Missing return statements in file default_launcher.hpp cause an 'illegal instruction'

[bertwim]

In boost/process/v2/posix/default_launcher.hpp there is a template function invoke_on_success defined as

template<typename Launcher, typename Init>
inline auto invoke_on_success(Launcher & launcher, const filesystem::path &executable,
                              const char * const * (&cmd_line),
                              Init && init, derived && )
-> decltype(init.on_success(launcher, executable, cmd_line))
{
      init.on_success(launcher, executable, cmd_line);
}

This function is supposed to have a return value, but a return statement is missing. This causes (on my installation, using recent versions of both g++ and clang++) an 'illegal instruction' followed by an inevitable crash.

Adding a return statement solved the problem:

template<typename Launcher, typename Init>
inline auto invoke_on_success(Launcher & launcher, const filesystem::path &executable,
                              const char * const * (&cmd_line),
                              Init && init, derived && )
-> decltype(init.on_success(launcher, executable, cmd_line))
{
      return init.on_success(launcher, executable, cmd_line);
}

The same problem occurs (in the same file default_launcher.hpp) with the routines
invoke_on_error, invoke_on_fork_error and invoke_on_exec_error

[klemens-morgenstern]

on_success is supposed to return void - are you using a custom initializer?

[bertwim]

Yes, that is what I do.
In fact, I was familiarizing myself with the boost.process api, using the code I found in How to use v2 initializers #466 by mark-5345. First I thought that I was still doing something wrong, and I reacted to that post.

But now it seems that I've struck a bug in the code. Either the functions mentioned should be declared void and return nothing, or they do return something, but then there needs to be a return.

In my original post, I made a small mistake indicating the culprint header file: it should be
boost/process/v2/posix/default_launcher.hpp. I had left out the posix part. I edited the original posting accordingly.

[klemens-morgenstern]

Well, I am declaring a return type for SFINAE purposes, but you're right that SIGILL shouldn't happen - if anything this should be caught at compile time.

[bertwim]

Yes, it should be caught at compile time. But primarily the code should be correct, of course.
From your responses, I take it that void is the correct return type, although with adding a return statement as indicated before seems to work ok.
May I assume that the correct fix will be applied for a next release?