data.table .internal.selfref ptr is NULL case using foreach

Start by running the supplied reproduction from bug.R with data.table, foreach, and doMC, then compare the %dopar% and %do% cases. Trace the .internal.selfref warning and determine whether the behavior belongs to data.table or foreach; done means the ownership and expected behavior are established with a regression test or documented resolution.

Hello I came across the following (bug?), I do not know if it comes from foreach or data.table or if I do something stupid.

library(data.table)                                            
library(foreach)                                               
library(doMC)                                                  
registerDoMC(cores = 2L)                                       
                                                               
options(datatable.verbose = TRUE)                              
DT <- data.table(x=1:2, y = 2:3, z = c("a", "b"))              
DT_list <- split(DT, by = "z")                                 
foreach(DT_i = DT_list) %dopar%                                
{                                                              
    fun <- function(DT) return(DT[, c("x", "y"), with = FALSE])
    res_i <- fun(DT_i)                                         
} -> res_list                                                  
res_list[[1]][,test := 1]     

$ R --vanilla -e 'source("bug.R")'         
> source("bug.R")          
Loading required package: iterators                    
Loading required package: parallel                     
Processing split.data.table with: x[j = list(.ll.tech.split = list(.SD)), by = "z", .SDcols = c("x", "y", "z")]
Finding groups using forderv ... 0.046s elapsed (0.573s cpu)                                                  
Finding group sizes from the positions (can be avoided to save RAM) ... 0.000s elapsed (0.000s cpu)           
lapply optimization is on, j unchanged as 'list(list(.SD))'                                                   
GForce is on, left j unchanged                         
Old mean optimization is on, left j unchanged.         
Making each group and running j (GForce FALSE) ...     
  memcpy contiguous groups took 0.000s for 2 groups    
  eval(j) took 0.000s for 2 calls                      
0.000s elapsed (0.002s cpu)                            
Finding groups using forderv ... 0.000s elapsed (0.001s cpu)                                                  
Finding group sizes from the positions (can be avoided to save RAM) ... 0.000s elapsed (0.001s cpu)           
lapply optimization is on, j unchanged as 'list(paste(unlist(lapply(.SD, as.character)), collapse = "."))'    
GForce is on, left j unchanged                         
Old mean optimization is on, left j unchanged.         
Making each group and running j (GForce FALSE) ...     
  memcpy contiguous groups took 0.000s for 2 groups    
  eval(j) took 0.000s for 2 calls                      
0.001s elapsed (0.006s cpu)                            
Detected that j uses these columns: <none>
 .internal.selfref ptr is NULL. This is expected and normal for a data.table loaded from disk. Please remember to always setDT() immediately after loading to prevent unexpected behavior. If this table was not loaded from disk or you've already run setDT(), please report to data.table issue tracker.
Assigning to all 1 rows    
RHS_list_of_columns == false                           
RHS for item 1 has been duplicated because NAMED is 2, but then is being plonked. length(values)==1; length(cols)==1)

I am using dev : data.table 1.12.3 IN DEVELOPMENT built 2019-07-16 10:38:13 UTC

Note that replacing %dopar% by %do% solves the issue so it may be a foreach bug ?
Only thing I am surprised about is that it is the first time I notice this and I usee foreach and data.table all the time