We could not reference mixed or upper case variables properly prior to
this change, as they were always lowercased before references were
pulled.
The new "option.casesensitive" for templates allows folks to turn on
variable name reference case sensitivity on a template by template
basis.
The grammar now allows for leading underscores for variable names. This
addresses a problem stemming from the fact that the imjournal module
will create variables using the given name from systemd. All "trusted"
variables in systemd begin with a leading underscore.
Add the json_null{,_array} tests (with config) from commit 635a46d9bfb67d7110484709bc839cdd20a3b443 and
984c7e238b2eea7cc6db83e041fe257646b61636 to DIST target so that these files are included within the release
tarball.
The bug prevented tcpflood from reestablishing tls connections
after they were dropped. Also added an option -f to control the
random drop probability value (default = 0.95). Which means the
connection will be dropped 5% of the time.
Some tests used valgrind explicitely even when the configure
option forbid it's use in the testbench. This breaks, for example,
testbench runs with clang's address sanitizer.
This fails on buildbot env, but not on development machines. This
temporarily disables the test. We do this so that we can continue
to build and do other tests for mmnormalize inside the testbench.
This may be restored if the unusual construct in liblognorm gets fixed.
However, this construct is replaced in liblognorm v2 by something better
(repeat), so it is not sure yet if the legacy part of liblognorm will get
a fix.
The installed version of valgrind seems to have a bug which causes
memory leaks to be detected inside gnutls library.
After research I decided to disable memoryleak checking for
this specific test.
On some architectures, calculation of edLen may lead to a
negative value when the value returned by rand() is less
than extraDataLen away from RAND_MAX. The negative value
passed to memset() triggers a glibc error. Forcing the
cast of rand() to be unsigned avoid the return of a negative
value.
Closes: #506
Signed-off-by: Louis Bouchard <louis.bouchard@canonical.com>
The commands are implemented in a way which assumes they have global
side effects (e.g. exit terminating the current test). Not sourcing
the script breaks this assumption.
Man pages are not generated anymore when "--disable-generate-man-pages" is
configured. Also changed man-page rulepatterns makefile. The method
using $< refering to the prerequisite did not work proberly on FreeBSD.
closes https://github.com/rsyslog/rsyslog/issues/387
Those tests were enhanced with the addition of liblogging-stdlog
features. When compiled without it, most of the test will fail.
Skip those.
Signed-off-by: Louis Bouchard <louis.bouchard@canonical.com>
