CLI support systems with no display

matplotlib.pyplot throws a RuntimeError when no display is present.  One
way around that is to set the backend to AGG.  But we don't want AGG when
we do have a display.  So we only want to set it opportunistically.  But
it also has to be set before anything else is imported.  Kind of convoluted.

So we really just import pyplot as opportunistically as possible, and then
rely on setting AGG backend when we're sure we don't need it in main().

gwinc/__main__.py

@@ -3,7 +3,6 @@ import os
 import signal
 import argparse
 import numpy as np
-from matplotlib import pyplot as plt
 from IPython.terminal.embed import InteractiveShellEmbed
 
 import logging
@@ -136,6 +135,18 @@ def main():
                 v = float(v)
             range_params[p] = v
 
+    if args.save:
+        # FIXME: this silliness seems to be the only way to have
+        # matplotlib usable on systems without a display.  There must
+        # be a better way.  'AGG' is a backend that works without
+        # displays.  but it has to be set before any other matplotlib
+        # stuff is imported.  and we *don't* want it set if we do want
+        # to show an interactive plot.  there doesn't seem a way to
+        # set this opportunistically.
+        import matplotlib
+        matplotlib.use('AGG')
+    from matplotlib import pyplot as plt
+
     ##########
     # main calculations
 

gwinc/plot.py

 from numpy import sqrt
-import matplotlib.pyplot as plt
-
-import logging
 
 PRIORITY_MAP = {
     'Quantum Vacuum': 0,
@@ -80,6 +77,7 @@ def plot_noise(
     f = noises['Freq']
 
     if ax is None:
+        import matplotlib.pyplot as plt
         fig = plt.figure()
         ax = fig.add_subplot(1, 1, 1)