pander: An R Pandoc Writer

The main aim of the pander R package is to provide a minimal and easy tool for rendering R objects into Pandoc's markdown. The package is also capable of exporting/converting complex Pandoc documents (reports) in various ways. Regarding the difference between pander and other packages for exporting R objects to different file formats, please refer to this section.

Current build and test coverage status: .

Some CRAN statistics:


The stable version can be installed easily in the R console like any other package:


On the other hand, I welcome everyone to use the most recent version of the package with quick-fixes, new features and probably new bugs. It's currently hosted on GitHub. To get the latest development version from GitHub of the devtools package from CRAN:


Or download the sources and build manually. If you're running R on Windows, you will need to install Rtools.


Few cool packages from CRAN are needed for installing and/or using pander:

And there are also a few optional suggested or supported R packages, such as:


pander heavily builds on Pandoc, which should be pre-installed before trying to convert your reports to different formats. Although main functions work without Pandoc, e.g. you can transform R objects into markdown or generate a markdown formatted report via Pandoc.brew or the custom reference class, but installing that great piece of software is suggested if you want to convert markdown to PDF/docx/HTML or other formats.

Starting v0.98.932 RStudio comes with a bundled Pandoc binary, so one can save the tedious steps of installing Pandoc.

If you do not have RStudio installed, please refer to the installation process of Pandoc, which is quite straightforward on most-popular operating systems: download and run the binary (a few megabytes), and get a full-blown document converter in a few seconds/minutes. On some Linux distributions, it might be a bit more complicated (as repositories tend to provide out-dated versions of Pandoc, so you would need cabal-install to install from sources). Please do not forget to restart your R session to update your PATH after installation!

Helper functions

The package contains numerous helper functions, which render user specified inputs in Pandoc's markdown format or apply some extra formatting on it. All Pandoc-related functions' names are starting with pandoc. For example pandoc.table is used for rendering tables in markdown. For a technical documentation, see the HTML help files of the package at Rdocumentation.

All pandoc functions generally prints to console and do not return anything by default. If you want the opposite, to get markdown in a string, call each function ending in .return, for example pandoc.table.return. For more details, please see the official documentation in e.g. ?pandoc.strong.

The full list of currently available primitive Pandoc-related functions are:

For example there is a helper function rendering R lists into markdown:

> l <- list(
+        "First list element",
+        paste0(1:5, '. subelement'),
+        "Second element",
+        list('F', 'B', 'I', c('phone', 'pad', 'talics')))
> pandoc.list(l, 'roman')

Which command produces the following output:

I. First list element 
    I. 1. subelement 
    II. 2. subelement 
    III. 3. subelement 
    IV. 4. subelement 
    V. 5. subelement  
II. Second element 
    I. F 
    II. B 
    III. I 
        I. phone 
        II. pad 
        III. talics   

<!-- end of list -->

Markdown tables

One of the most popular feature in pander is pandoc.table, rendering most tabular R objects into markdown tables with various options and settings (e.g. style, caption, cell highlighting, cell alignment, width). This section aims to provide quick introduction to most common options, but for more usage/implementation details and examples, please refer to specialized vignette, which can be accessed by vignette('pandoc_table') or available online here.

Let's start with a small example:

> pandoc.table(mtcars[1:3, 1:4])

Which command produces the following output by default:

      &nbsp;         mpg   cyl   disp   hp 
------------------- ----- ----- ------ ----
   **Mazda RX4**     21     6    160   110 

 **Mazda RX4 Wag**   21     6    160   110 

  **Datsun 710**    22.8    4    108    93 

Please note that all below features are also supported by the more concise pander generic S3 method!


