Evaluate and Process R Code

This function takes either a vector/list of strings with actual R code, which it to be parsed to separate elements. Each list element is evaluated in a special environment, and a detailed list of results is returned for each logical part of the R code: a character value with R code, resulting R object, printed output, class of resulting R object, possible informative/warning/error messages and anything written to stdout. If a graph is plotted in the given text, the returned object is a string specifying the path to the saved file. Please see Details below. If parse option set to FALSE, then the returned list's length equals to the length of the parsed input - as each string is evaluated as separate R code in the same environment. If a nested list of R code or a concatenated string (separated by \n or ;) is provided like list(c('runif(1)', 'runif(1)')) with parse=FALSE, then everything is evaled at one run so the length of returned list equals to one or the length of the provided nested list. See examples below.

evals(txt, parse = evalsOptions("parse"), cache = evalsOptions("cache"),
  cache.mode = evalsOptions("cache.mode"),
  cache.dir = evalsOptions("cache.dir"),
  cache.time = evalsOptions("cache.time"),
  cache.copy.images = evalsOptions("cache.copy.images"),
  showInvisible = FALSE, classes = evalsOptions("classes"),
  hooks = evalsOptions("hooks"), length = evalsOptions("length"),
  output = evalsOptions("output"), env = NULL,
  graph.unify = evalsOptions("graph.unify"),
  graph.name = evalsOptions("graph.name"),
  graph.dir = evalsOptions("graph.dir"),
  graph.output = evalsOptions("graph.output"),
  width = evalsOptions("width"), height = evalsOptions("height"),
  res = evalsOptions("res"), hi.res = evalsOptions("hi.res"),
  hi.res.width = evalsOptions("hi.res.width"), hi.res.height = 960 *
  (height/width), hi.res.res = res * (hi.res.width/width),
  graph.env = evalsOptions("graph.env"),
  graph.recordplot = evalsOptions("graph.recordplot"),
  graph.RDS = evalsOptions("graph.RDS"), log = evalsOptions("log"), ...)

a character vector containing R code. This could be a list/vector of lines of code or a simple string holding R code separated by ; or \n.


if TRUE the provided txt elements would be merged into one string and parsed to logical chunks. This is useful if you would want to get separate results of your code parts - not just the last returned value, but you are passing the whole script in one string. To manually lock lines to each other (e.g. calling a plot and on next line adding an abline or text to it), use a plus char (+) at the beginning of each line which should be evaluated with the previous one(s). If set to FALSE, evals would not try to parse R code, it would get evaluated in separate runs - as provided. Please see examples below.


caching the result of R calls if set to TRUE. Please note the caching would not work if parse set to FALSE or syntax error is to be found.


cached results could be stored in an environment in current R session or let it be permanent on disk.


path to a directory holding cache files if cache.mode set to disk. Default to .cache in current working directory.


number of seconds to limit caching based on proc.time. If set to 0, all R commands, if set to Inf, none is cached (despite the cache parameter).


copy images to new file names if an image is returned from the disk cache? If set to FALSE (default), the cached path would be returned.


return invisible results?


a vector or list of classes which should be returned. If set to NULL (by default) all R objects will be returned.


list of hooks to be run for given classes in the form of list(class = fn). If you would also specify some parameters of the function, a list should be provided in the form of list(fn, param1, param2=NULL) etc. So the hooks would become list(class1=list(fn, param1, param2=NULL), ...). See example below. A default hook can be specified too by setting the class to 'default'. This can be handy if you do not want to define separate methods/functions to each possible class, but automatically apply the default hook to all classes not mentioned in the list. You may also specify only one element in the list like: hooks=list('default' = pander_return). Please note, that nor error/warning messages, nor stdout is captured (so: updated) while running hooks!


any R object exceeding the specified length will not be returned. The default value (Inf) does not filter out any R objects.


a character vector of required returned values. This might be useful if you are only interested in the result, and do not want to save/see e.g. messages or printed output. See examples below.


environment where evaluation takes place. If not set (by default), a new temporary environment is created.


