RFC/WIP: preserve the type of the source upon collect(Enumerable) #3
Conversation
remove redundant eltype() methods
abstract of the enumerable that doesn't transform the elements of the sources but just rearranges/filters/duplicates them
Codecov Report
@@ Coverage Diff @@
## master #3 +/- ##
==========================================
+ Coverage 25.36% 27.82% +2.46%
==========================================
Files 14 14
Lines 347 327 -20
==========================================
+ Hits 88 91 +3
+ Misses 259 236 -23
Continue to review full report at Codecov.
|
alyst
changed the title
|
Ah, ok, this actually doesn't fix the problem with dataframes, because dataframes are handled by IterableTables, so I would need to patch the column type selection in |
add init_sink(Enumerable) method that defaults to Vector{eltype},
but keeps the source type for SimpleSourceEnumerable{T,<:AbstractArray}
|
I just outlined how the original issue could be addressed in the issue you had opened. I don't think any of the other stuff here would be needed, right?
|
|
Closing this for now, I don't think we want to merge this. |
No user-code workaround for queryverse/Query.jl#157 came to my mind, so I decided to propose a fix.
The first commit just adds
Tparam to theEnumerable, which allows reducingeltype()redundancy. I think it's a good change overall.The 2nd commit introduces
SimpleSourceEnumerable{T,S}, which is a type of enumerable that have a single source and doesn't transform elements, e.g. "where" and "orderby".The 3rd commits actually allows
SimpleSourceEnumerableto preserve the type of theAbstractArraysource uponcollect(), which covers theCategoricalArraycase without requiring CategoricalArrays dependency.Actually, it would make sense to try to preserve the source type for the other enumerables if the sink is of the same/similar type as the source(s).
Even if the elements are transformed, but the result is
CategoricalValue, it would make sense to collect them intoCategoricalArray.But that would require more general design or package dependency decisions.