Precompiled vignette on CRAN
On Jul 14, 2010, at 4:04 PM, Prof Brian Ripley wrote:
On Wed, 14 Jul 2010, Felix Sch?nbrodt wrote:
Hello, my package passes R CMD check without any warnings on my local machine (Mac OS), as well as on Uwe Ligges' Winbuilder. On RForge, however, we sometimes run into problems building the Sweave vignettes.
Just 'problems' is not helpful.
FWIW I think it was a red herring - after bugfixes to the package and installing (unstated*) dependencies it works (possibly use of require instead of library in the vignette might help - I don't know what the best practice is).
Now here's my question: is it necessary for a CRAN submission that the Sweave vignettes can be compiled on CRAN, or is it possible to provide the (locally compiled) pdf vignette to be included in the package?
This really is a question to ask the CRAN gatekeepers, but people are on vacation right now, so I've give some indication of my understanding. What does 'compiled' mean here? (Run through LaTeX? Run the R code?) There are examples on CRAN of packages which cannot re-make their vignettes without external files (e.g. LaTeX style files), or take hours (literally) to run the code. The source package should contain the PDF versions of the vignettes as made by the author. There is relevant advice in 'Writing R Extensions'. What the people who do the CRAN package checks do get unhappy about are packages which fail running the R code in their vignettes, since this often indicates a problem in the package which is not exercised by the examples nor tests. This gives a warning, as you will see in quite a few CRAN package checks.
In the case of TripleR R CMD check failed (i.e. error code != 0) because of the vignette. Cheers, Simon * - which reminds me -- what is the correct place to list vignette dependencies? "Suggests:" ? R CMD check simply fails in the vignette build when vignetts use unstated dependencies [via library()] - there is no explicit warning/error by check itself.