Guy Harris wrote:
On Apr 28, 2010, at 1:32 PM, Jeff Morriss wrote:
guy@xxxxxxxxxxxxx wrote:
http://anonsvn.wireshark.org/viewvc/viewvc.cgi?view=rev&revision=32519
User: guy
Date: 2010/04/19 04:38 PM
Log:
If that should truly "never happen", use DISSECTOR_ASSERT_NOT_REACHED()
so it's more clearly marked as a dissector bug.
(It apparently *does* happen - see bug 4698.)
This has the randpkt test failing on the buildbot.
...which means that the RSVP dissector has, and had even before that checkin, a bug, in that something that, according to a comment in the code, "should never happen" can, in fact, happen with a bogus packet; this just makes the bug more obvious.
Should it really be backported to 1.2.8?
Clearly marking something that "should never happen" but does happen as a dissector bug in the dissection is better than just putting a blob of
Unknown session type
into the protocol tree, so, yes, I'd backport it.
Or should the randpkt test accept dissector bugs as OK (like the fuzz
testing)?
The fuzz testing accepts dissector bug reports as OK? That seems like an error to me.
So someone else thought too when they noticed it. So we turned it on
for a while but then we got overwhelmed with bugs. See, for example,
3885, 3879, 3887, 3881. So then it got turned off again.