> We have other selectors available for RunArray specific semantic, like aRunArray values do:, aRunArray runsAndValuesDo:

Side note: If you see any real use case for [aRunArray values do:], please make a forwarding selector #valuesDo: for it. I really don't want to encourage LoD violations. Alternatively, provide #runsAndValues as a dictionary? Hm ...

Le jeu. 2 janv. 2020 à 22:09, Jakob Reschke <forums.jakob@resfarm.de> a écrit :

Am Do., 2. Jan. 2020 um 21:12 Uhr schrieb Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com>:

If we strictly adhere to this interpretation of method comment, detectMin: detectMax: detect:ifNone: and count: should also iterate on each and every element.

Do you agree that count: hardly benefits from such definition?

Instinctively I wanted to reply that the optimization in RunArray does not change the result of these operations contrary to collect: because they do not change elements, so equal elements could not suddenly become non-equal elements during the operation. But it got me thinking that I cannot imagine these usefully with side effects in the blocks... :-)

What about do: and inject:into: instead? Would you evaluate their blocks only once per run?

No.

It would mean changing the semantics.

I don't want to change the semantics, just the implementation.

We have other selectors available for RunArray specific semantic, like aRunArray values do:, aRunArray runsAndValuesDo:

Side effects in collect: and select: (and count: etc...) are border line.

Either we expect that those methods must enumerate exactly like do: in order to support side effects uniformely.

Your reaction perfectly demonstrates that this can be a legitimate expectation.

Or we consider that the enumeration is implementation defined, as long as the methods maps, filter, (count etc...).

I think that RunArray has a greater value in itself if we adopt the later, but it's only my own opinion so far.