system.file("include", package = "BH") returns empty string about bh

[eddelbuettel]

Do a dpkg -L r-cran-bh: it is an empty virtual shell. For a reason: BH has a footprint of 150mb installed and on Debian you do not need it. Use libboost-all-dev instead.

That said, why did you think boost_lib <- system.file("include", package = "BH") was a good idea? Besides the obvious gotcha of calling something _lib when it checks for includes.

Anyway, if this is handed down from rstan than you may have not choice but to install.packages("BH") and waste those 150mb if that is how rstan wants to access it. It should know better (we often use configure or cmake).

In any event this is not an issue of the CRAN package BH where you filed the issue but with the (trying to be helpful) Debian package r-cran-bh. I stand behind both so I can help so fire up if you have questions.

[robmoss]

Sorry for filing the issue here rather than Debian/Ubuntu bug trackers, I wasn't sure which was the most appropriate. It's obvious in hindsight.

That said, why did you think boost_lib <- system.file("include", package = "BH") was a good idea? Besides the obvious gotcha of calling something _lib when it checks for includes.

This is how rstan locates the boost header files when compiling a model (at least, on my Ubuntu and Debian machines).

I knew that r-cran-bh is an empty package, but I didn't appreciate that install.packages("BH") installed a second copy of the boost header files. So I assumed that the r-cran-bh package would (or should) work with rstan without any further actions on my part.

[eddelbuettel]

It is kinda the other way around:

• install.packages("BH") is an R function that does not know it is on Debian, it installs full BH
• r-cran-bh is a small hack attempting to save the second installation of Boost (besides libboost-*)
• rstan does not know about point 2) (not should it) so it ask "where is BH"
• what should really happy is for Boost to work with pkg-config because a ton of CRAN packages use that to find include and library locations

[robmoss]

Thank you very much for the explanation Dirk.