#general
- h
high-advantage-47550
02/17/2017, 5:49 PM(i want to explicitly resolve all conflicting transients) - h
happy-kitchen-89482
02/17/2017, 5:55 PMHi folks (in particular @enough-analyst-54434, who has experience with this…) - h
happy-kitchen-89482
02/17/2017, 5:55 PMI’ve noticed an issue with our codebase, that is currently biting me. - h
happy-kitchen-89482
02/17/2017, 5:56 PMPython, as we know, doesn’t like logical packages being split among multiple directories (something the JVM classloader has no problem with) - h
happy-kitchen-89482
02/17/2017, 5:57 PMHowever we have such cases in our codebase. Specifically, our contrib trees all start with
.pants.contrib - h
happy-kitchen-89482
02/17/2017, 5:57 PMThis leads to fun things like: - h
happy-kitchen-89482
02/17/2017, 5:57 PM
works./pants test contrib/python:: contrib/findbugs:: —tag=-integration - h
happy-kitchen-89482
02/17/2017, 5:57 PMBut
does not./pants test contrib/findbugs:: contrib/python:: —tag=-integration - h
happy-kitchen-89482
02/17/2017, 5:59 PMThis is because some tests in
import fromcontrib/python/tests/python
in the same source rootpython_libraries - h
happy-kitchen-89482
02/17/2017, 6:00 PMbut the python importer fails when it can’t find that imported module in
, which shadowscontrib/findbugs/tests/python/pants/contrib
for the purpose of importing.contrib/python/tests/python/pants/contrib - h
happy-kitchen-89482
02/17/2017, 6:00 PMOr at least this is my interpretation of the problem. - h
happy-kitchen-89482
02/17/2017, 6:00 PMI am no expert on python import logic. - r
rough-minister-58256
02/17/2017, 6:03 PMah, this is the “multiple source roots create packages with overlapping import namespaces” problem? - r
rough-minister-58256
02/17/2017, 6:04 PMultimately it comes down to sys.path ordering - h
happy-kitchen-89482
02/17/2017, 6:05 PMIn fact, I’m confused about how code under
is able to import code from the maincontrib
tree, since they both start withsrc/python
, and we have anpants
there, so it’s a package.__init__.py - h
happy-kitchen-89482
02/17/2017, 6:05 PMYes, but A) how does our code even work, and B) we should fix this, because the phenomenon above is bad - h
happy-kitchen-89482
02/17/2017, 6:06 PMAnd by “fix this” I mean “rename stuff in our codebase” - e
enough-analyst-54434
02/17/2017, 6:11 PMWell, super subtle issues aside, the contents of
is key__init__.py - e
enough-analyst-54434
02/17/2017, 6:12 PMNamespace package - e
enough-analyst-54434
02/17/2017, 6:12 PMJust making sure you're aware of that whole thing - e
enough-analyst-54434
02/17/2017, 6:15 PMIt's cooperative unfortunately so 1 miss ruins all - h
happy-kitchen-89482
02/17/2017, 6:23 PMOh, I am aware… I dived into that whole morass in my first ever Pants commit, back in 2011, when I added ANTLR support, and had it generate namespace package stuff in
...__init__.py - h
happy-kitchen-89482
02/17/2017, 6:23 PMBut all our
are empty, I think__init__.py - h
happy-kitchen-89482
02/17/2017, 6:23 PMAlso, that namespace packaging thing is frowned on by various factions of Pythonistas, and I can see why - h
happy-kitchen-89482
02/17/2017, 6:24 PMAh no - h
happy-kitchen-89482
02/17/2017, 6:25 PMwe have some non-empty__init__.py - h
happy-kitchen-89482
02/17/2017, 6:25 PMweird, I thought we enforced that - h
happy-kitchen-89482
02/17/2017, 6:26 PMOK, so apparently we do use namespace packages... - h
happy-kitchen-89482
02/17/2017, 6:26 PMAnd the python purists among us are OK with this? - h
happy-kitchen-89482
02/17/2017, 6:27 PMIn that case presumably the issue is that one of those two contrib trees isn’t being cooperative