On Tue, 8 Sep 2015, Andrew Cagney wrote:
these should be fixed if you add RETRANSMIT_INTERVAL_DEFAULT=10000 to
Makefile.inc.local and re-compile.
I'm puzzled.
These specific tests require their own custom build, or all our test
builds require this tweak? I prefer to test what we ship (which means
not requiring Makefile.inc.local tweaks).
It would be all tests, this is a compile time option.
I had the exact same argument as you, that I'd really want to test
what we ship. But the problem is that the changes it is causing in
output is quite unpredictable and hard to sanitize. For example:
- Things counting for seeing a payload "x times" get the wrong count
- Multiline differences while we see "500ms 1s 2s" retransmit values
- Sometimes the responder sees the retransmit of previous packet,
but sometimes it will see the current packet, leading to no proposal
chosen messages (which are ignored but change the output)
- tests that _do_ need to test retransmits cannot get whack
--noretransmits so these again become very unpredictable in output.
We could have put retransmit-timeout and retransmit-interval in
conn %default but it seemed foodgroups did not inherit those, so
this does not help OE tests.
So the compile time option was a compromise. If we see issues in
the wild that we dont see on our tests, we should revisit this
decision.
Paul
_______________________________________________
Swan-dev mailing list
Swan-dev@lists.libreswan.org
https://lists.libreswan.org/mailman/listinfo/swan-dev
