On Tue, 3 Oct 2017, Andrew Cagney wrote:

I still don't know the correct way to set up a basic configuration (technically 
I know the wrong way to set up a basic configuration - by driving whack 
directly). 
What should require just a few lines gets lost in multiple levels of 
indirection, each adding stuff I suspect isn't needed.

Which is a good reason for turning the test cases into better
documentation by writing things out.

PS: slow-retransmits is interesting for another reason, there's an impair 
mechanism that almost does the right thing - it fails because it (correctly) 
logs that it
impaired the re-transmit

The impair is meant for those cases where we expect a failure to happen.
A connection that is supposed to not work. In real life deployments, we
must (eternally) keep trying, but for within the test case we want to
confirm and abort. That's the only time the impair for retransmits
should be used.

Well mostly, I also use it when I know there is a bug causing a
failure, and while working on it I add it so the test failure doesn't
take many minutes. But once the bug is fixed, the impair should no
longer be in place.

slow-retransmits is a "hack" to prevent false positives on "harmless"
retransmits that cause output changes. Ideally, we would fix these
without changing the configuration. But we would have to be sure
to not filter these out for those tests where it matters. This is a
hard problem. It is not only the retransmit output of another line
for the re-send. It also changes all the packet counters, so the
output change bleeds into everything.

Paul
_______________________________________________
Swan-dev mailing list
[email protected]
https://lists.libreswan.org/mailman/listinfo/swan-dev

Reply via email to