Error in dyn.load(libLFile): unable to load shared object

Summary:

After fitting a larger number of models in one R session (roughly 100, didn't count and can't reconstruct), I got an error message saying that no more shared objects can be loaded.

Description:

Rstan wasn't used directly but through the brms package (the maintainer, @paul-buerkner, suggested that I post here). Not sure if relevant but this happens when I run R in Emacs ESS, e.g., when compiling a knitr document multiple times in a row without caching.

Perhaps, rstan could unload shared objects after sampling finishes and reload them as needed?

Reproducible Steps:

Start an R session. Fit many models. In my case about 100.

Current Output:

Error in dyn.load(libLFile) : 
  unable to load shared object '/tmp/RtmpJ7KuDT/file3745e4a2a37.so':
  `maximal number of DLLs reached...

Expected Output:

No error message.

RStan Version:

The version of RStan you are running (e.g., from packageVersion("rstan"))

2.16.2

R Version:

The version of R you are running (e.g., from R.version.string)

R version 3.4.1 (2017-06-30)

Operating System:

Ubuntu 16.04

Start by reproducing the report in an R 3.4.1 session on Ubuntu 16.04, fitting roughly 100 models through brms and observing the dyn.load error. Trace how rstan’s shared objects are loaded and retained; done means repeated fits no longer reach the maximal DLL limit, with coverage for the reported scenario.