Muhammad Ali and Adam Foster prototyped a dashboard display for DrProject as a term project this winter. It collects information on the number of tickets in various states, and the number of check-ins, then uses Matplotlib to construct time-series charts. It’s pretty cool, so Ali agreed to spend a week integrating into DrProject for deployment.
In order to make it work, DrProject has to load Matplotlib and PySqlite (the Python wrappers for SQLite) at the same time. After a day of wrestling with versions, packages, LD_LIBRARY_PATH, and other junk, Ali can get it to load one or the other, but never both. The symptoms are described here:
Basically, if LD_LIBRARY_PATH is left empty, Matplotlib will load, but PySqlite won’t. If LD_LIBRARY_PATH is set to /usr/local/lib (where the sqlite .so file is located), PySqlite is happy, but Matplotlib fails to load. If the directory containing the .so’s that Matplotlib needs are added to LD_LIBRARY_PATH, well, you can read Ali’s message, but it’s still broken.
We’d welcome help or guidance; we’d also be grateful if someone would come up with a packaging and deployment strategy that makes these problems just not happen .
Later: Ali writes, “The problem was that I had an incompatible version of one the packages that matplotlib needs to import (libcairo), installed in /usr/local/lib. By adding usr/local/lib to LD_LIBRARY_PATH, the system loaded up the incompatible libcairo, instead of its regular one. The fix was ‘uninstalling’ all incompatible packages (libcairo and libpango in my case) from /usr/local/lib. I’ve also learnt that in Linux terminology uninstalling is the same as ‘rm -f’.”