Chunk options and package options

The knitr package shares most options with Sweave, but some were dropped/changed and some new options were added. The default values are in the parentheses below. Note that the chunk label for each chunk is assumed to be unique, i.e., no two chunks share the same label. This is especially important for cache and plot filenames. Chunks without labels will be assigned labels like unnamed-chunk-i where i is the chunk number.

Chunk Options

Take Rnw files as an example: usually we write chunk options in the form tag=value like this:

<<mychunk, cache=TRUE, eval=FALSE, dpi=100>>=

And opts_chunk$set() can change the default global options in a document (e.g. put this in a code chunk: opts_chunk$set(comment=NA, fig.width=6, fig.height=6)), and \SweaveOpts{} will no longer be supported (it is good if you do not know what this means). A few special notes on the options:

  1. Chunk options must be written in one line; no line breaks are allowed inside chunk options;
  2. Avoid spaces ` ` and periods . in chunk labels and directory names; if your output is a TeX document, these characters can cause troubles (in general it is recommended to use alphabetic characters with words separated by - or _ and avoid other characters), e.g. setup-options is a good label, whereas setup.options and chunk 1 are bad; fig.path='figures/mcmc-' is a good prefix for figure output if this project is about MCMC, and fig.path='markov chain/monte carlo' is bad; non-alphanumeric characters except - and _ in figure filenames will be replaced with _ automatically;
  3. All option values must be valid R expressions just like how we write function arguments;
    • for example, options that take character values must be quoted as you do in R (e.g. should write fig.path="abc" instead of fig.path=abc, and out.width='\\textwidth' instead of out.width=\textwidth)
    • in theory, the chunk label should be quoted as well but the sake of convenience it will be automatically quoted if you did not quote it (e.g. <<2a>>= will become <<'2a'>>=)
    • for logical options, TRUE and FALSE are OK, and true/false will not work as you might have expected because true is not TRUE
    • you can write arbitrarily complicated expressions as you want as long as they are legitimate R code
    • Sweave users please read the transition page carefully because the syntax is different

Below is a list of chunk options in knitr. Note the options background and size only apply to .Rnw documents.

Code Evaluation

  • eval: (TRUE; logical) whether to evaluate the code chunk; it can also be a numeric vector to select which R expression(s) to evaluate, e.g. eval=c(1, 3, 4) or eval=-(4:5)

Text Results

  • echo: (TRUE; logical or numeric) whether to include R source code in the output file; besides TRUE/FALSE which completely turns on/off the source code, we can also use a numeric vector to select which R expression(s) to echo in a chunk, e.g. echo=2:3 means only echo the 2nd and 3rd expressions, and echo=-4 means to exclude the 4th expression
  • results: ('markup'; character) takes these possible values
    • markup: mark up the results using the output hook, e.g. put results in a special LaTeX environment
    • asis: output as-is, i.e., write raw results from R into the output document
    • hold: hold all the output pieces and push them to the end of a chunk
    • hide hide results; this option only applies to normal R output (not warnings, messages or errors)
    • note markup and asis are equivalent to verbatim and tex in Sweave respectively (you can still use the latter two, but they can be misleading, e.g., verbatim does not really mean verbatim in R, and tex seems to be restricted to LaTeX)
  • collapse: (FALSE; logical; applies to Markdown output only) whether to, if possible, collapse all the source and output blocks from one code chunk into a single block (by default, they are written to separate <pre></pre> blocks)
  • warning: (TRUE; logical) whether to preserve warnings (produced by warning()) in the output like we run R code in a terminal (if FALSE, all warnings will be printed in the console instead of the output document); it can also take numeric values as indices to select a subset of warnings to include in the output
  • error: (TRUE; logical) whether to preserve errors (from stop()); by default, the evaluation will not stop even in case of errors!! if we want R to stop on errors, we need to set this option to FALSE
    • when the chunk option include = FALSE, error knitr will stop on error, because it is easy to overlook potential errors in this case
  • message: (TRUE; logical) whether to preserve messages emitted by message() (similar to warning)
  • split: (FALSE; logical) whether to split the output from R into separate files and include them into LaTeX by \input{} or HTML by <iframe></iframe>; this option only works for .Rnw, .Rtex, and .Rhtml documents (it does not work for R Markdown)
  • include: (TRUE; logical) whether to include the chunk output in the final output document; if include=FALSE, nothing will be written into the output document, but the code is still evaluated and plot files are generated if there are any plots in the chunk, so you can manually insert figures; note this is the only chunk option that is not cached, i.e., changing it will not invalidate the cache
  • strip.white: (TRUE; logical) whether to remove the white lines in the beginning or end of a source chunk in the output
  • class.output: (NULL; character) useful for HTML output, appends classes that can be used in conjunction with css, so you can apply custom formatting.