All four Pandoc formats are supported by pander. From those (multiline, simple, grid, pipe/rmarkdown), I'd suggest sticking to the default multiline format with the most features, although rmarkdown v1.0 fanatics might better like the pipe table format. Please see a few examples below:

The default style is the multiline format as most features (e.g. multi-line cells and alignment) are supported:

> m <- mtcars[1:2, 1:3]
> pandoc.table(m)

      &nbsp;         mpg   cyl   disp 
------------------- ----- ----- ------
   **Mazda RX4**     21     6    160  

 **Mazda RX4 Wag**   21     6    160  

While simple tables are much more compact, but do not support line breaks in cells:

> pandoc.table(m, style = "simple")

      &nbsp;         mpg   cyl   disp 
------------------- ----- ----- ------
   **Mazda RX4**     21     6    160  
 **Mazda RX4 Wag**   21     6    160  

My personal favorite, the grid format is really handy for emacs users and it does support line breaks inside of cells, but cell alignment is not possible in most parsers:

> pandoc.table(m, style = "grid")

|       &nbsp;        |  mpg  |  cyl  |  disp  |
|    **Mazda RX4**    |  21   |   6   |  160   |
|  **Mazda RX4 Wag**  |  21   |   6   |  160   |

And the so called rmarkdown or pipe table format is often used directly with knitr, since it was supporters by the first versions of the markdown package:

> pandoc.table(m, style = "rmarkdown")

|       &nbsp;        |  mpg  |  cyl  |  disp  |
|    **Mazda RX4**    |  21   |   6   |  160   |
|  **Mazda RX4 Wag**  |  21   |   6   |  160   |

But once again, you should simply stick to the default multiline table format in most cases. Otherwise, it's wise to update the default table format via panderOptions.


It's really easy to add a caption to a table:

> pandoc.table(m, style = "grid", caption = "Hello caption!")

|       &nbsp;        |  mpg  |  cyl  |  disp  |
|    **Mazda RX4**    |  21   |   6   |  160   |
|  **Mazda RX4 Wag**  |  21   |   6   |  160   |

Table: Hello caption!

For more convenient and flexible usage, you might be interested in the special set.caption helper function. Call the function at any time, and the next table or plot will catch up the provided caption:

> set.caption("Hello caption!")
> pandoc.table(m)

      &nbsp;         mpg   cyl   disp 
------------------- ----- ----- ------
   **Mazda RX4**     21     6    160  

 **Mazda RX4 Wag**   21     6    160  

Table: Hello caption!

Unless permanent option is set for TRUE (by default), caption will be set only for next table. To disable permanently set caption, just call set.caption(NULL) or call set.caption with permanent parameter being set to FALSE.

Highlighting cells

One of the fanciest features in pander is the ease of highlighting rows, columns or any cells in a table. This is a real markdown feature without custom HTML or LaTeX-only tweaks, so all HTML/PDF/MS Word/OpenOffice etc. formats are supported.

This can be achieved by calling pandoc.table directly and passing any (or more) of the following arguments or calling the R function with the same names before rendering a table with either the pander generic S3 method or via pandoc.table:

The emphasize.italics helpers would turn the affected cells to italic, emphasize.strong would apply a bold style to the cell and emphasize.verbatim would apply a verbatim style to the cell. A cell can be also italic, bold and verbatim at the same time.

Those functions and arguments ending in rows or cols take a vector (like which columns or rows to emphasize in a table), while the cells argument take either a vector (for one dimensional "tables") or an array-like data structure with two columns holding row and column indexes of cells to be emphasized -- just like what which(..., arr.ind = TRUE) returns. A quick-example:

> t <- mtcars[1:3, 1:5]
> emphasize.italics.cols(1)
> emphasize.italics.rows(1)
> emphasize.strong.cells(which(t > 20, arr.ind = TRUE))
> pandoc.table(t)

      &nbsp;           mpg      cyl    disp       hp      drat 
------------------- ---------- ----- --------- --------- ------
   **Mazda RX4**     ***21***   *6*  ***160*** ***110*** *3.9* 

 **Mazda RX4 Wag**   ***21***    6    **160**   **110**   3.9  

  **Datsun 710**    ***22.8***   4    **108**   **93**    3.85 

For more examples, please see our "Highlight cells in markdown tables" blog post.

