Chunk Reference/Macro

Sweave has the feature of chunk reference to reuse chunks with the syntax <<chunk-label>> (without = compared to <<>>=), e.g.

<<chunk1>>=
1 + 1
@

<<chunk2>>=
<<chunk1>>
@

In chunk2, the code in chunk1 will be inserted. This feature is also available in knitr, but note knitr supports arbitrary (finite) levels of recursion in chunk references (Sweave only supports one level), i.e. one chunk can reference another chunk which references yet another chunk.

This same <<chunk-label>> syntax will work in markdown, to reuse a previous named markdown chunk within another chunk, even though your .Rmd editor will likely flag the line as an unexpected token.

There are still other approaches to reuse chunks in knitr.

1. use the same label as the previous chunk to be reused
2. use the chunk option ref.label to reference another chunk

Use the same label

An example for the first approach:

<<chunk1, echo=TRUE, results='hide'>>=
1 + 1
@

<<chunk1, echo=FALSE, results='markup'>>=
@

The second chunk is empty, so knitr will look for another chunk with the same label but is not empty, and use the code from that chunk. The key is to leave a chunk empty to use code from other chunks. One problem with this approach is that you cannot cache both chunks since their MD5 digests are different, and knitr only allows one set of cache files per label.

Use chunk option ref.label

An example for the second approach:

<<chunk1, echo=TRUE, results='hide'>>=
1 + 1
@

<<chunk2, ref.label='chunk1', echo=FALSE, results='markup'>>=
@

The second chunk uses a different label, so cache is no longer a problem. Obviously the second approach is a more general solution.

This feature enables us to separate R code and R output in the output document. For instance, we can use echo=FALSE in the body of an article to hide R code, and use chunk references in the appendix to show R code (with eval=FALSE, ref.label=...).