Code Decoration

  • tidy: (FALSE; logical) whether R code should be tidied up using the function tidy_source() in the formatR package; if it failed to tidy up, original R code will not be changed; tidy=TRUE is like keep.source=FALSE in Sweave, but it also tries not to discard R comments (N.B. this option does not work in certain cases; see http://yihui.name/formatR for more information)
    • tidy.opts: (NULL; list) a list of options to be passed to tidy.source(), e.g. tidy.opts=list(blank=FALSE, width.cutoff=60); these formatR options can also be set globally via options()
  • prompt: (FALSE; logical) whether to add the prompt characters in the R code (see prompt and continue in ?base::options; note that adding prompts can make it difficult for readers to copy R code from the output, so prompt=FALSE may be a better choice
    • this option may not work well when the chunk option engine is not R (#1274)
  • comment: ('##'; character) the prefix to be put before source code output; default is to comment out the output by ##, which is good for readers to copy R source code since output is masked in comments (set comment=NA to disable this feature)
  • highlight: (TRUE; logical) whether to highlight the source code (it is FALSE by default if the output is Sweave or listings)
  • size: ('normalsize'; character) font size for the default LaTeX output (see ?highlight in the highlight package for a list of possible values)
  • background: ('#F7F7F7'; character or numeric) background color of chunks in LaTeX output (passed to the LaTeX package framed); the color model is rgb; it can be either a numeric vector of length 3, with each element between 0 and 1 to denote red, green and blue, or any built-in color in R like red or springgreen3 (see colors() for a full list), or a hex string like #FFFF00, or an integer (all these colors will be converted to the RGB model; see ?col2rgb for details)
  • class.source: (NULL; character) useful for HTML output, appends classes that can be used in conjunction with css, so you can apply custom formatting.

There is a hidden option indent which stores the possible leading white spaces of the chunk, e.g. for the chunk below, indent is a character string of two spaces:


Currently this option is only used to indent markdown output, because leading white spaces have special meanings in markdown.


  • cache: (FALSE; logical) whether to cache a code chunk; when evaluating code chunks, the cached chunks are skipped, but the objects created in these chunks are (lazy-) loaded from previously saved databases (.rdb and .rdx) files, and these files are saved when a chunk is evaluated for the first time, or when cached files are not found (e.g. you may have removed them by hand); note the filename consists of the chunk label with an MD5 digest of the R code in the chunk (the MD5 string is a summary of the chunk text, and any changes in the chunk will produce a different MD5 digest); unlike the cacheSweave package which uses stashR, this package directly uses internal functions in base R for cache, and another difference is that results of the code will still be included in the output even when cache is used (whereas cacheSweave has no output when a chunk is cached), because knitr also caches the printed output of a code chunk as a character string
    • for advanced users, cache can be more granular and takes numeric values 0, 1, 2, 3; see the cache example page for “More granular cache”
  • cache.path: ('cache/'; character) a prefix to be used for the names of cache files (by default they are saved to a directory named cache relative to the current working directory; you can also use an absolute dir here, e.g. /home/foo/bar- or D:\\abc\\mycache, but it is not recommended since such absolute directories may not exist in other people’s systems, therefore it is recommended to use relative directories)
  • cache.vars: (NULL) a character vector of variable names to be saved in the cache database; by default all variables created in the current chunks are identified and saved, but we can manually set the variables to be saved
  • cache.lazy: (TRUE) whether to lazyLoad() or directly load() objects; for very large objects, lazyloading may not work, so cache.lazy=FALSE may be desirable (see #572)
  • cache.comments: (NULL) if FALSE, changing comments in R code chunks will not invalidate the cache database
  • cache.rebuild: (FALSE) if TRUE, the chunk is evaluated even if cache=TRUE and no changes are made. This is useful when given an R code argument. For example: cache.rebuild=!file.exists("path/to/file.ext") (see #238).
  • dependson: (NULL; character or numeric) a character vector of chunk labels to specify which other chunks this chunk depends on; this option applies to cached chunks only – sometimes the objects in a cached chunk may depend on other cached chunks, so when other chunks are changed, this chunk must be updated accordingly
    • if dependson is a numeric vector, it means the indices of chunk labels, e.g. dependson=1 means this chunk depends on the first chunk in the document, and dependson=c(-1, -2) means it depends on the previous two chunks (negative indices stand for numbers of chunks before this chunk, and note they are always relative to the current chunk)
    • please note this option does not work when set as a global chunk option via opts_chunk$set(); it must be set as a local chunk option
  • autodep: (FALSE; logical) whether to figure out the dependencies among chunks automatically by analyzing the global variables in the code (may not be reliable) so that dependson does not need to be set explicitly