Cell alignment

You can specify the alignment of the cells (left, right or center/centre) in a table directly by setting the justify parameter:

> pandoc.table(head(iris[,1:3], 2), justify = c('right', 'center', 'left'))

  Sepal.Length  Sepal.Width  Petal.Length  
-------------- ------------- --------------
           5.1      3.5      1.4           

           4.9       3       1.4           

Or pre-define the alignment for (all future) pandoc.table or the pander S3 generic method by a helper function:

> set.alignment('left', row.names = 'right')
> pandoc.table(mtcars[1:2,  1:5])

             &nbsp; mpg   cyl   disp   hp   drat  
------------------- ----- ----- ------ ---- ------
      **Mazda RX4** 21    6     160    110  3.9   

  **Mazda RX4 Wag** 21    6     160    110  3.9   

Just like with captions, you can also specify the permanent option to be TRUE to update the default cell alignment for all future tables. And beside using set.alignment helper function or passing parameters directly to pandoc.table, you may also set the default alignment styles with panderOptions.

What's even more fun, you can specify a function that takes the R object as its argument to compute some unique alignment for your table based on e.g. column values or variable types:

> panderOptions('table.alignment.default',
+   function(df)
+     ifelse(sapply(df, mean) > 2, 'left', 'right'))
> pandoc.table(head(iris[,1:3], 2))

Sepal.Length   Sepal.Width     Petal.Length
-------------- ------------- --------------
5.1            3.5                      1.4

4.9            3                        1.4

Table and cell width

pandoc.table can also deal with the problem of really wide tables. Ever had an issue in LaTeX or MS Word when tried to print a correlation matrix of 40 variables? Not a problem any more as you can split up the table with auto-added captions. The split.table option defaults to 80 characters:

> pandoc.table(mtcars[1:2, ], style = "grid", caption = "Hello caption!")

|       &nbsp;        |  mpg  |  cyl  |  disp  |  hp  |  drat  |  wt   |
|    **Mazda RX4**    |  21   |   6   |  160   | 110  |  3.9   | 2.62  |
|  **Mazda RX4 Wag**  |  21   |   6   |  160   | 110  |  3.9   | 2.875 |

Table: Hello caption! (continued below)

|       &nbsp;        |  qsec  |  vs  |  am  |  gear  |  carb  |
|    **Mazda RX4**    | 16.46  |  0   |  1   |   4    |   4    |
|  **Mazda RX4 Wag**  | 17.02  |  0   |  1   |   4    |   4    |

And too wide cells can also be split by line breaks. The maximum number of characters in a cell is specified by split.cells parameter (default to 30), can be a single value, vector (values for each column separately) and relative vector (percentages of split.tables parameter):

> df <- data.frame(a = 'Lorem ipsum', b = 'dolor sit', c = 'amet')
> pandoc.table(df, split.cells = 5)

  a     b    c  
----- ----- ----
Lorem dolor amet
ipsum  sit      

> pandoc.table(df, split.cells = c(5, 20, 5))

  a       b      c  
----- --------- ----
Lorem dolor sit amet

> pandoc.table(df, split.cells = c("80%", "10%", "10%"))

     a        b    c  
----------- ----- ----
Lorem ipsum dolor amet

If the koRpus package is installed, pandoc.table can even split the cells with hyphening support:

> pandoc.table(data.frame(baz = 'foobar'), use.hyphening = TRUE, split.cells = 3)


Minor features

Funtionality described in other sections is most notable, but pander/pandoc.table also has smaller nifty features that are worth mentioning:

> pandoc.table(mtcars[1:3, 1:4])

      &nbsp;         mpg   cyl   disp   hp 
------------------- ----- ----- ------ ----
   **Mazda RX4**     21     6    160   110 

 **Mazda RX4 Wag**   21     6    160   110 

  **Datsun 710**    22.8    4    108    93 

> pandoc.table(mtcars[1:3, 1:4], plain.ascii = TRUE)

                     mpg   cyl   disp   hp 
------------------- ----- ----- ------ ----
     Mazda RX4       21     6    160   110 

   Mazda RX4 Wag     21     6    160   110 

    Datsun 710      22.8    4    108    93 
> m <- mtcars[1:3, 1:5]
> m$mpg <- NA
> pandoc.table(m, missing = '?')

