[R/add_data_table] cleanup update_cell()

[JanMarvin]

Frist attempt to speed up update_cell().

The benchmark below is now ~22s (prev ~26s) on my desktop. Mostly spent in update_cell(). Writing the file initial only took 37 ms.

library(microbenchmark)

# a larger data frame with strings, missings and numbers
xlsxFile <- system.file("extdata", "readTest.xlsx", package = "openxlsx2")
wb1 <- wb_load(xlsxFile)
data <- wb_to_df(wb1, sheet = 3, skipEmptyRows = TRUE)

res <- microbenchmark::microbenchmark(
  {
    options("loop" = "Rcpp")
    wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data)
    # Rcpp <- wb$worksheets[[1]]$sheet_data$cc
  }, 
  {
    options("loop" = "R")
    wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data)
    # R <- wb$worksheets[[1]]$sheet_data$cc
  },
  
  times = 1
); res

# all.equal(Rcpp, R)

> res
Unit: seconds
                                                                                                         expr      min       lq     mean   median       uq      max neval
 {     options(loop = "Rcpp")     wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data) } 21.67788 21.67788 21.67788 21.67788 21.67788 21.67788     1
    {     options(loop = "R")     wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data) } 25.29638 25.29638 25.29638 25.29638 25.29638 25.29638     1

Replacing the loop with Rcpp definitively has benefits, though the entire function is still pretty slow.

[JanMarvin]

Once I fixed a couple of my initial thinkos (or better, make use of some functions we have developed since), updating time is now acceptably fast. The Rcpp loop brings nice speed improvements too :)

> library(microbenchmark)
> library(openxlsx2)
> 
> # a larger data frame with strings, missings and numbers
> xlsxFile <- system.file("extdata", "readTest.xlsx", package = "openxlsx2")
> wb1 <- wb_load(xlsxFile)
> data <- wb_to_df(wb1, sheet = 3, skipEmptyRows = TRUE)
> 
> res <- microbenchmark::microbenchmark(
+   {
+     options("loop" = "Rcpp")
+     wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data)
+     # Rcpp <- wb$worksheets[[1]]$sheet_data$cc
+   }, 
+   {
+     options("loop" = "R")
+     wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data)
+     # R <- wb$worksheets[[1]]$sheet_data$cc
+   },
+   
+   times = 1
+ ); res
Unit: seconds
                                                                                                         expr      min       lq     mean
 {     options(loop = "Rcpp")     wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data) } 1.837195 1.837195 1.837195
    {     options(loop = "R")     wb <- wb_workbook()$add_worksheet()$add_data(x = data)$add_data(x = data) } 7.250326 7.250326 7.250326
   median       uq      max neval
 1.837195 1.837195 1.837195     1
 7.250326 7.250326 7.250326     1

[JanMarvin]

I assume that the failing test is just because of whatever R core was doing with gsub

[JanMarvin]

Final benchmark runtime on desktop is ~1.96s. Still a lot longer than creating from scratch, but way better than the 25s from earlier today. Now the entire time is spent in the Rcpp loop (if anyone wants to spend another day improving this).