should evals try to unify the style of (base, lattice and ggplot2) plots? If set to TRUE, some panderOptions() would apply. By default this is disabled not to freak out useRs :)


set the file name of saved plots which is tempfile by default. A simple character string might be provided where %d would be replaced by the index of the generating txt source, %n with an incremented integer in graph.dir with similar file names and %t by some unique random characters. While running in Pandoc.brew other indices could be triggered like %i and %I.


path to a directory where to place generated images. If the directory does not exist, evals try to create that. Default set to plots in current working directory.


set the required file format of saved plots. Currently it could be any of grDevices': png, bmp, jpeg, jpg, tiff, svg or pdf.


width of generated plot in pixels for even vector formats


height of generated plot in pixels for even vector formats


nominal resolution in ppi. The height and width of vector images will be calculated based in this.


generate high resolution plots also? If set to TRUE, each R code parts resulting an image would be run twice.


width of generated high resolution plot in pixels for even vector formats


height of generated high resolution plot in pixels for even vector formats. This value can be left blank to be automatically calculated to match original plot aspect ratio.


nominal resolution of high resolution plot in ppi. The height and width of vector plots will be calculated based in this. This value can be left blank to be automatically calculated to fit original plot scales.


save the environments in which plots were generated to distinct files (based on graph.name) with env extension?


save the plot via recordPlot to distinct files (based on graph.name) with recodplot extension?


save the raw R object returned (usually with lattice or ggplot2) while generating the plots to distinct files (based on graph.name) with RDS extension?


an optionally passed logger name from futile.logger to record all info, trace, debug and error messages. Logging to the console can be done by specifying e.g. flog.namespace(), and log to a file by previously calling flog.appender and appender.file on the given logger name.


optional parameters passed to graphics device (e.g. bg, pointsize etc.)


As evals tries to grab the plots internally, pleas do not run commands that set graphic device or dev.off. E.g. running evals(c('png("/tmp/x.png")', 'plot(1:10)', 'dev.off()')) would fail. printing of lattice and ggplot2 objects is not needed, evals would deal with that automatically.

The generated image file(s) of the plots can be fine-tuned by some specific options, please check out graph.output, width, height, res, hi.res, hi.res.width, hi.res.height and hi.res.res parameters. Most of these options are better not to touch, see details of parameters below.

Returned result values: list with the following elements

  • src - character vector of specified R code.

  • result - result of evaluation. NULL if nothing is returned. If any R code returned an R object while evaluating then the last R object will be returned as a raw R object. If a graph is plotted in the given text, the returned object is a string (with class set to image) specifying the path to the saved image file. If graphic device was touched, then no other R objects will be returned.

  • output - character vector of printed version (capture.output) of result

  • type - class of generated output. 'NULL' if nothing is returned, 'error' if some error occurred.

  • msg - possible messages grabbed while evaluating specified R code with the following structure:

    • messages - character vector of possible diagnostic message(s)

    • warnings - character vector of possible warning message(s)

    • errors - character vector of possible error message(s)

  • stdout - character vector of possibly printed texts to standard output (console)

By default evals tries to cache results. This means that if evaluation of some R commands take too much time (specified in cache.time parameter), then evals would save the results in a file and return from there on next exact R code's evaluation. This caching algorithm tries to be smart as checks not only the passed R sources, but all variables inside that and saves the hash of those.

Technical details of the caching algorithm:

  • Each passed R chunk is parsed to single commands.

  • Each parsed command's part (let it be a function, variable, constant etc.) evaluated (as a name) separately to a list. This list describes the unique structure and the content of the passed R commands, and has some IMHO really great benefits (see examples below).

  • A hash if computed to each list element and cached too in pander's local environments. This is useful if you are using large data frames, just imagine: the caching algorithm would have to compute the hash for the same data frame each time it's touched! This way the hash is recomputed only if the R object with the given name is changed.

  • The list is serialized and an SHA-1 hash is computed for that - which is unique and there is no real risk of collision.

  • If evals can find the cached results in a file named to the computed hash, then it is returned on the spot.

  • Otherwise the call is evaluated and the results are optionally saved to cache (e.g. if cache is active, if the proc.time() of the evaluation is higher then it is defined in cache.time etc.).

