Rakudo: HyperIterables ignore loop phasers

[gfldex]

my @a = (^10).list; @a = @a.hyper(:batch(5)).map({ LAST say "last"; „∅“ });
# rakudo-moar d10d69774: ( no output )

# expexted output at the end of a batch:
# last

# FIRST and NEXT are ignored too
# It may be sensible to fail on LAST in a HyperIterable and provide a
# LAST-BATCH phaser to avoid confusion when non-parallel code is
# changed.

[lizmat]

FWIW, I’m making all phasers throw a NYI for the 2017.10 release. So we can figure out semantics until the next release.

On 22 Oct 2017, at 13:21, Wenzel P. P. Peppmeyer [e-mail] wrote:

my @a = (^10).list; @a = @a.hyper(:batch(5)).map({ LAST say "last"; „∅“ });

rakudo-moar d10d69774: ( no output )

expexted output at the end of a batch:

last

FIRST and NEXT are ignored too

It may be sensible to fail in LAST in a HyperIterable and provide a

LAST-BATCH phaser to avoid confusion when non-parallel code is

changed.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.

[AlexDaniel]

↑ https://github.com/rakudo/rakudo/commit/a042fd927c672cc2bf6d186e2d40f3ae69d88b86

[jnthn]

They're simply NYI (which is the case for many things with hyper/race). FIRST should fire on the first element, and LAST on the last element, just as with a non-hyper loop. Batches shouldn't come in to it. The batch with the first element carries a flag indicating that, and it's the same for the batch with the last element. I don't think that we should need to make any guarantee that FIRST will run before iterations in another worker, nor that LAST will run after, since a loop with such order dependencies couldn't be parallelized anyway.

I suspect most situations where a LAST-IN-BATCH could be useful would generally be better done with a race reduce, which will be one of the next race-related things I implement (if nobody beats me to it).

[chee]

could this issue be considered closed now that it throws a X::NYI?

[lizmat]

Yes, thanks for the nudge!