GitLab

Also fixed a small bug in the previous commit where $CPPFLAGS was not
being passed along with $CPP when checking the log4cplus version.

... also cleaned up the display of config.report some to avoid the
rather unsightly blank lines that were ejected quite frequently when a
give variable had empty contents.

... and don't generate isc/server_common/bind10_server.py from
isc/server_common/bind10_server.py.in. This fixes a builddir vs srcdir
path failure for the isc.server_common module.

The joker in the pack here is io_address.h, required by the DHCP
header files.  That refers to the headers-only version of Boost
ASIO included in the BIND 10 source kit, so the ASIO files need
to be copied as well.

Also fix configure argument name to match help documentation.

... as we only use it in an automake condition.

This has caused issues in the past (see other error messages for example),
and escaping with \ doesn't seem to help in those cases.

This reverts commit 2092528a.

It should not go to master yet. Reverting until it is through
review.

The switch doesn't have an effect yet.

Tell more details about what the problem is.

The bug is in offset pointer, not in the red-black tree.

After trying to persuade automake to disable optimisations selectively
on the library or object file and trying desperate ways like
https://lists.gnu.org/archive/html/automake/2006-09/msg00038.html,
it produced a different kind of segfault with that version of boost.
It's likely that part of boost was not matured in that version yet, so
require new boost for the shared memory support.

With old version of boost and some versions of gcc, the RBTree
implementation is not stable with optimisations turned on. Detect the
old boost if we use gcc (no idea how to trigger it with the whole
machinery of the crashing test, so testing against the version).

Fail in that case now, we'll do something more drastic in following
commits.

NameChangeRequests from src/bin/d2 into their own library,
libdhcp_ddns, in src/lib/dhcp_ddns.

Now awaiting the merging of #2980 (the hooks code) so that the
configuration validation code can check if the libraries are valid
before accepting the configuration.

Only OSX uses ".dylib" as a suffix for shared libraries.  Adjust
configure.ac so that the operating system can be better determined,
and ensure the hooks tests use the correct suffic for the test shared
libraries.

Absence of "libdl" on BSD systems required an addition to configure.ac
to check its presence.  Problems with converting "pointer to object"
to "pointer to void*" via reinterpret_cast on Solaris required an
extension to the Pointer_converter class.

This manages LibraryManagers created for the libraries loaded.

can be located properly. Unit test was failing during distcheck.

runUnload() still has to be added.

The hooks framework now has its own directory in "lib" instead of
living in "util".

it's basically empty yet.