在 R4.0.0 中合并数据帧比 R3.6.3 慢 60 倍？

Question

Merging 2 data frames in R4.0.0 is much slower than in R.3.6.3.

Reproducible example :

library(reshape2)

n <- matrix(1:6000000, nrow=15000, ncol=400)
colnames(n) <- 1:ncol(n)

meta <- data.frame("col1" = ncol(n):1, row.names = colnames(n))
start_time <- Sys.time()
test = sapply(1:nrow(n), 
               function(i) {
                 print(i)
                 nn = reshape2::melt(n[i,])
                 tmp = merge(nn, meta, by="row.names");
               }
)
end_time <- Sys.time()
end_time-start_time

This code takes 23 seconds in R3.6.3 and 23 minutes in R4.0.0 (on my machine) - making the merge 60 times slower . The 'melt' function is not the time consuming step.

This issue is also not related to sapply. You can reproduce the extreme speed difference using a for loop:

for(i in 1:nrow(n)){
  print(i)
  nn = reshape2::melt(n[i,])
  tmp = merge(nn, meta, by="row.names");
}

I'd be happy to hear your feedback? Am I missing something?

(I have previously asked this question in another context and have deleted the old question since it wasn't reproducible)

Answer 1

This is actually a bug in R4.0.0.

I have submitted it to R-core ( https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17794 ) and update here once it is fixed.

Update The bug has been fixed by the R-team (thanks to Martin Maechler,), is already rolled out in R-devel. and the fix will be included in the next release version of R.