[ocaml-5.1.0]: Bytecode -custom exes fail to link with cc a C++ compiler

[shayne-fletcher]

Since ocaml-5.1.0, bytecode -custom executables fail to link if given -cc a C++ compiler (e.g -cc "clang++" will fail but -cc "clang" works).

$ cat > hello_world.ml
let () = print_string "Hello world!\n"
^D
$ ocamlc.opt -o hello-world ./hello_world.ml -custom -cc "clang++"
clang-16: warning: treating 'c' input as 'c++' when in C++ mode, this behavior is deprecated [-Wdeprecated]
ld: Undefined symbols:
  _caml_builtin_cprim, referenced from:
      _caml_build_primitive_table in libcamlrun.a[14](dynlink.b.o)
      _caml_build_primitive_table_builtin in libcamlrun.a[14](dynlink.b.o)
  _caml_names_of_builtin_cprim, referenced from:
      _caml_build_primitive_table in libcamlrun.a[14](dynlink.b.o)
      _caml_main in libcamlrun.a[56](startup_byt.b.o)
clang-16: error: linker command failed with exit code 1 (use -v to see invocation)
File "./hello_world.ml", line 1:
Error: Error while building custom runtime system

This regression was introduce in this commit and the reason is as follows: -custom bytecode executables link with compiler generated C like

#if defined __cplusplus
extern "C" {
#endif
typedef long value;

typedef value (*caml_primitive)();

...
extern value caml_abs_float(void);
...

const caml_primitive caml_builtin_cprim[] = {caml_abs_float, ...};
const char* const caml_names_of_builtin_cprim = {"caml_abs_float", ...};

...

#if defined __cplusplus
}
#endif

Since caml_builtin_cprim and caml_names_of_builtin_cprim are const qualified, when compiled with a C++ compiler (>= C++11), they are given internal linkage ¹ (unlike C).

The fix is to arrange for the generated code to mark caml_builtin_cprim and caml_names_of_builtin_cprim as extern:

extern const caml_primitive caml_builtin_cprim[] = {caml_abs_float, ...};
extern const char* const caml_names_of_builtin_cprim = {"caml_abs_float", ...};

Footnotes

1. C++11: C.1.2 https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2012/n3337.pdf ↩

[xavierleroy]

Hello Shayne, nice to hear from you, and thanks for the report. The older I get, the less C and C++ make sense to me, but each in its own different way...

Do you really need to use a C++ compiler with ocamlc -custom ? Couldn't you just omit the -cc option, or give it a C compiler?

[shayne-fletcher]

hello xavier!

the context here is the buck2 hack, flow & pyre builds (among others) at meta.

Do you really need to use a C++ compiler with ocamlc -custom

it's a very complicated multi-language setup that we work with. swapping out cc is non-trivial to say the least. i've tried a lot of things to work-around with some success but no one fix that will work for all targets.

Couldn't you just omit the -cc option

this could be a possibility but would require some effort in flow to adapt to i think.

all things considered if we can get something like #12791 through that would be the ideal. an alternative is to revert 99a05d2