Skip to content

manage R function and data dependencies like 'make'

3 messages · Lutz Prechelt, Brian Ripley, Paul Gilbert

Original
Lutz Prechelt
# 
Dear R-devels,

I am looking for a package (or some other infrastructure) for the
following situation:

I am doing a larger data evaluation.
I have several dozen data files and do not want to keep the persistent
data in the R workspace (for robustness reasons).
I have several dozen R files, some for reading and preprocessing data
files, others for doing plots or analyses.
I will make frequent changes to both data files and R files while doing
the analysis.

I would like to automate mechanisms that allow 
- a data file reading function to suppress its actual work if neither
the data file nor the R file containing the function were modified since
the data file was last read
- an R file sourcing function to suppress its actual work if the R file
has not been modified
- and perhaps even: automate re-reading a data file upon access to the
corresponding dataframe iff the file has been modified since the
dataframe was created.

In short: Something like Unix's 'make', but for managing dependencies of
functions and dataframes in addition to files. In R. (And of course I am
very open for solutions that are more elegant than what I have sketched
above.)

I could not find something in the help and have rather few ideas for
good search terms.

I any such thing available?
(If no such infrastructure exists, what is the right R function for
accessing file modification dates?)

Thanks!

  Lutz
Brian Ripley
# 
R-devel has file_test() in utils (earlier versions had a private version 
in tools).  That has a '-nt' op to do what you need.

file.info() accesses modification dates.

Having said that, I would use 'make' as for example R's own test suites 
do.
On Wed, 7 Feb 2007, Lutz Prechelt wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Paul Gilbert
    

    

      
      #
    

  


  

      
I use make to do things like you describe. Here is an example target 
from one of my Makerules files:

$(T:%=compare.R$(REP).T%):compare.%: compare.R estimates.%
	@echo making $(notdir $(PWD)) $@ because $? changed ...
	@(cd  $(subst compare.,,$@) ; \
	    echo "z <- try( source('../compare.R')); \
	  if (!inherits(z, 'try-error')) q('yes', status=0) else \
	   {print(z); q('yes', status=1)} "  | \
	      R --slave >../$(FLAGS)/$@.log 2>&1)
	@mv $(FLAGS)/$@.log $(FLAGS)/$@


I realize this is out of context and possibly mangled by mail wrap, but 
if you are familiar with (GNU) make then it should give you a good idea 
what to do. In this example I am accumulating (intermediate) things in 
the .RData file and using flags as targets to indicate the status of 
different steps. Another possibility would be to rename the .RData files 
and use them as the targets. Let me know if you want a more complete 
example.

I really would encourage you to think of wrapping R (like a compiler) in 
make, rather than trying to re-implement something like make within R.

(I would be interested to see examples if anyone is using Ant to do this 
kind of thing.)

Paul

        
Lutz Prechelt wrote:

            

      
      
      
    

        

      
      
    

        
====================================================================================

La version fran?aise suit le texte anglais.

------------------------------------------------------------------------------------

This email may contain privileged and/or confidential inform...{{dropped}}