Optimize data transfer between pure python nodes

[chrisgorgo]

Currently when two pure python nodes (like nipy GLM fit and contrast estimate) are connected (A->B) node A saves the data to file and node B reads it. The reading phase could be skipped by sending the data directly to node B. This, however, creates some problems:

1. It will not work with "timestamp" hashing.
2. Big chunks of data will have to be kept in memory until receiving nodes will be executed.

[djarecka]

@chrisfilo - do you have any new comments on this issue?

[chrisgorgo]

Nope.

[satra]

this may be supported as part of efficiency implementation in 2.0 workflow engine, but will require some complex logic to implement properly. currently low priority.