Skip to content

Latest commit

 

History

History
120 lines (89 loc) · 2.82 KB

Debugging.md

File metadata and controls

120 lines (89 loc) · 2.82 KB
Raw

Debugging in R

  • print()/cat()/message()
  • browser()/recover()
  • debug()
  • trace()
  • traceback()
  • options() error = recover/browser/dump.frames

Proactive Debugging

Stop in a given function before it starts doing its work.

Reactive Debugging

When an error occurs, we want to stop and look around at the current state of the computations. This includes not only the line where the error occurred and the variables used in that expression, but also the state of other function calls that led to this error. While the error clearly occurs in function y, the problem may have been caused by function x calling y with a bad/unexpected argument which was computed in x erroneously.

We arrange to stop when there is an error

options(error = recover)

An alternative is to call traceback() immediately after an error occurred (i.e. before we issue another R command at the REPL).

Turning warnings into errors

Warnings are harder to catch. Warnings are typically displayed at the end of a computation, not when they occur. We can change this with

options(warn = 1)

To trap warnings as they occur and stop to examine the call stack, we might debug the warning() function and see where it is called. Alternatively, we can elevate all warnings to errors with

options(warn = 2)

Post-mortem Debugging

After an error occurs and the next prompt is shown, you can use traceback() to explore the call stack. This behaves much like debugging with options(error = recover). It merely changes the default where now you chose to examine the call stack, whereas with error = recover, you are always asked to explore it and have to exit.

Batch Debugging

We can run R in "batch" mode, i.e., you run R not interactively but as a background/asynchronous job/process. This is useful for very long running tasks what happens when there is an error.

tmp = mtcars[-31,]
by(tmp, list(tmp$cyl, tmp$am), function(x) lm(mpg ~ wt, x)))
by(tmp, list(tmp$cyl, tmp$am), lm, formula = mpg ~ wt)
debug(lm)
by(tmp, list(tmp$cyl, tmp$am), lm, formula = mpg ~ wt)
trace(lm, quote(print(data)), print = FALSE)
by(tmp, list(tmp$cyl, tmp$am), lm, formula = mpg ~ wt)

The following gives an error:

by(tmp, list(tmp$cyl, tmp$am), lm, formula = mpg ~ et)

library(ReadPDF)
getDocTitle("foo")
library(ReadPDF)
f = list.files("~/DSIProjects/Zoonotics-shared/LatestDocs/PDF/0000708828/", pattern = "xml$", full = TRUE)
getDocTitleString(f)
source("~/DSIProjects/ReadPDF/R/getTitle.R")

Note that some of the functions getDocTitle() calls may be in the ReadPDF namespace and unexported. In this case, our new functions won't find them as we have source()d these into the global environment.

Let's debug isTitleBad() to see why it thinks this txt is not a title.

debug(isTitleBad)