varying domains

[fabian-s]

• useful for indefinite integrals, maybe: have functions that represent integrals over different intervals

• necessary for function fragment data

• potentially re-useable for Gellar's vd-stuff in refund

• does that even make sense for fbase or should it be restricted to fevals?

edge cases for scalar fevals:

• in https://github.com/fabian-s/tidyfun/commit/0e1b335dc014f044fa47634c141c2b3316873107 feval_irreg of length 1 is treated as vd because it makes subsetting etc more easy

• this breaks c()'ing of a single irreg to a regular and vice versa because the check is too simplistic.

• should "scalar" feval_regs be converted to feval_irreg by default (....no, I guess!) But then need a lot of if-else for subsetting and c'ing

More generally:

• none of this works/will generalize for functions with multivariate argvals in its present form
• look at intervals package for abstraction layer / infrastructure for merging/comparing intervals?

[fabian-s]

... should also think about other types of support/domain -- e.g.

• proper Dates
• periodic data with circular domains
• (images on spheres to represent global data)

[fabian-s]

• not going to implement varying domains
• raw data on circular domain does not make sense (e.g. 24h accelerometry)
• summaries should just use periodic bases for that kind of data