[R-pkg-devel] weird C stack traces from win-builder, due to brms/rstan load?

[Duncan Murdoch]

On 05/07/2021 2:50 p.m., Bill Dunlap wrote:
> I think the stack trace is omitting the initial underscore in the symbol 
> name:
> 
>  > c++filt 
> _ZN4Rcpp8CppClassC1EPNS_6ModuleEPNS_10class_BaseERNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE
> Rcpp::CppClass::CppClass(Rcpp::Module*, Rcpp::class_Base*, 
> std::__cxx11::basic_string<char, std::char_traits<char>, 
> std::allocator<char> >&)

Thanks.  It doesn't work for me with the leading underscore, but I'm 
using a clang based compiler, not gcc, so maybe that explains it.  (I've 
tried most of the --format= options, without success.)

Duncan



> -Bill
> 
> On Mon, Jul 5, 2021 at 11:01 AM Duncan Murdoch <murdoch.duncan using gmail.com 
> <mailto:murdoch.duncan using gmail.com>> wrote:
> 
>     On 05/07/2021 10:32 a.m., Ben Bolker wrote:
>      >    I'm running into trouble checking a new release of the
>     'broom.mixed'
>      > package
>     <https://cran.r-project.org/web/packages/broom.mixed/index.html
>     <https://cran.r-project.org/web/packages/broom.mixed/index.html>>
>      > on win-builder.  I think it is some kind of rstan problem.
>      >
>      >     The broom.mixed package loads the brms package, and loads several
>      > pre-computed objects from the brms package; I don't think it actually
>      > tries to run anything of the functions from brms during the check
>     process.
>      >
>      >     At several points during checking (see
>      > <https://win-builder.r-project.org/LsDDMqpnX52n/00check.log
>     <https://win-builder.r-project.org/LsDDMqpnX52n/00check.log>>), I get
>      > voluminous "C stack trace" outputs: once when "Checking
>     dependencies in
>      > R code", once when calling require(brms) as part of an example, once
>      > when running the tests (not easy to determine the precise
>     location, and
>      > once when building the vignettes).
>      >
>      >     The trace starts like this:
>      >
>      > ==== C stack trace ===============================
>      >
>      >       SETCAR [0x0x6c838a90+32]
>      >     
>       ZN4Rcpp8CppClassC1EPNS_6ModuleEPNS_10class_BaseERNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE [0x0x6ac093ff+1151]
>      >       ZN4Rcpp6Module12classes_infoEv [0x0x6ac07467+327]
>      >       Z20Module__classes_infoP7SEXPREC [0x0x6abe1c4a+74]
>      >       Rf_NewFrameConfirm [0x0x6c7a7d18+33960]
>      >       R_initAssignSymbols [0x0x6c7f3259+53849]
>      >       Rf_eval [0x0x6c7fcd31+369]
>      > ...
>      >
>      >     and goes on for a while.
>      >
>      >     R CMD check runs fine on Linux with recent r-devel.
>      >
>      >     Has anyone seen this?  Other than scrubbing all references to
>     brms
>      > and rstan (a second try with one of the brms references avoided
>     instead
>      > runs into trouble when loading rstan ...), I don't see what I can
>     do ...
>      >
>      > <https://win-builder.r-project.org/YArr24tuYZ8G/00check.log
>     <https://win-builder.r-project.org/YArr24tuYZ8G/00check.log>>
>      >
> 
>     What error do you get before the stack trace?
> 
>     The Z... names look like C++ entry points to overloaded functions or
>     methods.  There's a command line function (c++filt) that will usually
>     turn those into their original readable form.  It's not working for me
>     on the names you've got; maybe because I'm on a Mac, and that was
>     compiled on Windows.  Perhaps someone else knows how to demangle those
>     names?
> 
>     Duncan Murdoch
> 
>     ______________________________________________
>     R-package-devel using r-project.org <mailto:R-package-devel using r-project.org>
>     mailing list
>     https://stat.ethz.ch/mailman/listinfo/r-package-devel
>     <https://stat.ethz.ch/mailman/listinfo/r-package-devel>
>