accessing rows via a factor column throws error

[mplatzer]

this used to work in data.table v1.9.4 but now throws an error in v1.9.6:

dt <- data.table(i=1:10, f=as.factor(1:10))
dt[f==3]

the thrown error is: Error in bmerge(i, x, leftcols, rightcols, io <- FALSE, xo, roll = 0, : x.'f' is a factor column being joined to i.'V1' which is type 'double'. Factor columns must join to factor or character columns.

the error can be avoided by adding additional conditions, e.g. dt[f==3 & TRUE] or by comparing to characters (as indicated by the error itself): dt[f=="3"]. not sure, whether this is a bug or a feature, but I though it's worthwhile reporting.

[ghost]

Just found something similar. I had working code with the previous CRAN release, so something must have been changed in 1.9.6.

set.seed(1)
DT <- data.table(V1=factor(c(1,2)),
                 V2=LETTERS[1:3],
                 V4=1:12)


# does not work
DT[V1 == 1, ]

# does work
DT[V1 == factor(1), ]

While in 1.9.4:

DT[V1 == factor(1), ]
Error in Ops.factor(V1, factor(1)) : level sets of factors are different

[caneff]

Running into this too. This seems to be very similar to isssue #957 but with factors instead of characters.

[caneff]

The problem seems to be that on line 471 of data.table.R, you check

mode(x[[isub2]]) != mode(RHS)

However, both factors and integers have mode "numeric", so it skips this part and moves on to the standard auto-indexing, where bmerge chokes. I'm not sure what the right substitution is for line 471 but that is where the fix should go.

[caneff]

Ping? This is breaking our existing code so we are stuck with 1.9.4 until this gets fixed.

[ghost]

Here http://stackoverflow.com/questions/32934400/gather-multiple-times-at-once someone mentions in the comments that this problem can be circumvented by turning off auto indexing in v1.9.6.