  • fig.path: ('figure/'; character) prefix to be used for figure filenames (fig.path and chunk labels are concatenated to make filenames); it may contain a directory like figure/prefix- (will be created if it does not exist); this path is relative to the current working directory; if the prefix ends in a trailing slash, e.g. output/figures/, figures will be saved in the specified directory without any changes to filename prefix, thus providing a relative filepath alternative to the package-level option base.dir
  • fig.keep: ('high'; character) how plots in chunks should be kept; it takes five possible character values or a numeric vector (see the end of this section for an example)
    • high: only keep high-level plots (merge low-level changes into high-level plots);
    • none: discard all plots;
    • all: keep all plots (low-level plot changes may produce new plots)
    • first: only keep the first plot
    • last: only keep the last plot
    • if set to a numeric vector: interpret value as index of (low-level) plots to keep
  • fig.show: ('asis'; character) how to show/arrange the plots; four possible values are
    • asis: show plots exactly in places where they were generated (as if the code were run in an R terminal);
    • hold: hold all plots and output them in the very end of a code chunk;
    • animate: wrap all plots into an animation if there are mutiple plots in a chunk;
    • hide: generate plot files but hide them in the output document
  • dev: ('pdf' for LaTeX output and 'png' for HTML/markdown; character) the function name which will be used as a graphical device to record plots; for the convenience of usage, this package has included all the graphics devices in base R as well as those in Cairo, cairoDevice and tikzDevice, e.g. if we set dev='CairoPDF', the function with the same name in the Cairo package will be used for graphics output; if none of the 20 built-in devices is appropriate, we can still provide yet another name as long as it is a legal function name which can record plots (it must be of the form function(filename, width, height)); note the units for images are always inches (even for bitmap devices, in which DPI is used to convert between pixels and inches); currently available devices are bmp, postscript, pdf, png, svg, jpeg, pictex, tiff, win.metafile, cairo_pdf, cairo_ps, CairoJPEG, CairoPNG, CairoPS, CairoPDF, CairoSVG, CairoTIFF, Cairo_pdf, Cairo_png, Cairo_ps, Cairo_svg, tikz and a series of quartz devices including quartz_pdf, quartz_png, quartz_jpeg, quartz_tiff, quartz_gif, quartz_psd, quartz_bmp which are just wrappers to the function quartz() with different file types
    • the options dev, fig.ext, fig.width, fig.height and dpi can be vectors (shorter ones will be recycled), e.g. <<foo, dev=c('pdf', 'png')>>= creates two files for the same plot: foo.pdf and foo.png
  • dev.args: (NULL) more arguments to be passed to the device, e.g. dev.args=list(bg='yellow', pointsize=10); note this depends on the specific device (see the device documentation); when dev has multiple elements, dev.args can be a list of lists of arguments with each list of arguments to be passed to each single device, e.g. <<dev=c('pdf', 'tiff'), dev.args=list(pdf = list(colormodel = 'cmyk', useDingats = TRUE), tiff = list(compression = 'lzw'))>>=
  • fig.ext: (NULL; character) file extension of the figure output (if NULL, it will be derived from the graphical device; see knitr:::auto_exts for details)
  • dpi: (72; numeric) the DPI (dots per inch) for bitmap devices (dpi * inches = pixels)
  • fig.width, fig.height: (both are 7; numeric) width and height of the plot, to be used in the graphics device (in inches) and have to be numeric
  • fig.asp: (NULL; numeric) the aspect ratio of the plot, i.e. the ratio of height/width; when fig.asp is specified, the height of a plot (the chunk option fig.height) is calculated from fig.width * fig.asp
  • out.width, out.height: (NULL; character) width and height of the plot in the final output file (can be different with its real fig.width and fig.height, i.e. plots can be scaled in the output document); depending on the output format, these two options can take flexible values, e.g. for LaTeX output, they can be .8\\linewidth, 3in or 8cm and for HTML, they may be 300px (do not have to be in inches like fig.width and fig.height; backslashes must be escaped as \\); for LaTeX output, the default value for out.width will be changed to \\maxwidth which is defined here
    • out.width can also be a percentage, e.g. '40%' will be translated to 0.4\linewidth when the output format is LaTeX
  • out.extra: (NULL; character) extra options for figures, e.g. out.extra='angle=90' in LaTeX output will rotate the figure by 90 degrees; it can be an arbitrary string, e.g. you can write multiple figure options in this option; it also applies to HTML images (extra options will be written into the <img /> tag, e.g. out.extra='style="display:block;"')
  • fig.retina: (1; numeric) this option only applies to HTML output; for Retina displays, setting this option to a ratio (usually 2) will change the chunk option dpi to dpi * fig.retina, and out.width to fig.width * dpi / fig.retina internally; for example, the physical size of an image is doubled and its display size is halved when fig.retina = 2
  • resize.width, resize.height: (NULL; character) the width and height to be used in \resizebox{}{} in LaTeX; these two options are not needed unless you want to resize tikz graphics because there is no natural way to do it; however, according to tikzDevice authors, tikz graphics is not meant to be resized to maintain consistency in style with other texts in LaTeX; if only one of them is NULL, ! will be used (read the documentation of graphicx if you do not understand this)
  • fig.align: ('default'; character) alignment of figures in the output document (possible values are left, right and center; default is not to make any alignment adjustments); note that for Markdown output, forcing figure alignments will lead to the HTML tag <img /> instead of the original Markdown syntax ![](), because Markdown does not have native support for figure alignments (see #611)
  • fig.env: ('figure') the LaTeX environment for figures, e.g. set fig.env='marginfigure' to get \begin{marginfigure}
  • fig.cap: (NULL; character) figure caption to be used in a figure environment in LaTeX (in \caption{}); if NULL or NA, it will be ignored, otherwise a figure environment will be used for the plots in the chunk (output in \begin{figure} and \end{figure})
  • fig.scap: (NULL; character) short caption; if NULL, all the words before . or ; or : will be used as a short caption; if NA, it will be ignored
  • fig.lp: ('fig:'; character) label prefix for the figure label to be used in \label{}; the actual label is made by concatenating this prefix and the chunk label, e.g. the figure label for <<foo-plot>>= will be fig:foo-plot by default
  • fig.pos: (''; character) a character string for the figure position arrangement to be used in \begin{figure}[fig.pos]
  • fig.subcap: (NULL) captions for subfigures; when there are multiple plots in a chunk, and neither fig.subcap nor fig.cap is NULL, \subfloat{} will be used for individual plots (you need to add \usepackage{subfig} in the preamble); see 067-graphics-options.Rnw for an example
  • fig.process: (NULL) a function to post-process a figure file; it should take a filename, and return a character string as the new source of the figure to be inserted in the output
  • fig.showtext: (NULL) if TRUE, call showtext::showtext.begin() before drawing plots; see the documentation of the showtext package for details
  • external: (TRUE; logical) whether to externalize tikz graphics (pre-compile tikz graphics to PDF); it is only used for the tikz() device in the tikzDevice package (i.e., when dev='tikz') and it can save time for LaTeX compilation
  • sanitize: (FALSE; character) whether to sanitize tikz graphics (escape special LaTeX characters); see documentation in the tikzDevice package