      &nbsp;         mpg   cyl   disp   hp   drat 
------------------- ----- ----- ------ ---- ------
   **Mazda RX4**      ?     6    160   110   3.9  

 **Mazda RX4 Wag**    ?     6    160   110   3.9  

  **Datsun 710**      ?     4    108    93   3.85 
> m <- data.frame(a="foo\nbar", b="pander")
> pandoc.table(m)

   a      b   
------- ------
foo bar pander

> pandoc.table(m, keep.line.breaks = TRUE)

 a    b   
--- ------
foo pander

To see all possible options, please check ?pandoc.table

And please note, that all above mentioned features are also supported by the pander generic S3 method and defaults can be updated via panderOptions for permanent settings.

Generic pander method

pander or pandoc (call as you wish) can deal with a bunch of R object types as being a pandocized S3 generic method with a variety of already supported classes:

> methods(pander)
 [1] pander.anova*           pander.aov*             pander.aovlist*         pander.Arima* *
 [6] pander.cast_df*         pander.character*       pander.clogit*          pander.coxph*           pander.cph*
[11] pander.CrossTable**      pander.Date*            pander.default*         pander.density*
[16] pander.describe*        pander.evals*           pander.factor*          pander.formula*         pander.ftable*
[21] pander.function*        pander.glm*             pander.Glm*             pander.gtable*          pander.htest*
[26] pander.image*           pander.irts*            pander.list*            pander.lm*              pander.lme*
[31] pander.logical*         pander.lrm*             pander.manova*          pander.matrix*          pander.microbenchmark*
[36] pander.mtable**            pander.nls*             pander.NULL*            pander.numeric*
[41] pander.ols*             pander.orm*             pander.polr*            pander.POSIXct*         pander.POSIXlt*
[46] pander.prcomp*          pander.randomForest*    pander.rapport*         pander.rlm*             pander.sessionInfo*
[51] pander.smooth.spline*   pander.stat.table*      pander.summary.aov*     pander.summary.aovlist* pander.summary.glm*
[56] pander.summary.lm*      pander.summary.lme*     pander.summary.manova*  pander.summary.nls*     pander.summary.polr*
[61] pander.summary.prcomp*  pander.summary.rms*     pander.summary.survreg* pander.summary.table*   pander.survdiff*
[66] pander.survfit*         pander.survreg*         pander.table*           pander.tabular*         pander.ts*
[71] pander.zoo*

If you think that pander lacks support for any other R class(es), please feel free to open a ticket suggesting a new feature or submit pull request and we will be happy to extend the package.

Besides the most basic R object types (vectors, matrices, tables or data frames), list-support might be interesting for you:

> pander(list(a = 1, b = 2, c = table(mtcars$am), x = list(myname = 1, 2), 56))

A nested list can be seen above with a table and all (optional) list names. As a matter of fact, pander.list is the default method of pander too, when you call it on an unsupported R object class:

> x <- chisq.test(table(mtcars$am, mtcars$gear))
> class(x) <- "I've never heard of!"
> pander(x)
 **WARNING**^[Chi-squared approximation may be incorrect]

  * **statistic**:


  * **parameter**:


  * **p.value**: _2.831e-05_
  * **method**: Pearson's Chi-squared test
  * ****: table(mtcars$am, mtcars$gear)
  * **observed**:

    &nbsp;   3   4   5
    ------- --- --- ---
     **0**  15   4   0

     **1**   0   8   5

  * **expected**:

    &nbsp;    3     4     5
    ------- ----- ----- -----
     **0**  8.906 7.125 2.969

     **1**  6.094 4.875 2.031

  * **residuals**:

    &nbsp;    3            4                 5
    ------- ------ ----------------- -----------------
     **0**  2.042  -1.17073226447712 -1.72300609401128

     **1**  -2.469 1.41534629268074  2.08301279585419

  * **stdres**:

    &nbsp;    3            4                5
    ------- ------ ----------------- ----------------
     **0**  4.395  -2.32338345119108 -2.9429523715087

     **1**  -4.395 2.32338345119108  2.9429523715087

<!-- end of list -->

So pander showed a not known class in an (almost) user-friendly way. And we got some warnings too styled with Pandoc footnote! If that document is exported to e.g. HTML or pdf, then the error/warning message could be found on the bottom of the page with a link. Note: there were two warnings in the above call - both captured and returned! Well, this is the feature of Pandoc.brew, see below.

