Nested max(int, int) expressions cause compiler error in RStan / StanHeaders 2.32.5

[ebuhle]

Summary:

Compilation fails with rstan 2.32.5 and StanHeaders 2.32.5 under R 4.3.x if the model code contains an expression of the form max(max(int, int), int), i.e. nested calls to max() with integer arguments.

Description:

After recently updating from rstan 2.32.2 and StanHeaders 2.26.28 to the current 2.32.5 versions of both packages in R 4.3.2, I was unable to compile some previously working Stan models. Through a lot of trial and error, removing code blocks until compilation succeeded and then adding them back until it failed again, I managed to isolate the issue to nested integer-max calls. If a model contains an expression like max(max(int, int), int), the compiler throws an error. If the inner max(int, int) is replaced by a constant integer or a previously defined int variable, or if the arguments to max() are real instead of int, compilation works.

A colleague and I have reproduced this in R 4.3.0 and 4.3.2 on two machines running different versions of Windows.

Reproducible Steps:

test1.stan <- "transformed data {
  int test = max(max(1, 1), 1);
}
"

test1 <- rstan::stan_model(model_code = test1.stan)  # error

test2.stan <- "transformed data {
  int test = max(1, 1);
}
"

test2 <- rstan::stan_model(model_code = test2.stan)  # no error

Current Output:

The first example produces the following compiler output followed by an error:

make cmd is
  make -f "C:/R/R-43~1.2/etc/x64/Makeconf" -f "C:/R/R-43~1.2/share/make/winshlib.mk" -f "C:/~/.R/Makevars.win" CXX='$(CXX17) $(CXX17STD)' CXXFLAGS='$(CXX17FLAGS)' CXXPICFLAGS='$(CXX17PICFLAGS)' SHLIB_LDFLAGS='$(SHLIB_CXX17LDFLAGS)' SHLIB_LD='$(SHLIB_CXX17LD)' SHLIB="file7fc28f11320.dll" WIN=64 TCLBIN= OBJECTS="file7fc28f11320.o"