Note any number of plots can be recorded in a single code chunk, and this package does not need to know how many plots are in a chunk in advance – it can figure out automatically, and name these images as fig.path-label-i where i is incremental from 1; if a code chunk does not actually produce any plots, knitr will not record anything either (the graphics device is open only when plots are really produced); in other words, it does not matter if fig.keep='high' but no plots were produced.

Low-level plotting commands include lines() and points(), etc. To better understand fig.keep, consider the following chunk:

plot(1)         # high-level plot
abline(0, 1)    # low-level change
plot(rnorm(10)) # high-level plot
## many low-level changes in a loop (a single R expression)
for(i in 1:10) {
    abline(v = i, lty = 2)

Normally this produces 2 plots in the output (i.e. when fig.keep='high'); for fig.keep='none', no plots will be saved; for fig.keep='all', 4 plots are saved; for fig.keep='first', the plot produced by plot(1) is saved, and for fig.keep='last', the last plot with 10 vertical lines is saved.

There are two hidden options which are not designed to be set by the users: fig.cur (the current figure number or index when there are multiple figures) and fig.num (the total number of figures in a chunk). The purpose of these two options is to help knitr deal with the filenames of multiple figures as well as animations. In some cases, we can make use of them to write animations into the output using plot files which are saved manually (see the graphics manual for examples).