This is a quite secure way of caching, but if you would encounter any issues, just set cache to FALSE or tweak other cache parameters. While setting cache.dir, please do think about what you are doing and move your graph.dir accordingly, as evals might result in returning an image file path which is not found any more on your file system!

Also, if you have generated a plot and rendered that to e.g. png before and later try to get e.g. pdf - it would fail with cache on. Similarly you cannot render a high resolution image of a cached image, but you have to (temporary) disable caching.

The default evals options could be set globally with evalsOptions, e.g. to switch off the cache just run evalsOptions('cache', FALSE).

Please check the examples carefully below to get a detailed overview of evals.


a list of parsed elements each containing: src (the command run), result (R object: NULL if nothing returned, path to image file if a plot was generated), printed output, type (class of returned object if any), informative/wawrning and error messages (if any returned by the command run, otherwise set to NULL) and possible stdoutt value. See Details above.

See Also

eval.msgs evalsOptions

  • evals
# parsing several lines of R code
txt <- readLines(textConnection('x <- rnorm(100)
  warning('Lorem ipsum foo-bar-foo!')
  qplot(rating, data = movies, geom = 'histogram')
  y <- round(runif(100))
  cor.test(x, y)
  crl <- cor.test(runif(10), runif(10))
  table(mtcars$am, mtcars$cyl)
  ggplot(mtcars) + geom_point(aes(x = hp, y = mpg))'))

## parsing a list of commands
txt <- list('df <- mtcars',
 c('plot(mtcars$hp, pch = 19)','text(mtcars$hp, label = rownames(mtcars), pos = 4)'),
 'ggplot(mtcars) + geom_point(aes(x = hp, y = mpg))')

## the same commands in one string but also evaluating the `plot` with `text`
## (note the leading '+' on the beginning of `text...` line)
txt <- 'df <- mtcars
 plot(mtcars$hp, pch = 19)
 +text(mtcars$hp, label = rownames(mtcars), pos = 4)
 ggplot(mtcars) + geom_point(aes(x = hp, y = mpg))'
## but it would fail without parsing
evals(txt, parse = FALSE)

## handling messages
evals('message(20);message(20)', parse = FALSE)

## adding a caption to a plot
evals('set.caption("FOO"); plot(1:10)')
## `plot` is started with a `+` to eval the codes in the same chunk
## (no extra chunk with NULL result)
evals('set.caption("FOO"); +plot(1:10)')

## handling warnings
evals('chisq.test(mtcars$gear, mtcars$hp)')
evals(list(c('chisq.test(mtcars$gear, mtcars$am)', 'pi',
  'chisq.test(mtcars$gear, mtcars$hp)')), parse = FALSE)
evals(c('chisq.test(mtcars$gear, mtcars$am)',
  'chisq.test(mtcars$gear, mtcars$hp)'))

## handling errors
evals('Old MacDonald had a farm\\dots')
evals('## Some comment')
evals(c('runiff(20)', 'Old MacDonald had a farm?'))
evals(list(c('runiff(20)', 'Old MacDonald had a farm?')), parse = FALSE)
evals(c('mean(1:10)', 'no.R.function()'))
evals(list(c('mean(1:10)', 'no.R.function()')), parse = FALSE)
evals(c('no.R.object', 'no.R.function()', 'very.mixed.up(stuff)'))
evals(list(c('no.R.object', 'no.R.function()', 'very.mixed.up(stuff)')), parse = FALSE)
evals(c('no.R.object', 'Old MacDonald had a farm\\dots', 'pi'))
evals('no.R.object;Old MacDonald had a farm\\dots;pi', parse = FALSE)
evals(list(c('no.R.object', 'Old MacDonald had a farm\\dots', 'pi')), parse = FALSE)

