posix: Using bp::std_out in conjunction with bp::limit_handles

[ceggers-arri]

Boost version: 1.72.0 Platform: Linux

I use bp::limit_handles in order to avoid leaking file descriptors to child processes. This works fine, but I was unsuccessful combining this with redirection of std_out to a string future.

 /* works fine */
std::vector<char> stdOut(4096);
auto child = bp::system("/bin/echo", "1",
                          bp::std_out > boost::asio::buffer(stdOut),
                          bp::limit_handles);

/* string is empty */
std::future<std::string> stdOut;
auto child = bp::system("/bin/echo", "1",
                          bp::std_out >stdOut,
                          bp::limit_handles);

In the second example, the future is valid, but the string is empty. When I remove bp::limit_handles, stdOut contains "1\n".

I recognized that there is a slight difference between async_out_buffer ...

template<int p1, int p2, typename Buffer>
struct async_out_buffer : ::boost::process::detail::posix::handler_base_ext,
                          ::boost::process::detail::posix::require_io_context,
                          ::boost::process::detail::uses_handles
{
...

... and async_out_future:

template<int p1, int p2, typename Type>
struct async_out_future : ::boost::process::detail::posix::handler_base_ext,
                          ::boost::process::detail::posix::require_io_context
{
...

While the former derives from bp::detail::use_handles (which seems to avoid closing this file descriptor), the latter doesn't do this.

Is this a by intention or a bug? Is there a way redirecting std_out to a string together with closing all other file handles?

Side question: How can I determine the number of output bytes for the first example (with the vector)?

regards Christian

[klemens-morgenstern]

This is a bug. thanks for reporting.

[garethsb]

@klemens-morgenstern also bug for async_pipe... https://github.com/boostorg/process/blob/master/include/boost/process/detail/posix/pipe_out.hpp is also missing uses_handles?

This commit https://github.com/boostorg/process/commit/f8c0dd4da58a2cf5db110d5a7e116adbe09f24a8 seems to have changes for pipe_out... but this wasn't taken in the merge commit https://github.com/boostorg/process/commit/fa81cecffc16fc557a3b99bcf8ecb6a7c9825937 😞

[garethsb]

I spotted a few other issues with limit_handles implementation...

This seems to be an important typo, on_success presumably intended:

https://github.com/boostorg/process/blob/2b39b56efb1cc7a3efe8d8815868ed0060111e7a/include/boost/process/detail/windows/handles.hpp#L165

This doc seems to be untrue (ah, already reported as https://github.com/boostorg/process/issues/212):

https://github.com/boostorg/process/blob/8222a57744a5df9beddcc7bddc168af66c85e3c7/include/boost/process/handles.hpp#L94-L99

This doc is confusing:

https://github.com/boostorg/process/blob/8222a57744a5df9beddcc7bddc168af66c85e3c7/include/boost/process/handles.hpp#L85

I assume that's meant to be "sets all handles to be inherited only explicitly"?

This comment has a typo, get_used_handles presumably intended:

https://github.com/boostorg/process/blob/2b39b56efb1cc7a3efe8d8815868ed0060111e7a/include/boost/process/detail/used_handles.hpp#L25