  • interval: (1; numeric) number of seconds to pause between animation frames
  • aniopts: ('controls,loop') extra options for animations; see the documentation of the animate package
  • ffmpeg.bitrate (1M; character) to be passed to the -b:v argument of FFmpeg to control the quality of WebM videos created from FFmpeg
  • ffmpeg.format (webm; character) the video format of FFmpeg, i.e. the filename extension of the video

Code Chunk

  • code: (NULL; character) if provided, it will override the code in the current chunk; this allows us to programmatically insert code into the current chunk; e.g. a chunk option code = capture.output(dump('fivenum', '')) will use the source code of the function fivenum to replace the current chunk
  • ref.label: (NULL; character) a character vector of labels of the chunks from which the code of the current chunk is inherited (see the demo for chunk reference)

Child Documents

  • child: (NULL; character) a character vector of filenames of child documents to be run and input into the main document

Language Engines

  • engine: ('R'; character) the language name of the code chunk; currently other possible values are 'python' and 'awk'/'gawk'; the object knit_engines in this package can be used to set up engines for other languages

Option templates

  • opts.label: (NULL; character) the label of options set in opts_template (see ?opts_template); this option can save some typing efforts for sets of frequently used chunk options

Extracting source code

  • purl: (TRUE; logical) when running purl() to extract source code from a source document, whether to include or exclude a certain code chunk

Other Chunk Options

  • R.options: (NULL) local R options for a code chunk; these options are set temporarily via options() before the code chunk, and restored after the chunk

Package Options

The package options can be changed using the object opts_knit; for example,

opts_knit$set(progress = TRUE, verbose = TRUE)

See ?opts_knit for the alternative approach to setting package options using the R base function options().

All package options are:

  • animation.fun: (hook_ffmpeg_html) a hook function to create animations in HTML output; the default hook uses FFmpeg to convert images to an MP4 video
  • aliases: (NULL) a named character vector to specify the aliases of chunk options, e.g. c(h = 'fig.height', w = 'fig.width') tells knitr that the chunk option h really means fig.height, and w is an alias for fig.width; this option can be used to save some typing efforts for long option names
  • base.dir: (NULL) an absolute directory under which the plots are generated
  • base.url: (NULL) the base url for HTML pages
  • child.path: ('') the search path for child documents; by default child documents are searched for relative to the directory of the parent document
  • concordance: (FALSE) whether to write a concordance file to map the output line numbers to the input line numbers; this enables one to navigate from the output to the input and can be helpful especially when TeX error occurs (this feature is mainly for RStudio)
  • eval.after: (fig.cap) a character vector of option names; these options will be evaluated after a chunk is evaluated, and all other options will be evaluated before a chunk (e.g. for chunk option fig.cap=paste('p-value is', t.test(x)$p.value), it will be evaluated after the chunk according to the value of x if eval.after='fig.cap')
  • global.par: (FALSE) if TRUE, the par() settings from the last code chunk will be preserved and applied to the next code chunk (of course, this only applies to base graphics); by default, knitr opens a new graphics device to record plots and close it after evaluating the code, so par() settings will be discarded
  • header: the text to be inserted into the output document before the document begins (e.g. after \documentclass{article} in LaTeX, or <head> in HTML); this is useful for defining commands and styles in the LaTeX preamble or HTML header; the beginning of document is found using the pattern defined in knit_patterns$get('document.begin')
  • latex.options.color, latex.options.graphicx (NULL): options for the LaTeX packages color and graphicx, respectively
  • out.format: (NULL) possible values are latex, sweave, html, markdown and jekyll; it will be automatically determined based on the input file, and this option will affect which set of hooks to use (see ?render_latex for example); note this option has to be set before knit() runs (i.e. it does not work if you set it in the document), or alternatively, you can use the render_* series inside the document to set up the hooks
  • progress: (TRUE) whether to display a progress bar when running knitr; note it also depends on the R option KNITR_PROGRESS (it this variable is set to FALSE via options(KNITR_PROGRESS = FALSE), the progress option will be set to FALSE when knitr is loaded)
  • root.dir: (NULL) the root directory when evaluating code chunks; if NULL, the directory of the input document will be used
  • self.contained: (TRUE) whether the output document should be self-contained (TeX styles written in the tex document, and CSS styles in HTML document); this option only applies to Rnw and Rhtml documents, and does not apply to R Markdown documents (Rmd)
  • unnamed.chunk.label: (unnamed-chunk) the label prefix for unnamed chunks
  • upload.fun: (identity) a function that takes a filename as its input, processes it and returns a character string when the output format is HTML or Markdown; typically it is a function to upload a image and return the link to the image, e.g. opts_knit$set(upload.fun = imgur_upload) can upload a file to http://imgur.com (see ?imgur_upload)
  • verbose: (FALSE) whether to show verbose information (e.g., R code in each chunk and message logs) or just show chunk labels and options
  • width: (75) it is used to set the R session option width that may affect the width of source code and text output