-
dearblue authored
The purpose is two-fold: 1. to be able to specify a pointer directly when user data is used When using `mrb_protect()`, it is necessary to allocate objects by `mrb_obj_cptr()` function when using user data. Adding `mrb_protect_raw()` will make it simpler to reimplement `mrbgems/mruby-error`. 2. to correctly unwind callinfo when an exception is raised from a C function defined as a method (the main topic) If a method call is made directly under `mrb_protect()` and a C function is called, control is returned from `mrb_protect()` if an exception occurs there. In this case, callinfo is not restored, so it is out of sync. Moreover, returning to mruby VM (`mrb_vm_exec()` function) in this state will indicate `ci->pc` of C function which is equal to `NULL`, and subsequent `JUMP` will cause `SIGSEGV`. Following is an example that actually causes `SIGSEGV`: - `crash.c` ```c #include <mruby.h> #include <mruby/compile.h> #include <mruby/error.h> static mrb_value level1_body(mrb_state *mrb, mrb_value self) { return mrb_funcall(mrb, self, "level2", 0); } static mrb_value level1(mrb_state *mrb, mrb_value self) { return mrb_protect(mrb, level1_body, self, NULL); } static mrb_value level2(mrb_state *mrb, mrb_value self) { mrb_raise(mrb, E_RUNTIME_ERROR, "error!"); return mrb_nil_value(); } int main(int argc, char *argv[]) { mrb_state *mrb = mrb_open(); mrb_define_method(mrb, mrb->object_class, "level1", level1, MRB_ARGS_NONE()); mrb_define_method(mrb, mrb->object_class, "level2", level2, MRB_ARGS_NONE()); mrb_p(mrb, mrb_load_string(mrb, "p level1")); mrb_close(mrb); return 0; } ``` - compile & run ```console % `bin/mruby-config --cc --cflags --ldflags` crash.c `bin/mruby-config --libs` % ./a.out zsh: segmentation fault (core dumped) ./a.out ``` After applying this patch, it will print exception object and exit normally. The `mrb_protect()`, `mrb_ensure()` and `mrb_rescue_exceptions()` in `mrbgems/mruby-error` have been rewritten using `mrb_protect_raw()`.
891e8522