## graph options
evals('plot(1:10)', graph.output = 'jpg')
evals('plot(1:10)', height = 800)
evals('plot(1:10)', height = 800, hi.res = TRUE)
evals('plot(1:10)', graph.output = 'pdf', hi.res = TRUE)
evals('plot(1:10)', res = 30)
evals('plot(1:10)', graph.name = 'myplot')
evals(list('plot(1:10)', 'plot(2:20)'), graph.name = 'myplots-%d')
evals('plot(1:10)', graph.env = TRUE)
evals('x <- runif(100);plot(x)', graph.env = TRUE)
evals(c('plot(1:10)', 'plot(2:20)'), graph.env = TRUE)
evals(c('x <- runif(100)', 'plot(x)','y <- runif(100)', 'plot(y)'), graph.env = TRUE)
    c('x <- runif(100)', 'plot(x)'),
    c('y <- runif(100)', 'plot(y)')),
  graph.env = TRUE, parse = FALSE)
evals('plot(1:10)', graph.recordplot = TRUE)
## unprinted lattice plot
evals('histogram(mtcars$hp)', graph.recordplot = TRUE)

## caching
system.time(evals('plot(mtcars)'))                # running again to see the speed-up :)
system.time(evals('plot(mtcars)', cache = FALSE)) # cache disabled

## caching mechanism does check what's inside a variable:
x <- mtcars
x <- cbind(mtcars, mtcars)
x <- mtcars

## stress your CPU - only once!
evals('x <- sapply(rep(mtcars$hp, 1e3), mean)')   # run it again!

## play with cache
evals('histogram(rep(mtcars$hp, 1e5))')
## nor run the below call
## that would return the cached version of the above call :)
f <- histogram
g <- rep
A <- mtcars$hp
B <- 1e5
evals('f(g(A, B))')#'

## or switch off cache globally:
evalsOptions('cache', FALSE)
## and switch on later
evalsOptions('cache', TRUE)

## evaluate assignments inside call to evals
## changes to environments are cached properly and retreived
evalsOptions('cache.time', 0)
x <- 2
evals('x <- x^2')[[1]]$result
evals('x <- x^2; x + 1')[[2]]$result
evalsOptions('cache.time', 0.1)

## returning only a few classes
txt <- readLines(textConnection('rnorm(100)
  list(x = 10:1, y = 'Godzilla!')
evals(txt, classes = 'numeric')
evals(txt, classes = c('numeric', 'list'))

## hooks
txt <- 'runif(1:4); matrix(runif(25), 5, 5); 1:5'
hooks <- list('numeric' = round, 'matrix' = pander_return)
evals(txt, hooks = hooks)
## using pander's default hook
evals(txt, hooks = list('default' = pander_return))
evals('22/7', hooks = list('numeric' = round))
evals('matrix(runif(25), 5, 5)', hooks = list('matrix' = round))

## setting default hook
evals(c('runif(10)', 'matrix(runif(9), 3, 3)'),
  hooks = list('default'=round))
## round all values except for matrices
evals(c('runif(10)', 'matrix(runif(9), 3, 3)'),
  hooks = list(matrix = 'print', 'default' = round))

# advanced hooks
hooks <- list('numeric' = list(round, 2), 'matrix' = list(round, 1))
evals(txt, hooks = hooks)

# return only returned values
evals(txt, output = 'result')

# return only messages (for checking syntax errors etc.)
evals(txt, output = 'msg')

# check the length of returned values and do not return looong R objects
evals('runif(10)', length = 5)

# note the following will not be filtered!
evals('matrix(1,1,1)', length = 1)

# if you do not want to let such things be eval-ed in the middle of a string
# use it with other filters :)
evals('matrix(1,1,1)', length = 1, classes = 'numeric')

# hooks & filtering
  hooks = list('matrix' = pander_return),
  output = 'result')

# eval-ing chunks in given environment
myenv <- new.env()
evals('x <- c(0,10)', env = myenv)
evals('mean(x)', env = myenv)
# note: if you had not specified 'myenv', the second 'evals' would have failed
evals('x <- c(0,10)')

# log
x <- evals('1:10', log = 'foo')
# trace log
evalsOptions('cache.time', 0)
x <- evals('1:10', log = 'foo')
x <- evals('1:10', log = 'foo')
# log to file
t <- tempfile()
flog.appender(appender.file(t), name = 'evals')
x <- evals('1:10', log = 'evals')
# permanent log for all events
evalsOptions('log', 'evals')
flog.threshold(TRACE, 'evals')
# }
Documentation reproduced from package pander, version 0.6.0, License: AGPL-3 | file LICENSE

Community examples

Looks like there are no examples yet.