make would use
g++  -std=gnu++17 -I"C:/R/R-43~1.2/include" -DNDEBUG   -I"C:/R/R-4.3.2/library/Rcpp/include/"  -I"C:/R/R-4.3.2/library/RcppEigen/include/"  -I"C:/R/R-4.3.2/library/RcppEigen/include/unsupported"  -I"C:/R/R-4.3.2/library/BH/include" -I"C:/R/R-4.3.2/library/StanHeaders/include/src/"  -I"C:/R/R-4.3.2/library/StanHeaders/include/"  -I"C:/R/R-4.3.2/library/RcppParallel/include/" -DRCPP_PARALLEL_USE_TBB=1 -I"C:/R/R-4.3.2/library/rstan/include" -DEIGEN_NO_DEBUG  -DBOOST_DISABLE_ASSERTS  -DBOOST_PENDING_INTEGER_LOG2_HPP  -DSTAN_THREADS  -DUSE_STANC3 -DSTRICT_R_HEADERS  -DBOOST_PHOENIX_NO_VARIADIC_EXPRESSION  -D_HAS_AUTO_PTR_ETC=0  -include "C:/R/R-4.3.2/library/StanHeaders/include/stan/math/prim/fun/Eigen.hpp"  -std=c++1y    -I"C:/rtools43/x86_64-w64-mingw32.static.posix/include"     -O2 -Wall  -mfpmath=sse -msse2 -mstackrealign  -c file7fc28f11320.cpp -o file7fc28f11320.o
if test "zfile7fc28f11320.o" != "z"; then \
  if test -e "file7fc28f11320-win.def"; then \
    echo g++  -shared -s -static-libgcc -o file7fc28f11320.dll file7fc28f11320-win.def file7fc28f11320.o  "C:/R/R-4.3.2/library/rstan/lib/x64/libStanServices.a" -L"C:/R/R-4.3.2/library/StanHeaders/libs/x64" -lStanHeaders -L"C:/R/R-4.3.2/library/RcppParallel/lib/x64" -ltbb -LC:/R/R-4.3.2/library/RcppParallel/lib/x64 -ltbb -ltbbmalloc -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib/x64" -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib"  -L"C:/R/R-43~1.2/bin/x64" -lR ; \
    g++  -shared -s -static-libgcc -o file7fc28f11320.dll file7fc28f11320-win.def file7fc28f11320.o  "C:/R/R-4.3.2/library/rstan/lib/x64/libStanServices.a" -L"C:/R/R-4.3.2/library/StanHeaders/libs/x64" -lStanHeaders -L"C:/R/R-4.3.2/library/RcppParallel/lib/x64" -ltbb -LC:/R/R-4.3.2/library/RcppParallel/lib/x64 -ltbb -ltbbmalloc -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib/x64" -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib"  -L"C:/R/R-43~1.2/bin/x64" -lR ; \
  else \
    echo EXPORTS > tmp.def; \
    nm file7fc28f11320.o | sed -n 's/^.* [BCDRT] / /p' | sed -e '/[.]refptr[.]/d' -e '/[.]weak[.]/d' | sed 's/[^ ][^ ]*/"&"/g'  >> tmp.def; \
    echo g++  -shared -s -static-libgcc -o file7fc28f11320.dll tmp.def file7fc28f11320.o  "C:/R/R-4.3.2/library/rstan/lib/x64/libStanServices.a" -L"C:/R/R-4.3.2/library/StanHeaders/libs/x64" -lStanHeaders -L"C:/R/R-4.3.2/library/RcppParallel/lib/x64" -ltbb -LC:/R/R-4.3.2/library/RcppParallel/lib/x64 -ltbb -ltbbmalloc -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib/x64" -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib"  -L"C:/R/R-43~1.2/bin/x64" -lR ; \
    g++  -shared -s -static-libgcc -o file7fc28f11320.dll tmp.def file7fc28f11320.o  "C:/R/R-4.3.2/library/rstan/lib/x64/libStanServices.a" -L"C:/R/R-4.3.2/library/StanHeaders/libs/x64" -lStanHeaders -L"C:/R/R-4.3.2/library/RcppParallel/lib/x64" -ltbb -LC:/R/R-4.3.2/library/RcppParallel/lib/x64 -ltbb -ltbbmalloc -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib/x64" -L"C:/rtools43/x86_64-w64-mingw32.static.posix/lib"  -L"C:/R/R-43~1.2/bin/x64" -lR ; \
    rm -f tmp.def; \
  fi \
fi

Error in compileCode(f, code, language = language, verbose = verbose) :
C:/R/R-4.3.2/library/RcppEigen/include/Eigen/src/Core/ProductEvaluators.h:35:90: required from 'Eigen::internal::evaluator<Eigen::Product<Lhs, Rhs, Option> >::evaluator(const XprType&) [with Lhs = Eigen::Product<Eigen::CwiseBinaryOp<Eigen::internal::scalar_product_op<double, double>, const Eigen::CwiseNullaryOp<Eigen::internal::scalar_constant_op<double>, const Eigen::Matrix<double, 1, -1> >, const Eigen::Transpose<Eigen::Matrix<double, -1, 1> > >, Eigen::Matrix<double, -1, -1>, 0>; Rhs = Eigen::Matrix<double, -1, 1>; int Options = 0; XprType = Eigen::Product<Eigen::Product<Eigen::CwiseBinaryOp<Eigen::internal::scalar_product_op<double, double>, const Eigen::CwiseNullaryOp<Eigen::internal::scalar_constant_op<double>, const Eigen::Matrix<double, 1, -1> >, const Eigen::Transpose<Eigen::Matrix<double, -1, 1> > >, Eigen::Matrix<double, -1, -1>, 0>, Eigen::Matrix<double, -1, 1>, 0>]'C:/R/R-4.3.2/library/RcppEigen/include/Eigen/src/Core/Product.h:132:22: required from 'Eigen::in

Error in sink(type = "output") : invalid connection

Expected Output:

Successful compilation producing a stanmodel object.

RStan Version:

rstan 2.32.5 and StanHeaders 2.32.5

R Version:

R 4.3.0 or 4.3.2

Operating System:

Windows

[WardBrian]

Thanks for reporting, this was a bug in our math library.

@bgoodri this is another really easy backport: replace return_type_t<T1, T2> with auto for the prim overload of max (and min, too, which has the same issue)

[bgoodri]

Will do