palladin
commented
9 years ago I'll start by removing some of them. The three main elimination criteria are 1) Sequentially 2) Global synchronization 3) On demand consumption (Pull based) In other words CloudFlow combinators must be Parallel and Distributed friendly and Push based by nature.
Hard-difficult combinators:
- xxx Back combinators are not Parallel friendly
- mapi (xxxi), xxxIndex, skip, scan are order sensitive and very expensive in a Parallel and Distributed setting
- xxxWhile functions require global synchronization
- zip , xxxn (like map2, zip3 etc) are Pull based (zip and mapxx combinators are very easy to be implemented as producers but not for intermediate step combinators. Maybe in the future we can use intermediate Buffers/Queue to implement some of them as intermediate combinators).
Easy combinators: compareWith, concat, contains, where, average, sumBy, mapFold, max, min, sort, append, truncate, chunkBySize
That's all for now!
As @dsyme suggested in mbraceproject/MBrace.Core#100, it would be good to add more functions to CloudFlow.
Here is a list of most of the F# 4.0 collection functions and which are already implemented for CloudFlow, if we could trim this to a list of functions that would be useful to add to CloudFlow then people could get started on implementing them.
appendaverageByaveragechoosechunkBySizecollectcompareWithconcatcontainscountBydistinctBydistinctemptyexactlyOneexceptexists2existsfilterfindBackfindIndexBackfindIndexfindfold2foldBack2foldBackfoldforall2forallgroupByheadindexedinitisEmptyitemiter2iteriteri2iterilastlengthmap2map3mapFoldBackmapFoldmapmapi2mapimaxBymaxminByminnthpairwisepermutepickreduceBackreducereplicaterevscanBackscansingletonskipWhileskipsortByDescendingsortBysortDescendingsortWithsortsplitIntosumBysumtailtakeWhiletaketruncatetryFindBacktryFindIndexBacktryFindIndextryFindtryHeadtryItemtryLasttryPickunfoldwherewindowedzip3zip