[Rd] Display lines of code from the top-level script or subscript in non-interactive R Session with Rprof

[Duncan Murdoch]

On 03/08/2022 4:44 a.m., Alexander Keth via R-devel wrote:
> Hello there,
> 
> 
> I am running R in a production environment. My goal is to profile all production jobs, which are run in non interactive R sessions via Rscript, in the form job-xyz ran for xxx amount of time and spend yyy seconds with code execution of line # (for every line of code). In general the R code is run with a main script which calls various subscripts. The jobs make heays use of external packages (e.g. dplyr, DBI, data.table and so on).
> 
> I re-installed all packages with --with-keep.source. Subscripts are sourced in the main-script via eval(parse("path/to/subscript.R")) to enable line-profiling with Rprof. The call to Rprof is Rprof("rprof.out", line.profiling = TRUE, memory.profiling = TRUE).
> 
> 
> Unfotunately, the majority of the code relies on heavy package use (e.g. dplyr, data.table and so on). Thus most of the code lines in Rprof refer to the source-code within those packages and not the 'top-level' source code in the main-script or the subscripts. So far the only solution I came up with is to scrape the Rprof output using the profile package (https://github.com/r-prof/profile), extract the top-level call stack function calls (remove top level eval calls before) and auto-magically match the function calls with the function calls performed in the main-script and subscripts. However, this process is obviously not perfect and very error prone...
> 
> 
> Is there any better way to do things?

I think reinstalling uninteresting packages --with-keep-source was a 
mistake.  If you use the standard builds of those, and only keep source 
in your own code, most of the detail will come from there.

I'm not familiar with the profile package, but the utils::summaryRprof() 
function with `lines = "show"` will give a display that concentrates on 
the timing by line in the code that has source references. I think from 
your description you want to look at the "by.total" table, but maybe you 
want the "by.line" table.

Duncan Murdoch