But the output of different statistical methods are tried to be prettyfied. Some the above call normally returns like:

> pander(chisq.test(table(mtcars$am, mtcars$gear)))

 Test statistic   df      P value    
---------------- ---- ---------------
     20.94        2   2.831e-05 * * *

Table: Pearson's Chi-squared test: `table(mtcars$am, mtcars$gear)`

 **WARNING**^[Chi-squared approximation may be incorrect]

A few other examples on the supported R classes:

> pander(t.test(extra ~ group, data = sleep))

 Test statistic   df    P value   Alternative hypothesis 
---------------- ----- --------- ------------------------
     -1.861      17.78  0.07939         two.sided        

Table: Welch Two Sample t-test: `extra` by `group`

> ## Dobson (1990) Page 93: Randomized Controlled Trial (examples from: ?glm)
> counts <- c(18, 17, 15, 20, 10, 20, 25, 13, 12)
> outcome <- gl(3, 1, 9)
> treatment <- gl(3, 3)
> m <- glm(counts ~ outcome + treatment, family = poisson())
> pander(m)

     &nbsp;        Estimate   Std. Error   z value   Pr(>|z|) 
----------------- ---------- ------------ --------- ----------
  **outcome2**     -0.4543      0.2022     -2.247    0.02465  

  **outcome3**      -0.293      0.1927      -1.52     0.1285  

 **treatment2**   1.338e-15      0.2      6.69e-15      1     

 **treatment3**   1.421e-15      0.2      7.105e-15     1     

 **(Intercept)**    3.045       0.1709      17.81   5.427e-71 

Table: Fitting generalized (poisson/log) linear model: counts ~ outcome + treatment

> pander(anova(m))

    &nbsp;       Df   Deviance   Resid. Df   Resid. Dev 
--------------- ---- ---------- ----------- ------------
   **NULL**      NA      NA          8         10.58    

  **outcome**    2     5.452         6         5.129    

 **treatment**   2   2.665e-15       4         5.129    

Table: Analysis of Deviance Table

> pander(aov(m))

    &nbsp;       Df   Sum Sq    Mean Sq   F value   Pr(>F) 
--------------- ---- --------- --------- --------- --------
  **outcome**    2     92.67     46.33     2.224    0.2242 

 **treatment**   2   8.382e-31 4.191e-31 2.012e-32    1    

 **Residuals**   4     83.33     20.83      NA        NA   

Table: Analysis of Variance Model

> pander(prcomp(USArrests))

    &nbsp;       PC1     PC2      PC3      PC4   
-------------- ------- -------- -------- --------
  **Murder**   0.0417  -0.04482 0.07989  -0.9949 

 **Assault**   0.9952  -0.05876 -0.06757 0.03894 

 **UrbanPop**  0.04634  0.9769  -0.2005  -0.05817

   **Rape**    0.07516  0.2007   0.9741  0.07233 

Table: Principal Components Analysis

> pander(density(mtcars$hp))

   &nbsp;      Coordinates   Density values 
------------- ------------- ----------------
  **Min.**       -32.12          5e-06      

 **1st Qu.**      80.69        0.0004068    

 **Median**       193.5         0.001665    

  **Mean**        193.5         0.002214    

 **3rd Qu.**      306.3         0.00409     

  **Max.**        419.1         0.006051    

Table: Kernel density of *mtcars$hp* (bandwidth: 28.04104)

> ## Don't like scientific notation?
> panderOptions('round', 2)
> pander(density(mtcars$hp))

   &nbsp;      Coordinates   Density values 
------------- ------------- ----------------
  **Min.**       -32.12            0        

 **1st Qu.**      80.69            0        

 **Median**       193.5            0        

  **Mean**        193.5            0        

 **3rd Qu.**      306.3            0        

  **Max.**        419.1           0.01      

Table: Kernel density of *mtcars$hp* (bandwidth: 28.04104)

And of course tables are formatted (e.g. auto add of line breaks, splitting up tables, hyphenation support or markdown format) based on the user specified panderOptions.

