Deal with new CRAN clang-UBSAN additional issue

[seananderson]

https://www.stats.ox.ac.uk/pub/bdr/memtests/clang-UBSAN/sdmTMB/sdmTMB-Ex.Rout

Error arises at:

> ### Name: visreg_delta
> ### Title: Plot sdmTMB models with the 'visreg' package
> ### Aliases: visreg_delta visreg2d_delta
> 
> ### ** Examples
> 
> if (inla_installed() &&
+   require("ggplot2", quietly = TRUE) &&
+   require("visreg", quietly = TRUE)) {
+ 
+   pcod_2011$fyear <- as.factor(pcod_2011$year)
+   fit <- sdmTMB(
+     density ~ s(depth_scaled) + fyear,
+     data = pcod_2011, mesh = pcod_mesh_2011,
+     spatial = "off",
+     family = tweedie()
+   )
+   visreg::visreg(fit, xvar = "depth_scaled")
+   visreg::visreg(fit, xvar = "fyear")
+   visreg::visreg2d(fit, xvar = "fyear", yvar = "depth_scaled")
+ 
+ }
/data/gannet/ripley/R/test-clang/RcppEigen/include/Eigen/src/Core/MathFunctions.h:372:33: runtime error: nan is outside the range of representable values of type 'int'

Perhaps painful in the short run, but one option is to submit the pull request to ggeffects so that ggpredict works and drop visreg support.

[seananderson]

For future reference, these 2 commits fixed the valgrind and clang-ASAN/UBSAN errors: https://github.com/pbs-assess/sdmTMB/commit/a7026aba725f4f926af70c4e3eb35a0f94b48d37 https://github.com/pbs-assess/sdmTMB/commit/de1efa2c8848985e943064a9595ad92b98325e56

My Docker file:

FROM wch1/r-debug
RUN apt-get -y install cmake
RUN apt-get -y install libnlopt-dev
RUN apt-get -y install libudunits2-dev
RUN apt-get -y install libgdal-dev
RUN apt-get -y install libproj-dev
RUN RDvalgrind -e "install.packages('s2')"
RUN RDvalgrind -e "install.packages('sf')"
RUN RDvalgrind -e "install.packages('lme4')"
RUN RDvalgrind -e "install.packages('mgcv')"
RUN RDvalgrind -e "install.packages('TMB')"
RUN RDvalgrind -e "install.packages('fmesher')"
RUN RDvalgrind -e "install.packages('assertthat')"
RUN RDvalgrind -e "install.packages('rlang')"
RUN RDvalgrind -e "install.packages('ggeffects')"
RUN RDvalgrind -e "install.packages('dplyr')"
RUN RDvalgrind -e "install.packages('visreg')"
RUN RDvalgrind -e "install.packages('devtools')"
RUN RDvalgrind -e "install.packages('clisymbols')"
RUN RDvalgrind -e "install.packages('fishMod')"
RUN RDvalgrind -e "install.packages('mvtnorm')"
RUN RDvalgrind -e "install.packages('glmmTMB')"
RUN RDvalgrind -e "install.packages('future')"
RUN RDvalgrind -e "install.packages('future.apply')"
RUN RDvalgrind -e "install.packages('ggforce')"
RUN RDvalgrind -e "install.packages('estimability')"
RUN RDvalgrind -e "install.packages('effects')"
RUN RDvalgrind -e "install.packages('emmeans')"
RUN RDvalgrind -e "install.packages('splancs')"

With another using RDcsan instead of RDvalgrind.

Do this on an Intel machine. M1/M2 Macs will be very slow with Docker here.

Increase Docker's RAM to 8 GB!

The error:

g++: fatal error: Killed signal terminated program cc1plus 

Means the system ran out of memory when building a package!

Build it:

docker build --tag "r-debugging" .

Run it:

docker run --rm -v ~/src/sdmTMB:/sdmTMB -ti --security-opt seccomp=unconfined r-debugging

Put a test script in scratch/test.R within sdmTMB. Then:

cp -r sdmTMB testdir
cd testdir
RDvalgrind -e "devtools::install()"
RDvalgrind -d valgrind -e "source('scratch/test.R')"

Valgrind is slow. Keep examples short. All examples took ~11 hours!

Or run all examples:

RDvalgrind -e "devtools::install()"
RDvalgrind -d valgrind -e 'devtools::run_examples()'

Or alternatively:

RDvalgrind  -e "rcmdcheck::rcmdcheck(check_dir = './check')"
# cd as needed to 'check'
RDvalgrind -d valgrind --vanilla < "sdmTMB.Rcheck/sdmTMB-Ex.R"

For, RDSan, I ran R CMD check:

From other Docker:

RDcsan -e "rcmdcheck::rcmdcheck()"

This only took ~25 minutes.

Run tests with better location reporting:

RDcsan -e 'devtools::test("testdir", reporter = testthat::LocationReporter)'