Directly support multiple inputs and outputs

[RutgerK]

By lack of a mailing-list or anything i'm just going to write an idea here as an issue, i hope that's alright.

The issue (https://github.com/sgillies/rasterio/issues/6) by AsgerPeterson about iterating over the blocks provides a nice example of how to open and write block for block.

Would it be within the scope of RasterIO to extent this to support multiple input and output datasets with a similar interface? It introduces a lot of potential problems like projection differences, performance hits if the blocks aren't the same etc.

A really simple implementation i made a year ago demonstrates this idea, see for example:
http://nbviewer.ipython.org/gist/RutgerK/7941518

My implementation is full of assumptions of the datasets being identical in terms of properties, making it just a very simple wrapper to eliminate the gdal.Open() overhead.

So to modify Asger's example a bit:

with rasterio.open('source1.tif',
                   'source2.tif') as src:

    kwargs = src.meta
    with rasterio.open(('destination1.tif', 'w', **kwargs),
                       ('destination2.tif', 'w', **kwargs),
                       ('destination3.tif', 'w', **kwargs)) as dst:

        for window in src.get_chunks( 1 ):
            data = src.read_band( 1, window)
            dst['destination1'].write_band( 1, data['source1'] + data['source2'], window)
            etc.

In some cases you could achieve a similar effect by combining inputs in a VRT, but this doesn't work for outputs for example.

[sgillies]

@RutgerK: it's an interesting idea, but I think it complicates rasterio and is better left out of the library. I suspect it would be less code to extend rasterio yourself with a new function that took a list of filenames and made itself into a context manager for all those files, closing them on exit, etc.