allow codec to be none for old versions of ffmpeg

[aestrivex]

The save_movie() function calls ffmpeg with the -c argument which breaks very old versions of ffmpeg such as versions which can be found on old installations of RHEL/CentOS. This PR allows the omission of the -c argument when the codec kwarg is None.

[larsoner]

Would be more DRY to do something like:

cmd = ['ffmpeg', '-i', frame_fmt, '-r', str(framerate)]
if codec is not None:
    cmd += ['-c', codec]
cmd += [dst]

[aestrivex]

OK, done that way.

[larsoner]

I assume you've tested this and it works for your code?

[aestrivex]

Yes it works, it is very slow to update though for brains with a large number of vertices.

I tried it on fsaverage5 and it was quite a lot faster.

[larsoner]

I wonder where the bottleneck is... feel free to open an improvement PR if you find a better way. We spent a bit of time getting the API right, but changing the under-the-hood mechanics shouldn't be too bad from an API standpoint. Maybe MoviePy would speed things up.

[coveralls]

Coverage decreased (-0.16%) to 72.43% when pulling f47af04 on aestrivex:allow_no_codec_ffmpeg_compat into 7d7373b on nipy:master.

[aestrivex]

Part of the bottleneck may be in repeatedly saving images, but I think a far larger bottleneck is in the large amount of interpolation done for a large amount of vertices (which the user can adjust currently).

I have some other code (which is not trivially portable to non X systems) which captures movies from mayavi by using ffmpeg to do screen capture rather than repeatedly taking screenshots. I tried both methods, and found that doing the direct screen capture does save a lot of overhead. My implementation might have been inefficient though, I will try it here and see if it does any better.

[larsoner]

We do want to keep compatibility with other platforms, though, so I'm not sure that would end up working.