Disk checkpoint storing all time step data

[Ig-dolci]

No description provided.

[jrmaddison]

Is there a use case for storing the restart dependencies with a SingleDiskStorageSchedule? This does let you replay the forward, but is there a reason to do that?

[jrmaddison]

Suggested change

	def __init__(self, move_data=False, store_only_adj_deps=False):
	def __init__(self, move_data=False, *, store_only_adj_deps=True):

Default False would be an API break, changing existing behaviour. New arguments should be keyword only.

[Ig-dolci]

Yes. I just updated it.

[Ig-dolci]

Is there a use case for storing the restart dependencies with a SingleDiskStorageSchedule? This does let you replay the forward, but is there a reason to do that?

It is nothing related to SingleDiskStorageSchedule itself. The goal here is to make SingleDiskStorageSchedule more flexible for use cases like pyadjoint, where an adjoint dependency data set cannot always be correctly constructed during the taping process. See this related pyadjoint issue

[jrmaddison]

The goal here is to make SingleDiskStorageSchedule more flexible for use cases like pyadjoint, where an adjoint dependency data set cannot always be correctly constructed during the taping process. See this related pyadjoint issue

Thanks, I see. There's nothing inherently wrong with adding a store_ics argument here if that's helpful, although it might fix the pyadjoint bug for the wrong reason. I'll comment there.