CmdStan broken now?

Developers
1

Help. @seantalts, @wds15: I can’t take the current state of develop and build any Stan program.

make build

results in pages of compiler errors like:

  "___cxa_throw", referenced from:
      stan::lang::parse(std::__1::basic_ostream<char, std::__1::char_traits<char> >*, std::__1::basic_istream<char, std::__1::char_traits<char> >&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, stan::io::program_reader const&, stan::lang::program&, bool) in stanc.o
      std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > >::reserve(unsigned long) in stanc.o
      stanc_helper(int, char const**, std::__1::basic_ostream<char, std::__1::char_traits<char> >*, std::__1::basic_ostream<char, std::__1::char_traits<char> >*) in stanc.o
      stan::io::program_reader::read(std::__1::basic_istream<char, std::__1::char_traits<char> >&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > const&, int&, bool, std::__1::set<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::less<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > >&) in stanc.o
      std::__1::basic_filebuf<char, std::__1::char_traits<char> >::seekoff(long long, std::__1::ios_base::seekdir, unsigned int) in stanc.o
      std::__1::basic_filebuf<char, std::__1::char_traits<char> >::sync() in stanc.o
      std::__1::basic_filebuf<char, std::__1::char_traits<char> >::underflow() in stanc.o
      ...
  "___gxx_personality_v0", referenced from:
      std::__1::vector<stan::lang::expression, std::__1::allocator<stan::lang::expression> >::~vector() in stanc.o
      stan::lang::fun_scalar_type(stan::lang::function_decl_def const&, bool) in stanc.o
      stan::lang::generate_indexed_expr_user(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, std::__1::vector<stan::lang::expression, std::__1::allocator<stan::lang::expression> >, std::__1::basic_ostream<char, std::__1::char_traits<char> >&) in stanc.o
      stan::lang::generate_expression(stan::lang::expression const&, bool, std::__1::basic_ostream<char, std::__1::char_traits<char> >&) in stanc.o
      stan::lang::generate_eigen_index_expression(stan::lang::expression const&, std::__1::basic_ostream<char, std::__1::char_traits<char> >&) in stanc.o
      stan::lang::generate_initializer(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, std::__1::vector<stan::lang::expression, std::__1::allocator<stan::lang::expression> > const&, stan::lang::expression const&, stan::lang::expression const&) in stanc.o
      stan::lang::generate_quoted_expression(stan::lang::expression const&, std::__1::basic_ostream<char, std::__1::char_traits<char> >&) in stanc.o
      ...
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make: *** [bin/stanc] Error 1

I’m trying to fix some bugs and this is making it tough. Is someone currently tasked to fixing CmdStan? Is there a pull request open to fix this?

2

Yeah, I noticed this the other day (and so did the CmdStan Performance Tests, but I apparently didn’t configure email appropriately for that yet). The PR in which I was asking for your help should fix this - https://github.com/stan-dev/cmdstan/pull/629. Maybe we can continue the conversation there?

3

Yup. Thanks. This is high priority.

4

I just approved the PR. I don’t think there’s much discussion to do on that PR.

If this gets pulled in, does CmdStan still work for Linux and Windows? (with the standard install)

5

It should work again, yes.

Our makefiles are in a not too good state… (we use the C compile commands to compile C++ code; not sure why).

6

This was a holdout from pre-1.0. I’ve tried to fix it a few times, but it’s always disruptive to users to change conventions. And… it’s also hard to fix so it’s correct on 3 platforms.

We should change it if we can. There’s an open PR on math that fixes it all at the math level. We can change CmdStan independently. I think we should since we’re now in this weird state where we have optional config for MPI and GPU.

@seantalts, will it be easier to tell when things are broken? It’s really, really disruptive when I can’t trust develop to work off of (for CmdStan, Stan, and Math). It’s fine if it’s for a short period where we’re orchestrating a few PRs across repos, but a note saying that we are would help? Btw, monorepo helps a bit, but doesn’t actually solve this problem.