Define asynchronous traversal rules

[maxfischer2781]

Chain traversal currently uses an arbitrary but strictly synchronous traversal method that is piecewise depth-first. For asynchronous, lazy and parallel traversal, an asynchronous model is required. Ideally, this does preserver the result of synchronous traversal.

[maxfischer2781]

The current behaviour is depth-first and sensitive to join/fork.

1 >> 2 >> (3 >> 4, 5 >> 6) >> 7

• Each parent is traversed for all input before proceeding to the child.
  • All input is passed to 1 before the result is passed to 2.
  • All input is passed to the fork (...) before the result is passed to 7.
  • Implementation: all output is materialised using a list before passing this on.
• Each peer is visited at once before proceeding to the next element.
  • All input is passed to 3 >> 4 before 5 >> 6.
  • Implementation: all input is pushed into elements before passing on to the next element.
• Multiple output is collected in order.
  • The input to 7 is a flat sequence of the output of 3 >> 4 followed by 5 >> 6.

[maxfischer2781]

Notes for a/synchronous [A/S] traversal:

• Data passed between elements is either 1-type or n-type.
  • 1-types are always individual, arbitrary objects
  • n-types are iterators [S: list]
    • for parallel chains multiple iterators are required for the same data, see itertools.tee