matplotlib backend set in /algorithms/spot_finding/per_image_analysis.py

[asmit3]

Context

While trying to debug code in dials.stills_process, I was unable to show plots using matplotlib when I put a breakpoint in the code (atleast not interactively). The reason seems to stem from

dials/algorithms/spot_finding/per_image_analysis.py

Lines 8 to 15 in 63d2fef

	try:
	import matplotlib
	# http://matplotlib.org/faq/howto_faq.html#generate-images-without-having-a-window-appear
	matplotlib.use("Agg") # use a non-interactive backend
	from matplotlib import pyplot
	except ImportError:
	pyplot = None

Here the matplotlib backend is set as part of a global import which turns off interactive plotting. Thus when a program imports something from this file, the global import lines for matplotlib turn off interactive plotting (this is what's happening with dials.stills_process). In dials.stills_process this is happening indirectly through imports made from indexer.py --> max_cell.py

Question

I was wondering if there is a reason for declaring the backend at the top in per_image_analysis.py? Can anything be done to not set the backend here and instead do so when plotting is done within the program ?

Just for demonstration, here is a barebones program that does not currently show plots. This is similar to the case in dials.stills_process

from libtbx.phil import parse

phil = parse("""
  include scope dials.algorithms.indexing.indexer.index_only_phil_scope
""", process_includes=True).extract()
from matplotlib import pyplot as plt; plt.plot(range(10), range(10)); plt.show()

[Anthchirp]

I think it is reasonable to move the entire import matplotlib down to the actual plotting code blocks. Then I would also drop the try:/except:/Sorry bit as the import is required and any ImportError can be dealt with upstream. That way we wouldn't even import matplotlib when the code in per_image_analysis.py is used without plotting.
@rjgildea, @ndevenish what do you think?

[Anthchirp]

Doing this would make mocking harder, but I don't see any test directly importing this file, so there is no immediate downside as far as I can see.

[ndevenish]

Reopening, because it seems this is throwing lots of warnings because matplotlib versions/support and backends. Need to have another look.

Might be worth backing out change and putting the other one in for now.

[ndevenish]

I've switched to the moving commit 5f59558 which hopefully solves without the issues of the other one.