updated PYTHON_INSTALL_DIR generation #636
Conversation
|
unfortunately the Travis CI build seems to fail due to a shortcoming of python's virtualenv: |
|
looks like Travis CI was okay with that one. requesting human review now. |
|
Thanks for the pull request. As you can probably imagine, we'll have to scrutinize this one pretty hard since it affects all Python packages in ROS. I'll give some feedback soon. |
|
I can confirm that the provided patch fixes #635. Thanks!! |
|
I would like to mention that this patch did not work for me on gentoo
|
|
@spaghetti- Can you please also try the follow up PR #831. And if it doesn't work provide some more information what the problem is, what the various paths are in your case in order to allow iterating on the patch. |
|
Yes of course. However @dirk-thomas is the right approach having |
|
There can't be more than one location for a single package. Should it needs to be all in either |
|
I will close this as a duplicate of #831. Please continue the discussion there. |
This is preparation for ros#636. I noticed that there is duplicated code for loading options for the offline and GRPC offline node because they are needed while constructing the map builder for the non-GRPC offline node (and that step is the only difference between the offline node and the GRPC offline node). I got around this by passing a map builder factory to `RunOfflineNode` instead, so we can deduplicate the code for loading options by doing it inside `RunOfflineNode`.
In response to #632 #635
This patch uses Python to report PYTHON_INSTALL_DIR, and it therefore allows cmake/python.cmake and python/catkin/builder.py to share most of the same code. It eliminates the debian-specific directive and works with python2 and python3. It has been tested on Gentoo, Debian, and Windows.
More testing is in order to confirm its complete functionality.