Creating Pandoc documents

The package is also capable of creating complex Pandoc documents (reports) from R objects in multiple ways:

Brew to Pandoc

The brew package, which is a templating framework for report generation, has not been updated since 2011, but it's still used in bunch of R projects based on its simple design and useful features in literate programming. For a quick overview, please see the following documents if you are not familiar with brew:

In short: a brew document is a simple text file with some special tags. Pandoc.brew uses only two of them (as building on a personalized version of Jeff's really great brew function):

This latter tries to be smart in some ways:

Besides this, the custom brew function can do more and also less compared to the original brew package. First of all, the internal caching mechanism of brew has been removed and rewritten for some extra profits besides improved caching.

For example now multiple R expressions can be passed between the <%= ... %> tags, and not only the text results, but the evaluated R objects are also (invisibly) returned in a structured list. This can be really useful while post-processing the results of brew. Quick example:

> str(Pandoc.brew(text ='
+   Pi equals to `<%= pi %>`.
+   And here are some random data:
+   `<%= runif(10) %>`
+ '))

Pi equals to _3.142_.
And here are some random data:
_0.6631_, _0.849_, _0.06986_, _0.3343_, _0.5209_, _0.3471_, _0.866_, _0.05548_, _0.8933_ and _0.2121_

List of 2
 $ :List of 4
  ..$ type  : chr "text"
  ..$ text  :List of 2
  .. ..$ raw : chr "Pi equals to _3.142_.\nAnd here are some random data:\n"
  .. ..$ eval: chr "Pi equals to _3.142_.\nAnd here are some random data:\n"
  ..$ chunks:List of 2
  .. ..$ raw : chr "_3.142_"
  .. ..$ eval: chr "_3.142_"
  ..$ msg   :List of 3
  .. ..$ messages: NULL
  .. ..$ warnings: NULL
  .. ..$ errors  : NULL
 $ :List of 2
  ..$ type   : chr "block"
  ..$ robject:List of 6
  .. ..$ src   : chr "runif(10)"
  .. ..$ result: num [1:10] 0.6631 0.849 0.0699 0.3343 0.5209 ...
  .. ..$ output: chr "_0.6631_, _0.849_, _0.06986_, _0.3343_, _0.5209_, _0.3471_, _0.866_, _0.05548_, _0.8933_ and _0.2121_"
  .. ..$ type  : chr "numeric"
  .. ..$ msg   :List of 3
  .. .. ..$ messages: NULL
  .. .. ..$ warnings: NULL
  .. .. ..$ errors  : NULL
  .. ..$ stdout: NULL
  .. ..- attr(*, "class")= chr "evals"

This document was generated by Pandoc.brew based on inst/README.brew so the above examples were generated automatically by running:

Pandoc.brew(system.file('README.brew', package = 'pander'))

The output is set to stdout by default, which means that the resulting text is written to the R console. But setting the output to a text file and running Pandoc on that to create a HTML, odt, docx or other document in one go is also possible. To export a brewed file to other then Pandoc's markdown, please use the convert parameter. For example:

text <- paste('# Header',
              'What a lovely list:\n<%= as.list(runif(10)) %>',
              'A wide table:\n<%= mtcars[1:3, ] %>',
              'And a nice chart:\n\n<%= plot(1:10) %>',
          sep = '\n')
Pandoc.brew(text = text, output = tempfile(), convert = 'html')
Pandoc.brew(text = text, output = tempfile(), convert = 'pdf')

So to brew this README with all R chunks automatically converted to html, please run:

Pandoc.brew(system.file('README.brew', package='pander'), output = tempfile(), convert = 'html')


The package bundles some examples for Pandoc.brew to let you check its features pretty fast. These are:

To brew these examples on your machine, try to run the followings commands:

Pandoc.brew(system.file('examples/minimal.brew', package='pander'))
Pandoc.brew(system.file('examples/minimal.brew', package='pander'), output = tempfile(), convert = 'html')

Pandoc.brew(system.file('examples/short-code-long-report.brew', package='pander'))
Pandoc.brew(system.file('examples/short-code-long-report.brew', package='pander'), output = tempfile(), convert = 'html')

Pandoc.brew(system.file('examples/graphs.brew', package='pander'))
Pandoc.brew(system.file('examples/graphs.brew', package='pander'), output = tempfile(), convert = 'html')

For easier access, I have uploaded some exported documents of the above examples as well:

Please check out pdf, docx, odt and other formats by changing the above convert option on your machine, and do not forget to give some feedback!

