Hi Ben thanks for sharing, looks interesting. I have never used transpose, so I was wondering what is the typical use case? with lists? The man page says

     This is particularly useful in tasks that require splitting a
     character column and assigning each part to a separate column.
     This operation is quite common enough that a function tstrsplit
     is exported.

I have used tstrsplit, so I guess I understand the use case with character, but I wonder if you could add an illustrative example, which would show how transpose is different / more flexible / useful with lists ?

The main use cases I have in my mind is when you receive a rowwise format or want to lapply some function rowwise. Not promoting to maximum type but returning lists would be favorable there to avoid additional casting.

dt = data.table(name=c("Anna", "Bob"), age=c(30,20))
fun = function(x) sprintf("Hi %d year old %s", x[[2]], x[[1]])
transpose(dt, list.cols = TRUE)[, lapply(.SD, fun)]
#                     V1                 V2
#                 <char>             <char>
# 1: Hi 30 year old Anna Hi 20 year old Bob
transpose(dt, list.cols = FALSE)[, lapply(.SD, fun)]
# Error in sprintf("Hi %d year old %s", x[[2]], x[[1]]) : 
#   invalid format '%d'; use format %s for character objects

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.50%. Comparing base (8de09b2) to head (d0d8285).

@@           Coverage Diff           @@
##           master    #5805   +/-   ##
=======================================
  Coverage   97.50%   97.50%           
=======================================
  Files          80       80           
  Lines       14876    14884    +8     
=======================================
+ Hits        14505    14513    +8     
  Misses        371      371           

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

Hi Ben thanks for sharing, looks interesting. I have never used transpose, so I was wondering what is the typical use case? with lists? The man page says

     This is particularly useful in tasks that require splitting a
     character column and assigning each part to a separate column.
     This operation is quite common enough that a function tstrsplit
     is exported.

I have used tstrsplit, so I guess I understand the use case with character, but I wonder if you could add an illustrative example, which would show how transpose is different / more flexible / useful with lists ?

transpose() is the workhorse behind tstrsplit, tstrsplit is "just" strsplit()|>transpose().

but there are many other times we wind up with a list() of rows that we want as columns instead. base R code looks like lapply(seq_along(l), \(j) sapply(l, `[[`, j)).

transpose() is not only much cleaner but much faster and handles type conversion and fills ragged data

transpose() is the workhorse behind tstrsplit, tstrsplit is "just" strsplit()|>transpose().

but there are many other times we wind up with a list() of rows that we want as columns instead. base R code looks like lapply(seq_along(l), \(j) sapply(l, `[[`, j)).

transpose() is not only much cleaner but much faster and handles type conversion and fills ragged data

Can an example of this usage (and new usage) be please added to transpose.Rd in this PR?
(it would help me at least)

FWIW here are benchmarks version vs base solution

l = list(sample(1e5), sample(letters, 1e5, TRUE))
dt = bench::mark(
  base = lapply(seq(length(l[[1]])), function(x) lapply(l, `[[`, x)),
  dt = transpose(l, list.cols=TRUE),
  iterations = 100
)
setDT(dt)
dt[, .(expression, min, median, "max" = lapply(time, max), `itr/sec`, n_itr, n_gc), .I]
#        I   expression          min       median    max   itr/sec n_itr  n_gc
#    <int> <bench_expr> <bench_time> <bench_time> <list>     <num> <int> <num>
# 1:     1         base      270.3ms      333.9ms  769ms  2.773482   100   241
# 2:     2           dt       15.5ms       23.3ms  315ms 29.969829   100    29

@ben-schwen base was run only two times while PR was run 16 times. Therefore I would look for a "max" statistic (which is missing) rather than min or median (presented in your post). You may find this post relevant: jangorecki/rollbench#1 (comment)
I would go with, ideally, system.time() + scaled up data, or eventually microbenchmark.

if you want to see how time/memory vary with N, please try using atime tdhock/atime#15

above shows that both base R and dt are linear time and memory, different by constant factors.

above shows the data size N which both can handle, given 1 second time limit or 1000kb memory limit. (dt is faster by constant factors but uses a constant factor more memory)

list.cols=TRUE also works for data.table (/frame) input, though it's not clear from the current PR if that's intentional.

As I mentioned in the updated NEWS, transposing a table to get it in row-major form is indeed quite common & useful. This is especially useful for I/O with non-column-major data structures (as is common for streaming applications which receive one row at a time). As noted in the linked issue, the workarounds were quite tedious before, after this PR, we can quite easily just transpose(<row-major>) |> lapply(unlist) |> setDT() to get a nice data.table out of something we receive, and transpose(<column-major>, list.cols=TRUE) to convert an R table into something to pass along.

In short, this feature is quite great and something I "didn't know I needed" all along!

would be great to add vignette documentation for this use case "As noted in the linked issue, the workarounds were quite tedious before, after this PR, we can quite easily just transpose() |> lapply(unlist) |> setDT() to get a nice data.table out of something we receive, and transpose(, list.cols=TRUE) to convert an R table into something to pass along"

would be great to add vignette documentation for this use case "As noted in the linked issue, the workarounds were quite tedious before, after this PR, we can quite easily just transpose() |> lapply(unlist) |> setDT() to get a nice data.table out of something we receive, and transpose(, list.cols=TRUE) to convert an R table into something to pass along"

Do you have a specific vignette in mind? I don't see any mention of transpose() in the current vignettes. Maybe what we really need is a datatable-utils vignette demonstrating transpose() as well as fcase(), fifelse(), ... odds-and-end-functions we have