Reducing R package lib size further?...
danielinteractive opened this issue · 6 comments
Hi @kaskr and all,
first of all thanks a lot for the TMB package, it is really useful and great!
Despite having developed code now with TMB for a couple of months, the whole compilation options are still a bit black box for me I feel. Right now we are preparing a package (https://github.com/openpharma/mmrm) for CRAN release. I am worried that the large lib size (still 73 Mb with latest version) will be a problem, as we already get a note for that.
I already trimmed the files down to just one cpp file which includes the TMB header. Anything else which we can do?
Any suggestions would be awesome, thanks!
Cheers
Daniel
Hi @danielinteractive,
I don't have any ideas for the reducing things at the TMB end, but in the past I have removed the testthat element of a package to reduce the file size below CRAN's 5MB limit, and relied on running those tests on github. It's also possible that you would just have to give up on CRAN and continue to rely on github as the installation source.
-Ian
@danielinteractive The large shared objects are due to debugging information. For example, glmmTMB.so is currently 69M with a default build. By removing -g from the compile line it gets to 4.9M. The '-g' usually appears in R CMD config CXXFLAGS which one can modify (by using a custom Makevars file).
The size can be reduced even more by removing symbol information. By adding -Wl,-s to the linker (see https://github.com/kaskr/adcomp/wiki/FAQ) it gets to 3.7M.
Unfortunately, none of these solutions are allowed by CRAN.
There might be some more info in this thread:
https://stat.ethz.ch/pipermail/r-devel/2016-October/073271.html
Thanks a lot @iantaylor-NOAA and @kaskr for your helpful replies!
Luckily and surprisingly at the same time this issue did not come up on CRAN pretest runs ... I hope that it is not an issue. (I wonder if CRAN installs packages with the --strip option and therefore reduces size automatically?
@danielinteractive I came across this Makevars https://github.com/jkbest2/spatq/blob/aef9ee5f8eca811ee11c484ee803ed7485775195/src/Makevars, which uses a solution by Dirk Eddelbuettel (link in file) that seems to help (by magic!).
@paddy7wb yeah I have seen that, but unfortunately it is not allowed on CRAN, it is explicitly mentioned in the Writing R extensions.
I think we can close this.