Regression in posix::executor - fd not closed

[KiNgMaR]

Hi,

8d2bd87 (i.e. upgrading from boost 1.68 to boost 1.69) has introduced a regression where an fd for one of the child process pipes is not closed when there's nothing to read from it.

This leads to fd exhaustion after a while, e.g. with the following piece of code:

    bp::ipstream stdout;
    bp::child diffProcess(bp::search_path("diff"),
        "--unified=1",
        static_cast<std::string>(tmpFile1),
        static_cast<std::string>(tmpFile2),
        bp::std_out > stdout);

    diffProcess.wait();

This patch is a workaround for the single case that we ran into, but there are probably more cases where the fd is not closed either. Please check :)

diff --git a/include/boost/process/detail/posix/executor.hpp b/include/boost/process/detail/posix/executor.hpp
index 1390a58..8b86ed1 100644
--- a/include/boost/process/detail/posix/executor.hpp
+++ b/include/boost/process/detail/posix/executor.hpp
@@ -282,7 +282,10 @@ class executor
                 set_error(std::error_code(err, std::system_category()), "Error read pipe");
         }
         if (count == 0)
+        {
+            ::close(source);
             return  ;
+        }

         std::error_code ec(data[0], std::system_category());
         std::string msg(data[1], ' ');

Thank You & BR,
Ingmar

[reddwarf69]

I have not looked at the patch. But I can confirm I have found a file descriptor leaking issue using 1.69 and this patch fixes it.

[pfaffe]

+1
Fixes master as well!

[SimonEbner]

I experience the same bug and it looks like it's still not patched in boost 1.70.beta1 😢 Does anybody know if this will be addressed in time for 1.70 ?

[klemens-morgenstern]

It is solved in my current develop branch, I sadly did not make the 1.70 release. It can be fixed by inserting ::close(p[0]) after _read_error(p[0]); there: https://github.com/klemens-morgenstern/boost-process/blob/master/include/boost/process/detail/posix/executor.hpp#L432