Integer cycling tasks for iterative processes

[cylc]

NIWA has a river catchment model calibration ensemble, which involves iterating an ensemble of tasks - same tasks, same input files, but with different content in the input files each time, and no cycle times.

Currently this can be done in cylc by (a) using Jinja2 to generate a suite of asynchronous tasks including all tasks for all iterations from the outset., which isn't ideal for large ensembles and many iterations; or (b) using a cycling suite in which the cycle times are meaningless, which isn't ideal because the user has to use date-time arithmetic to compute the final cycle time, and all task logs will be stamped with cycle times.

It would be better to do this with a cycling task type that just increments an integer tag instead of incrementing a date-time based cycle time. In principle this will be easy. Cylc already has a "repeating asynchronous" type for satellite processing, but these have arbitrary tags determined at run time.

Integer cycling tasks should be implemented as another cycling task type in which successive instance tags lie on an integer sequence instead of a date-time sequence. The corresponding cycler module will be quite trivial. The only difficultly - and this could be a bit messy - is to check for and generalize any assumption of date-time tag values wherever cycling tasks are processed throughout the code.

[hjoliver]

This will be addressed with #119, as integer cycling is an analogue of general date time cycling as far as the interface to cylc internals is concerned.

[hjoliver]

This has been done on the ISO8601 feature branch - we just need to review the notation, and add a few tests.

[hjoliver]

we just need to review the notation, and add a few tests

I'll try to get to this soon, but it's not critical for cylc-6.

[hjoliver]

Note comments on notation:

• clock trigger offset as duration #1059 (comment)
• clock trigger offset as duration #1059 (comment)
• clock trigger offset as duration #1059 (comment)
• clock trigger offset as duration #1059 (comment)

[hjoliver]

Closing this; remaining minor issues taken by #1129 and #1130.