Live report generation

pander package has a special reference class called Pandoc which could collect some blocks in a live R session and export the whole document to Pandoc/PDF/HTML etc. Without any serious further explanations, please check out the below (self-commenting) example:

## Initialize a new Pandoc object
myReport <- Pandoc$new()

## Add author, title and date of document
myReport$author <- 'Gergely Daróczi'
myReport$title  <- 'Demo'

## Or it could be done while initializing
myReport <- Pandoc$new('Gergely Daróczi', 'Demo')

## Add some free text
myReport$add.paragraph('Hello there, this is a really short tutorial!')

## Add maybe a header for later stuff
myReport$add.paragraph('# Showing some raw R objects below')

## Adding a short matrix

## Or a table with even
myReport$add.paragraph('Hello table:')
myReport$add(table(mtcars$am, mtcars$gear))

## Or a "large" data frame which barely fits on a page

## And a simple linear model with Anova tables
ml <- with(lm(mpg ~ hp + wt), data = mtcars)

## And do some principal component analysis at last

## Sorry, I did not show how Pandoc deals with plots:

## Want to see the report? Just print it:

## Exporting to PDF (default)

## Or to docx in tempdir():
myReport$format <- 'docx'

## You do not want to see the generated report after generation?
myReport$export(open = FALSE)

Capturing evaluation information with evals

When working on the rapport package, I really needed some nifty R function that can evaluate R expression along with capturing errors and warnings. Unfortunately the evaluate package had only limited features at that time, as it could not return the raw R object, but only the standard output with messages. So I wrote my own function, and soon some further feature requests arose, like identifying if an R expression results in a plot etc. This section aims to give a quick introduction to the functionality of evals, but for more usage/implementation details, please refer to specialized vignette, which can be accessed by vignette('evals', package='pander') or available online here.

But probably it's easier to explain what evals can do with a simple example:

> evals('1:10')
[1] "1:10"

 [1]  1  2  3  4  5  6  7  8  9 10

[1] " [1]  1  2  3  4  5  6  7  8  9 10"

[1] "integer"





[1] "evals"

So evals can evaluate a character vector of R expressions, and it returns a list of captured stuff while running those:

Besides capturing this nifty list of important circumstances, evals can automatically identify if an R expression is returning anything to a graphical device, and can save the resulting image in a variety of file formats along with some extra options, like applying a custom theme on base, lattice or ggplot2 plots:

> evals('hist(mtcars$hp)')[[1]]$result

So instead of a captured R object (which would be NULL in this situation by the way), we get the path of the image where the plot was saved:

Well, this is not a standard histogram usually returned by the hist function, right? As mentioned before, evals have some extra features like applying the user defined theme on various plots automatically. Please see the graphs.brew example above for further details, or check the related global options. If you do not like this feature, simply add evalsOptions('graph.unify, FALSE)to your.Rprofile`.

Further features are described in the technical docs, and now I'll only give a brief introduction to another important feature of evals.


As pander::evals is using a custom caching algorithm in the means of evaluating R expressions, it might be worthwhile to give a short summary of what is going on in the background when you are running e.g. Pandoc.brew, the "live report generation" engine or evals directly:

As pander does not cache based on raw sources of chunks and there is no easy way of enabling/disabling caching on a chunk basis, the users have to live with some great advantages and some minor tricky situations - which latter cannot be solved theoretically in my opinion, but I'd love to hear your feedback.

The caching hash is computed based on the structure and content of the R commands instead of the used variable names or R expressions, so let us make some POC example to show the greatest asset:

x <- mtcars$hp
y <- 1e3
evals('sapply(rep(x, y), mean)')

It took a while, huh? :)

Let us create some custom functions and variables, which are not identical to the above call:

f <- sapply
g <- rep
h <- mean
X <- mtcars$hp * 1
Y <- 1000

And now try to run something like:

evals('f(g(X, Y), h)')

Yes, it was returned from cache!

