> On Jun 7, 2017, at 6:15 PM, dragon512 <notificati...@github.com> wrote: > > @jpeach I don't know about the -D options. I did not test to see if -D works > or not ( I would assumed you guy had tested it already).
This patch enables the -D option in the generic option parsing, so I assumed that you needed it :) > The code removed here is preventing a custom directory for sysconfdir() to be > used. (ie the test would fail as trafficserver would exit with message > assuming we are using some -D option. This was added by Sudheer for Yahoo for the "-C config_verify” option. AFAIK no-one except Yahoo uses this, and there are no tests. > Removing this allowed some testing work to run correctly. OK so it sounds like it makes something work for you? In general I’m not that confident that setting the config directory is fully working. It worked enough for what Yahoo needed at the time, but it is hard to say more than that. > At the moment we are still making tweaks to get AuTest to work with layouts > that have absolute paths ( ie they ignore TS_ROOT) so they can work correct > with the testing system. A number of companies have complied about this > issue. This is one step in few PR yet to come to correct this issue. At the > moment an intern is looking in to traffic_ctrl and what we need to fix to get > this to use the correct paths. The only case that I’m aware of is that there is no convenient way to tell traffic_ctl which socket to use if you have a custom runtie layout. That doesn’t seem hard to fix, though deciding on the right way to specify the path deserves some though. > I can see about getting some "test" to validate the -D option works, and if > we all agree this is a supported option, see about documenting it AFAICT this option does nothing, so I don’t understand what started working when you removed it.
