Debug

R

How to debug an R code

• How to Easily and Efficiently Conquer Errors in Your Code
• Chapter 11 Debugging R code from the book What They Forgot to Teach You About R
• Debugging in R – How to Easily Overcome Errors in Your Code?. Examples are provided.

• R debugging tutorial (youtube) and accompanying document in github. options(error = recover) is useful! We can return to default error handling with options(error = NULL).
  • Note that options(error = recover) only works in an interactive session.
  • In an non-interactive mode, see traceback() for interactive and non-interactive R sessions

    options(error=function() { traceback(2); if(!interactive()) quit("no", status = 1, runLast = FALSE) })
    

• browser() and other resources

trace() and browser(), findLineNum() and setBreakpoint()

• Temporarily add browser() to a function. R debugging tutorial. trace(pkg:::foo, edit = TRUE) & untrace(). Suppose we run classpredict::classPredict() and got an error where the traceback() or options(error = recover) shows the error came from classpredict:::GetPredictionResults(). We can do the following. This is useful if the bug happened in an un-exported function from some package, not just in our R code.

  undebug(classPredict) # not necessary now
  
  trace(classpredict:::GetPredictionResults, edit = TRUE)
  # add browser() to a place we suspect the error came from
  # click 'Save' to exit
  
  classPredict(...)
  # hopefully we found the reason
  
  untrace(classpredict:::GetPredictionResults)
  

• trace(what, tracer, exit, at, print, signature, where = topenv(parent.frame()), edit = FALSE)

• browser(text = "", condition = NULL, expr = TRUE, skipCalls = 0L)

  # It still runs like for-loop
  lapply(1:3, function(i){ browser(); rnorm(i)})
  

• findLineNum() and setBreakpoint() if we use source() to get the functions. Or eval(parse(text=x)).
• as.list(body(myFun)).

• Debugging in R – How to Easily Overcome Errors in Your Code? See the trace() example there. It seems the function definition line (e.g. function(x)) means at=1.

  set.seed(5656)
  trace_test <- function(x){
    r <- log(x)
    if(r>=10){
      return(r)
    } else{
      return(2*r)
    }
  }
  val <- rnorm(50)
  trace_test(val)
  
  trace(trace_test,quote( if(any(is.nan(r))){browser()} ), at=3, print=F)
  trace_test(val)
  
  # Note that the following 'at' option will result in an error
  trace(trace_test,quote( if(any(is.nan(r))){browser()} ), at=2)
  trace_test(val)
  # Error in eval(expr, p) : object 'r' not found
  
  # fix of the above trace() example
  trace(trace_test,quote( if(any(x < 0)){browser()} ), at=2)
  trace_test(val)
  

• Chapter 11 Debugging R code > trace(). It is not necessary to use the at option.
• trace() inserts arbitrary code at any position in an existing function. trace() is occasionally useful when you’re debugging code that you don’t have the source for. More complicated than debug(). To insert a browser() function, we don't need to modify the original function. browser() debug statement R

  myFun <- function() {
      x <- 8:1
      y <- 1:8
      plot(y~x)
      lines(y~x)
      text(x,y, letters[1:8], pos=3)
  }
  as.list(body(myFun)) 
  # [[1]]
  # `{`
  # 
  # [[2]]
  # x <- 8:1
  # 
  # [[3]]
  # y <- 1:8
  # 
  # [[4]]
  # plot(y ~ x)
  # 
  # ... More ... 
  trace(myFun, browser, at = 4)
  myFun()   
  untrace(myFun)
  

  There are several examples in ?trace().

  ##  Very simple use. Not useful
  trace(sum)
  hist(rnorm(100)) # shows about 3-4 calls to sum()
  untrace(sum)
  
  ## Show how pt() is called from inside power.t.test():
  trace(pt) ## would show ~20 calls, but we want to see more:
  trace(pt, tracer = quote(cat(sprintf("tracing pt(*, ncp = %.15g)\n", ncp))),
        print = FALSE) # <- not showing typical extra
  power.t.test(20, 1, power=0.8, sd=NULL)  ##--> showing the ncp root finding:
  untrace(pt)
  
  f <- function(x, y) {
      y <- pmax(y, 0.001)
      if (x > 0) x ^ y else stop("x must be positive")
  }
  ## arrange to call the browser on entering and exiting
  ## function f
  trace("f", quote(browser(skipCalls = 4)),
        exit = quote(browser(skipCalls = 4)))
  trace("f", quote(if(any(y < 0)) yOrig <- y),
        exit = quote(if(exists("yOrig")) browser(skipCalls = 4)),
        print = FALSE)
  
