[Rd] wrap_logical warning message when loading objects created in R 3.6 in an R 3.5 session

[Duncan Murdoch]

On 23/08/2019 4:22 p.m., petr smirnov wrote:
> Hi,
> 
> I am experiencing a warning message when I load a large R object
> created in an R 3.6 session in R 3.5.* , as follows:
> 
> ```
> Warning message:
> In load(“GDSCv2.RData”) :
>   cannot unserialize ALTVEC object of class ‘wrap_logical’ from package
> ‘base’; returning length zero vector
> ```
> Of relevant information may be that the large R object (a data
> structure defined in my Bioconductor package PharmacoGx), was in part
> created including data frames which were cast from data.tables. I
> noticed that the ALTVEC class had caused some errors previously in the
> data.table package.
> 
> I have two questions:
> 1. Should I be concerned about this warning? I cannot seem to find
> what effect it has on the data loaded.

That object is not being loaded.  If the object is important to you, 
then you should be concerned.

> 2. Could you point me towards narrowing down the cause of this issue?

 From time to time the format of saved objects changes.  There was such 
a change in version 3.5.0 of R. Older versions cannot read files in a 
format that was invented after they were released.

I would have expected 3.5.0 to be able to read all the current formats, 
but it appears you have an object new to 3.6.0 that needs something that 
doesn't exist in 3.5.0.  The solution is probably to use 3.6.x to read 
that file, and then save it with "version=2" as an argument.  That 
should cause it to use the older format.

Duncan Murdoch


> Ideally, everyone would upgrade R promptly, but even our own
> institute's HPC cluster is still on 3.5, and the warning does not
> inspire confidence for some of the less technical members of our group
> who are using the datasets.