OSX this seems to be true - r
rough-minister-58256
07/21/2015, 7:01 PMI have a pex lib I use for adhoc data viz that I just created with ‘pex <lib> <lib> <lib>’ and it built flawlessly in my already-setup env - r
rough-minister-58256
07/21/2015, 7:01 PM$ unzip -l ~/bin/seaborn.pex |grep ".deps" |awk '{print $4}' |awk -F/ '{print $2}' | sort | uniq | awk -F- '{print $1}' | perl -pe 's/\n/ /g' matplotlib mock nose numpy pandas pyparsing python_dateutil pytz scipy seaborn six