why not have input/output volumes

[sjpotter]

Currently, it seems the way that one has to transfer data from the input VM image image to the output container image is for the detectives to tar up data and for then have it be extracted in the triggered provisioner.

The issue is that this only lets one easily copy data that is within a single tree, not something where one has to pick and choose files.

It further prevents easy modification of the artifact that is created.

For instance, take the demo case with Apache. If one wants to modify the Apache instance, one either has to extract the configuration tarball, modify it and recreate it, or add another layer just to handle that configuration change.

Instead of dealing with output tarballs, wouldn't it be easier to create output volumes where data can be copied to during the detective stage and leveraged during the provisioning stage (i.e. instead of an ADD <tarball> / it be ADD <directory that was volume> /

I this way, one could

1. more easily programatically decide what will be fetched

2. enable one's created artifact to be checked into a repo, modified as needed over time, and rebuilt on demand.

thoughts?