- User Since
- Apr 3 2020, 4:04 PM (42 w, 2 d)
Sat, Jan 23
I like the variants you listed that involve overloading the input_defs arg on solid. I think that kind of overloading is pretty Pythonic / expected. When reading that code, it's very difficult to misinterpret what's going on.
@schrockn those concerns make a lot of sense.
Fri, Jan 22
@schrockn do you have thoughts here? I think I would be OK with any of the options raised by @alangenfeld. I vaguely recall that you had some concerns about getting fancy with auto-creation/parameterization of types.
It would be nice to have a better story for these named tuples. Unfortunately, it might require mucking around with metaclasses to support overriding init if we want to do runtime type checking.
Thu, Jan 21
Wed, Jan 20
what led away from the approach in https://dagster.phacility.com/D5784 ? Just updating test call sites?
I think just raising the outer error and improving our display / printing of it makes the most sense to me.
Tue, Jan 19
I agree that, given that converting from ObjectStoreOperation to HandledOutput / LoadedInput would be lossy, probably best to punt.
Sat, Jan 16
Fri, Jan 15
shouldn't this be modeled w/ IOManagers instead of Nothings?
Thu, Jan 14
Wed, Jan 13
Looks good. Left a comment.
That was quick! Love the big red portion.
oops meant to hit accept
Tue, Jan 12
CC @schrockn in case he wants to weigh in.
A couple small comments - otherwise looks great!