A nice one (for Bayesian users of Sage's R)

[Emmanuel Charpentier]

A nice little puzzle for Sage's R users trying to use R 3.3.1 (forthcoming in Trac#20523) : the (still damnably useful, notwithstanding rstan) rjags package no longer installs.

rjags calls the system's jags binary and system's jags modules (compiled as .so objects in Unix). However, there is currently (i. e. sage 7.4 beta4) a conflict between the system's and Sage's notion of what BLAS should do : my (Debian) system's BLAS wants to use  a gotoblas function, not found in what Sage uses for BLAS.

A possible workaround is to install a Sage-specific version of JAGS from the source, configured with something like "./configure --prefix=$SAGE_ROOT/local") in a Sage shell (and compiled and installed in the Sage shell...). This worked like a charm for me.

I do not know Sage's build system well enough to know if forcing Sage to use system's numerical livraries (by forcing, e. g. to use the local ATLAS libraries) should be enough.

HTH,

--
Emmanuel Charpentier

[Dima Pasechnik]

if you want to use system's BLAS/ATLAS, set SAGE_ATLAS_LIB to point to the appropriate location

(usually, /usr/lib, I suppose), and rebuild from scratch

for more details see http://doc.sagemath.org/html/en/installation/source.html#environment-variables

It is a bug in the R package in question that it has hardcoded the location of BLAS to the system's one.

[Emmanuel Charpentier]

Thank you Dima ! I'll try that.

--
Emmanuel Charpentier