About the kickback:

As pander (or rather: evals) does not really deal with what is written in the provided sources but rather checks what is inside that, there might be some tricky situations where you would expect the cache to work, but it would not. Short example: we are computing and saving to a variable something heavy in a chunk (please run these in a clean R session to avoid conflicts):

evals('x <- sapply(rep(mtcars$hp, 1e3), mean)')

It is cached, just run again, you will see.

But if you would create x in your global environment with any value (which has nothing to do with the special environment of the report!) and x was not defined in the report before this call (and you had no x value in your global environment before), then the content of x would result in a new hash for the cache - so caching would not work. E.g.:

x <- 'foobar'
evals('x <- sapply(rep(mtcars$hp, 1e3), mean)')

I really think this is a minor issue (with very special coincidences) which cannot be addressed cleverly - but could be avoided with some cautions (e.g. run Pandoc.brew in a clean R session like with Rscript or littler - if you are really afraid of this issue). And after all: you loose nothing, just the cache would not work for that only line and only once in most of the cases.

Other cases when the hash of a call will not match cached hashes:

But the e.g. following do work from cache fine:

x  <- mtcars$hp
xx <- mtcars$hp*1

General options

The package comes with a variety of globally adjustable options, which have an effect on the result of your reports. You can query and update these options with the panderOptions function:

Besides localization of numeric formats or the styles of tables, lists and plots, there are some technical options as well, which would effect e.g. caching or the format of rendered image files. You can query/update those with the evalsOptions function as the main backend of pander calls is a custom evaluation function called evals.

The list of possible options are:

Difference from other rendering packages

How does pander differ from Sweave, brew, knitr, R2HTML and the other tools of literate programming? First of all pander can be used as a helper with any other literate programming solution, so you can call pander inside of knitr chunks.

But if you stick with pander's literate programming engine, then there's not much need for calling ascii, xtable, Hmisc, tables etc. or even pander in the R command chunks to transform R objects into markdown, HTML, tex etc. as Pandoc.brew automatically results in Pandoc's markdown, which can be converted to almost any text document format. Conversion can be done automatically after calling pander reporting functions (Pander.brew or Pandoc).

Based on the fact that pander transforms R objects into markdown, no "traditional" R console output is shown in the resulting document (nor in markdown, nor in exported docs), but all R objects are transformed to tables, list etc. Well, there is an option (show.src) to show the original R commands before the formatted output, and pander calls can be also easily tweaked to return the printed version of the R objects - if you would need that in some strange situation - like writing an R tutorial. But really think that nor R code, nor raw R results have anything to do with an exported report.

Of course all warnings, messages and errors are captured while evaluating R expressions just like stdout besides the raw R objects. So the resulting report also includes the raw R objects for further edits if needed - which is a very unique feature.

Graphs and plots are automatically identified in code chunks and saved to disk in a png file linked in the resulting document. This means that if you create a report (e.g. brew a text file) and export it to PDF/docx etc. all the plots/images would be there. There are some parameters to specify the resolution of the image and also the type (e.g. jpg, svg or pdf) besides a wide variety of theme options. About the latter, please check the graphs.brew example above.

And pander uses its built-in (IMHO quite decent) caching engine. This means that if the evaluation of some R commands takes too long time (which can be set by option/parameter), then the results are saved in a file and returned from there on next similar R code's evaluation. This caching algorithm tries to be smart, as it not only checks the passed R sources, but the content of all variables and functions, and saves the hash of those. This is a quite secure way of caching (see details above), but if you would encounter any issues, just switch off the cache. I've not seen any issues for years :)


I have created some simple LISP functions which would be handy if you are using the best damn IDE for R. These functions and default key-bindings are shipped with the package, feel free to personalize.

As time passed these small functions grew heavier (with my Emacs knowledge) so I ended up with a small library:


I am currently working on pander-mode which is a small minor-mode for Emacs. There are a few (but useful) functions with default keybindings:

Few options of pander-mode: M-x customize-group pander

To use this small lib, just type: M-x pander-mode on any document. It might be useful to add a hook to markdown-mode if you find this useful.