by loops that should be possible fail because dogroups grows beyond 2**31

[Kodiologist]

> data.table(i = 1:100)[, verbose = T, by = i, {message(".GRP: ", .GRP); list(v = logical(1e7))}]
Detected that j uses these columns: <none> 
Finding groups using forderv ... forder.c received 100 rows and 1 columns
0.000s elapsed (0.000s 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 '{'
Old mean optimization is on, left j unchanged.
Making each group and running j (GForce FALSE) ... .GRP: 1
.GRP: 2
dogroups: growing from 10000000 to -2147483648 rows
Error in `[.data.table`(data.table(i = 1:100), , verbose = T, by = i,  : 
  negative length vectors are not allowed

I know that data.table still doesn't support long vectors (#3957), but this operation should only require 100 * 1e7 = 1e9 rows, which is well under 2**31 - 1 ≈ 2.1e9. And if the user really does ask for too many rows, the error message can surely be improved.

Output of sessionInfo()

R version 4.2.1 (2022-06-23)                                                    
Platform: x86_64-pc-linux-gnu (64-bit)                                          
Running under: Ubuntu 20.04.4 LTS                                               
                                                                                
Matrix products: default                                                        
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.9.0                         
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.9.0                            
                                                                                       
locale:                                                                                       
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C                                                           
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8                                                           
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8                                                          
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                                                                        
 [9] LC_ADDRESS=C               LC_TELEPHONE=C                                                                   
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C                                                              

attached base packages:                                 
[1] stats     graphics  grDevices utils     datasets  methods   base                                             

other attached packages:                                
[1] data.table_1.14.8                                   

loaded via a namespace (and not attached):                                                                       
[1] compiler_4.2.1

[ben-schwen]

This is triggered by the internal estimate of rows of the result, which exceeds the 2**31-1 in your example. Subsequently, all column vectors are grown to that estimated length.

Apart from supporting long vectors which would include basically touching all C-files, we could try to improve that estimate

[Kodiologist]

It might be easier and more reliable to just cap the estimate at 2**31 - 1.

[ben-schwen]

Just capping the estimate at 2**31-1 unfortunately just kills the R process (if the cap is exceeded) which doesn't seem like an improvement to me

[Kodiologist]

But the cap wouldn't be exceeded here, right? So wouldn't it make this code work? Better error messages for trying to make a data table that has too many rows would also be helpful, clearly, but that's supplementary.

[ben-schwen]

In your example, it definitely works and gives the right output.

But slightly changing your example to data.table(i = 1:220)[, verbose = T, by = i, {message(".GRP: ", .GRP); list(v = logical(1e7))}] kills the R process at group 147