Add `never` as a possible `time_arg` value.

[EvanKirshenbaum]

It's really stupid that in order to use interactive.py, in order for it not to disappear immediately, you need to specify a --min-time argument with a fixed, but large value (e.g., 1day).

It would make more sense for time_arg to allow something like never to return an infinite Time (or at least a hugely large one). This will require ArgUnits to require a second map parallel to patterns (which maps unit names to unit expressions for a given dimension) holding, for each dimension, a mapping of names to quantities.

The default value for --min-time, currently a fixed 5*minutes, should instead be taken from the Task. Note that --min-time is added in Exerciser.add_common_args_to(), so this will require that this (called in Exerciser.setup_task()) get passed the Task.

Migrated from internal repository. Originally created by @EvanKirshenbaum on Feb 04, 2023 at 3:04 PM PST.

[EvanKirshenbaum]

This issue was referenced by the following commit before migration:

• f6fc813f5e80428aac84358ce5cad3e45f84b020 by @EvanKirshenbaum on Apr 19, 2023 at 4:54 PM PDT

[EvanKirshenbaum]

Looking at the code, I think it may make more sense to do it fully to refactor cmd_line to actually have separate objects per dimension. Right now, ArgUnits is defined using two static maps:

class ArgUnits:
    known: Final[dict[type[Quantity], dict[str, UnitExpr]]] = defaultdict(dict)
    patterns: Final[dict[type[Quantity], Pattern]] = {}

where all interactions are via class methods that require a type[Quantity] argument to say which dimension they're talking about. (The exception is units_arg, which runs through all of ArgUnits.known).

What I propose to do is to turn ArgUnits into something like DimArg[Q], where DimArg has a mapping of known instances, one per quantity type, each containing their own units, constants, patterns, and default strings for error messages. With this mapping, we can also do the functionality of units_arg by querying all of the instances.

If we give DimArg a __call__(str) method, we should be able to do something like either

voltage_arg = DimArg[Voltage](units={ ... })

or, possibly better,

voltage_arg = DimArg.for_dim(Voltage)
voltage_arg.add_units({ ... })

This is closer to what's done now, and has the advantage that there can be multiple calls, possibly even in user code. Yeah, I think that's the way to go.

To implement constants, you can just have a mapping from string to quantity (added to by passing in a mapping from quantity to sequence of strings), e.g., {"never": 1000000*day} that's looked at first in parse_arg().

Migrated from internal repository. Originally created by @EvanKirshenbaum on Apr 18, 2023 at 4:51 PM PDT.

[EvanKirshenbaum]

I haven't added the notion of constants, but I have enabled the "Let a task say it should live forever" behavior we actually need for the problem at hand.

I changed things so that default_min_time can now be None. In this case, the task should live forever, unless --max-time (or default_max_time) is a Time. I also moved all of these defaults onto the Task rather than being in Exerciser, so that Task.default_min_time can be 0*ms while DisplayOnly.default_min_time can be None.

I further made the logic a bit more flexible by adding --hold-display-for, which specifies a minimum time to keep the display open once the Task ends.

Finally, I added logging, so that the user knows when things will go down.

Migrated from internal repository. Originally created by @EvanKirshenbaum on Apr 19, 2023 at 5:27 PM PDT.