GitLab

This reverts commit e5fd505f
which does not work as expected.

    Several libraries depend on boost_system lib according to
    otool, but lacked linking to $(BOOST_LIBS) in their Makefile.am.
    This has been corrected:

    src/lib/asiodns/Makefile.am
    src/lib/config/Makefile.am
    src/lib/dhcp/Makefile.am
    src/lib/dhcp_ddns/Makefile.am
    src/lib/dhcpsrv/Makefile.am
    src/lib/eval/Makefile.am
    src/lib/stats/Makefile.am

Optimization of boost:asio code is now disabled when building with
BOOST_ERROR_CODE_HEADER_ONLY under GCC versions 5.2.0 through 5.3.0.

src/lib/asiolink/asio_wrapper.h
    New file which wraps boost/asio/asio.hpp around compilation
    logic to suppress optimization under GNU 5.2.0 thru 5.3.0.

The remaining changes either removed the inclusion of <asio/asio.hpp> or
replaced it with the inclusion of <<asiolink/asio_wrapper.h>.  Inclusion
file order was also altered to better comply with our coding guidelines.

The reference could be removed because the TEST_DATA_DIR was unused in the
asiodns unit tests. In fact the folder it pointed to did not exist. Removal
of the call to addDataPath allowed for not including and compiling the
dns/tests/unittest_util.h and dns/tests/unittest_util.cc.

The major reason for this change was to remove the dependency on the files
located in the dns/tests because due to the bug in automake they could not
be included in build using an absolute path specified with $(top_srcdir).

Note that when the AUTOMAKE_OPTIONS = subdir-objects is specified, the
object files are created in the subdirectories. When the location of the
source file is specified using a variable like $(top_srcdir) the name of
the variable is treated as a name of the subdirectory, not a path to the
file located out of the tree. Hence, we may need to avoid including files
like this when subdir-objects is in force.

The unit test, IOFetchTest.TcpSendReceive65535, was failing
routinely on a NetBSD VM.  The VM was running so slowly that
this test was timing out.  Typically it takes less than
500ms. On this VM it was taking over 4.5s and then hitting
a safety valve timeout.  That timeout value was increased.