[Rd] DOCUMENTATION(?): parallel::mcparallel() gives various types of "Error in unserialize(r) : ..." errors if value is of type raw

[Tomas Kalibera]

Thanks for the report, fixed in documentation of mcparallel. Raw vectors 
are reserved for internal use and cannot be returned by expressions 
passed to mcparallel. But one can indeed wrap a raw vector into a list.

FYI this internal use is not in readChild/mccollect, but in 
sendMaster/mcparallel.

Best
Tomas

On 06/21/2018 05:05 AM, Henrik Bengtsson wrote:
> I stumbled upon the following:
>
> f <- parallel::mcparallel(raw(0L))
> parallel::mccollect(f)
> # $`77083`
> # NULL
>
> but
>
> f <- parallel::mcparallel(raw(1L))
> parallel::mccollect(f)
> # Error in unserialize(r) : read error
> traceback()
> # 2: unserialize(r)
> # 1: parallel::mccollect(f)
>
> (restarting because the above appears to corrupt the R session)
>
> f <- parallel::mcparallel(raw(2L))
> parallel::mccollect(f)
> # Error in unserialize(r) : unknown input format
>
> I can reproduce this on Linux using R 3.5.0 all the way back to R
> 3.3.2 (didn't try further) and R development (unstable) (2018-06-19
> r74919).
>
> Diving into the code of parallel::mccollect(), it looks that the 'raw'
> data type has a special purpose:
>
>                    r <- readChild(pid)
>                    if (is.integer(r) || is.null(r))
>                      fin[pid == pids] <- TRUE
>                    if (is.raw(r))
>                      res[which(pid == pids)] <- list(unserialize(r))
>
> However, I don't see this behavior document in ?parallel::mcparallel.
> The closest I get is that under the 'Value' section it says:
>
>     "mccollect returns any results that are available in a list. [...]"
>
> Not sure if that should be interpreted as only values that are of type
> 'list' should be used.  OTH, the example use values of type double.
>
> /Henrik
>
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel