gmedian failing on simple case

[caneff]

On 1.10.4:

dt <- data.table(x=c(1,1,1),y=c(3,0,0))
setkey(dt, x)
dt[, median(y), by=x]

Yields the following error:
Error in gmedian(y) : negative length vectors are not allowed

If dt doesn't have a key set, it doesn't throw the error (but I can't tell if gmedian is even being called in that case). Also calling stats::median instead of median above works because it bypasses gmedian directly.

[franknarf1]

I'm running 1.10.4 on R 3.2.5 and don't see that error. To see if gmedian is being used, dt[, median(y), by=x, verbose=TRUE] .. not sure if that's what you meant. There's also the global option datatable.verbose.

Fyi, they prefer that you test on the devel version, currently 1.10.5 https://github.com/Rdatatable/data.table/wiki/Support

[caneff]

Sorry, yes I should have updated to the latest version. Updated to 1.10.5, still happens. But it is weird.

I start with R --vanilla. sessionInfo() yields:

R version 3.2.2 (2015-08-14)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 14.04.5 LTS

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     

loaded via a namespace (and not attached):
[1] gmotd_1.0

When I start it up, I paste in this exact block:

library(data.table)
dt <- data.table(x=c(1,1,1),y=c(3,0,0))
setkey(dt, x)
dt[, median(y), by=x]

Approximately 3/4s of the time I do this, I get the error mentioned above. The other 1/4 of the times I try it works for me. I have ran other code that uncovered this issue for me in the first place on networked machines and it produced the same errors, so I don't think this is some weird memory corruption thing. Thoughts?

[kmillar]

What appears to be happening is that in [.data.table, if 'on' is missing, then it gets set to integer(0), which eventually gets passed down as the 'o' parameter to the C implementation of gforce.

In there we have:
maxgrpn = INTEGER(getAttrib(o, install("maxgrpn")))[0]
but since integer(0) has no attributes, this becomes INTEGER(R_NilValue)[0], which could return anything. When it returns a negative value, we get the error seen above.

While I've been able to track this down, I don't understand the code well enough to fix it. Can someone work out a patch for it?

[MichaelChirico]

Linking this knitr issue I created which may be related (though it seems to be a knitr issue in my case):

yihui/knitr#1457