GitLab

The message when the DHCPv6 database backend is started now includes
both the type and the name of the database.

Parameters are hard-coded for the moment.

This should fix compilation, where it complains:

rrttl_unittest.cc:79: warning: this decimal constant is unsigned only in ISO C90

Conflicts:
	doc/devel/mainpage.dox
	src/lib/dhcp/Makefile.am
	src/lib/dhcp/lease_mgr.cc
	src/lib/dhcp/lease_mgr.h
	src/lib/dhcp/tests/Makefile.am
	src/lib/dhcp/tests/lease_mgr_unittest.cc

A number of other files were modified to resolve incompatibilities
in the way lease managers were created, and to promote the dummy
"memfile" lease manager to a fully-fledged lease manager with
unit tests.

There was a gap when one thing could overflow to negative numbers.

Having duid.{h, cc} in libdhcp++ causes dependency between two libraries.
Due to this dependency multi-threaded build fails because libs are created
by different threads simultaneously and race condition occurs. Libraries
should be moved to separate folders and have separate makefiles. This
complies with BIND10 structure and will make this issue goe away.

Also, simplify the code little bit and make it run faster in the case of
TTL without any units.

For some reason, on Solaris this causes types "uintXX_t" to become
undefined.  The "using" statement was removed and an explicit
namespace prepended to declarations of shared_ptr and dynamic_pointer_cast.

The allocation engine is now pointed to by a shared pointer.  Previously
it was pointed toi by a raw pointer and not deleted in the destructor.
Also, use an automatic variable to hold the DHCP6 server in the
unit tests.

It doesn't need to check the number of digits. It seems the tests pass
now without problems as well. Also, it might be problematic with large
number of leading zeroes.

Also, simplify comment with description.

From http://bind10.isc.org/ticket/2468#comment:2 to help troubleshoot later.

Okayed via jabber.

a) Corrected some method headers
b) Added additional diagnostic output should the test fail to
   open the database.

... to ensure that it is always called when a method terminates.

and suggest checking output from msgq using -v.

previously it could result in exception (by the attempt of killing sockcreator
as a chuser'ed process) during initial setup failure, leading to unclean
shutdown.  the revised code prevents it from happening, and also makes the
code concise by unifying the common logic.

so, it now just prints some hints to stderr and exits.

it even suggests restarting as a root.

it now includes textual representation of errno, and, if it's EACCES
("permission denied") the messsage suggesting restarting as a super user.