Fix transform slowness

[OfekShilon]

The source of transform slowness is this call to deparse on large inputs, from within name_dots. This seems to be a column-naming-convenience utility aimed at other scenarios (perhaps meld of data tables?) and in the call from transform its results are dropped altogether.

It is possible to bypass this costly call entirely by some dedicated argument passed to the data.table ctor and premeated to name_dots via ..., but a simpler and probably good enough solution would be to cap deparse at a single line by adding nlines=1 argument.

Impact:
(1) Prefix:

> df <- data.frame(x = runif(n = 1e7))
> dt <- as.data.table(df)
> system.time(df <- transform(df, y = round(x)))
   user  system elapsed 
  0.122   0.031   0.152 
> system.time(dt <- transform(dt, y = round(x)))
   user  system elapsed 
 19.658   0.157  19.808 

(2) Postfix:

> df <- data.frame(x = runif(n = 1e7))
> dt <- as.data.table(df)
> system.time(df <- transform(df, y = round(x)))
   user  system elapsed 
  0.147   0.001   0.147 
> system.time(dt <- transform(dt, y = round(x)))
   user  system elapsed 
  0.268   0.039   0.308 

Still twice as slow as transform.data.frame but nevertheless ~60x speedup. And probably enough for the (hopefully) few uses of transform.data.table.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (685d51d) 97.48% compared to head (0bacebc) 99.51%.

❗ Current head 0bacebc differs from pull request most recent head 1e03fe7. Consider uploading reports for the commit 1e03fe7 to get more accurate results

Additional details and impacted files

@@             Coverage Diff             @@
##           1-15-99    #5493      +/-   ##
===========================================
+ Coverage    97.48%   99.51%   +2.02%     
===========================================
  Files           80       80              
  Lines        14862    14763      -99     
===========================================
+ Hits         14488    14691     +203     
+ Misses         374       72     -302     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ColeMiller1]

Good find! A more direct approach that seems to work is to change this:

data.table/R/data.table.R

Line 2304 in 0895fa2

ans = do.call("data.table", c(list(`_data`), e[!matched]))

to:

ans = as.data.table(c(`_data`, e[!matched]))

On my computer, it is slightly slower than base (~5% to 10% slower). As far as the PR, a news item would be needed. I'm not sure there's a framework for performance tests, so a test might not be needed.

[MichaelChirico]

Thanks @OfekShilon and @ColeMiller1 for the fix! Merging into a 'dev' branch that will become master as soon as 1.15.0 hits CRAN.

@OfekShilon PTAL at the NEWS, I edited it a bit given my understanding, we can fix in follow-up if I mis-paraphrased.