The original plan was to always update plot objects in place, by
calling line.set_data(), quad.set_array(), etc. This is not
possible for AMR plots since the number of patches changes from
one frame to the next.
So far, we have 3 major types of plots with conflicting requirements:
- Single-patch matplotlib plot: can pre-specify axes, update in-place.
- Multi-patch matplotlib plot: can pre-specify axes, but cannot update in-place since number and extent of plot objects varies from one frame to next
- yt plot: cannot pre-specify axes or figure, but can update in-place. In-place update is required for animation!
So, we must use in-place updates for yt, we could use it for single-patch, and we can't use it for multi-patch.
Decision for now: for matplotlib we always clear the axes and generate new plot objects, while for yt we always update in-place.
When to convert from Solution to ds for yt plots? Only inside plot_item, so that we can assume we're dealing with Solutions almost everywhere.
Unlike in visclaw, this loop is at the lowest level, inside the loop over
items. We can do this because the matplotlib argument zorder can be used
to avoid patch plots of different items covering each other.
- plot subset of all frames
- image comparison tests (using matplotlib's decorator)
- test for fill_between
- test for show_patch_boundaries
- test for axis titles
- caching data across multiple items
- automatically get and use field names from Riemann
- set up coverage and coveralls on Travis