  ## Enter the browser just before stop() is called.  First, find
  ## the step numbers
  untrace(f) # (as it has changed f's body !)
  as.list(body(f))
  as.list(body(f)[[3]]) # -> stop(..) is [[4]]
  trace("f", quote(browser(skipCalls = 4)), at = list(c(3,4)))
  f(-1,2)
  
  ## trace a utility function, with recover so we
  ## can browse in the calling functions as well.
  trace("as.matrix", recover)
  
  ## turn off the tracing (that happened above)
  untrace(c("f", "as.matrix"))
  

How to quit debugging

press Q and then ENTER

list or undebug all debugged functions

https://stackoverflow.com/a/12807637

Using assign() in functions

For example, insert the following line to your function

 assign(envir=globalenv(), "GlobalVar", localvar)

options(error)

options(
  error = rlang::entrace, 
  rlang_backtrace_on_error = "branch")

Debug lapply()/sapply()

• https://stackoverflow.com/questions/1395622/debugging-lapply-sapply-calls
• https://stat.ethz.ch/R-manual/R-devel/library/utils/html/recover.html. Use options(error=NULL) to turn it off.

Debugging with RStudio

• https://www.rstudio.com/resources/videos/debugging-techniques-in-rstudio/
• https://github.com/ajmcoqui/debuggingRStudio/blob/master/RStudio_Debugging_Cheatsheet.pdf
• https://support.rstudio.com/hc/en-us/articles/205612627-Debugging-with-RStudio

Debug R source code

Build R with debug information

• R -> Build R from its source on Windows
• http://www.stats.uwo.ca/faculty/murdoch/software/debuggingR/ (defunct)
• http://www.stats.uwo.ca/faculty/murdoch/software/debuggingR/gdb.shtml (defunct)
• Build R with debug information (see the discussion here). Cf output messages from running ./configure and make using the default options.

$ ./configure --help
$ ./configure --enable-R-shlib --with-valgrind-instrumentation=2 \
                               --with-system-valgrind-headers \
               CFLAGS='-g -O0 -fPIC' \
               FFLAGS='-g -O0 -fPIC' \
               CXXFLAGS='-g -O0 -fPIC' \
               FCFLAGS='-g -O0 -fPIC' 
$ make -j4
$ sudo make install

• My note of debugging cor() function
• Using gdb to debug R packages with native code (Video) The steps to debug is given below.

# Make sure to create a file <src/Makevars> with something like: CFLAGS=-ggdb -O0
# Or more generally
# CFLAGS=-Wall -Wextra -pedantic -O0 -ggdb
# CXXFLAGS=-Wall -Wextra -pedantic -O0 -ggdb
# FFLAGS=-Wall -Wextra -pedantic -O0 -ggdb

$ tree nidemo
$ R CMD INSTALL nidemo
$ cat bug.R
$ R -f bug.R 
$ R -d gdb
(gdb) r
> library(nidemo)
> Ctrl+C
(gdb) b nid_buggy_freq
(gdb) c  # continue
> buggy_freq("nidemo/DESCRIPTION") # stop at breakpoint 1
(gdb) list
(gdb) n # step through
(gdb) # press RETURN a few times until you see the bug
(gdb) d 1 # delete the first break point
(gdb) b Rf_error # R's C entry point for the error function
(gdb) c
> buggy_freq("nidemo/DESCRIPTION")
(gdb) bt 5 # last 5 stack frames
(gdb) frame 2
(gdb) list
(gdb) p freq_data
(gdb) p ans
(gdb) call Rf_PrintValues(ans)
(gdb) call Rf_PrintValues(fname)
(gdb) q
# Edit buggy.c

$ R CMD INSTALL nidemo # re-install the package
$ R -f bug.R
$ R -d gdb
(gdb) run
> source("bug.R") # error happened
(gdb) bt 5 # show the last 5 frames
(gdb) frame 2
(gdb) list
(gdb) frame 1
(gdb) list
(gdb) p file
(gdb) p fh
(gdb) q
# Edit buggy.c

$ R CMD INSTALL nidemo
$ R -f bug.R

• Compiled code from "R packages" by Hadley Wickham
• Debugging C/C++ code from Bioconductor (case study)
• Same idea for the Rcpp situation. See What are productive ways to debug Rcpp compiled code loaded in R (on OS X Mavericks)?

.Call

• Writing R Extensions manual.
• R’s C interface from Advanced R by Hadley Wickham

Registering native routines

https://cran.rstudio.com/doc/manuals/r-release/R-exts.html#Registering-native-routines

Pay attention to the prefix argument .fixes (eg .fixes = "C_") in useDynLib() function in the NAMESPACE file.

Example of debugging cor() function

Note that R's cor() function called a C function cor().

stats::cor
....
 .Call(C_cor, x, y, na.method, method == "kendall")

A step-by-step screenshot of debugging using the GNU debugger gdb can be found on my Github repository https://github.com/arraytools/r-debug.