r263478 - /stable/10/sys/net/route.h breaks net-snmp

Gleb Smirnoff glebius at FreeBSD.org
Mon Apr 7 13:33:50 UTC 2014


On Mon, Apr 07, 2014 at 12:28:01PM +0200, Oliver Lehmann wrote:
O> > Strange. This is how it builds on my CURRENT:
O> >
O> > checking for struct rtentry.rt_dst... no
O> > checking for struct rtentry.rt_hash... no
O> > checking for struct rtentry.rt_next... no
O> > checking for struct rtentry.rt_refcnt... no
O> > checking for struct rtentry.rt_unit... no
O> > checking for struct rtentry.rt_use... no
O> > checking type of rtentry structure... unknown
O> > checking for struct rtentry... rtentry
O> >
O> > After that, it compiles fine.
O> >
O> > In your log it detects 'struct ortentry' and surprisingly prefers it.
O> 
O> Please check 10-STABLE.
O> - ortentry is visible
O> - rtentry is not visible
O> I guess this is the reason why. I didn't checked CURRENT, just grabbed
O> up the log via google (I have no CURRENT system atm). My issue is on
O> STABLE where you merged this to. I wonder why this API-breakage got
O> merged anyway?!

The ortentry wasn't part of any API or ABI. It was just defined. I am
very surprised that net-snmp picks it and utilizes it during compilation.
If it uses the structure anywhere, then it obtains invalid info from it.
So that is actually a bug on net-snmp side taking that structure.

I guess that in CURRENT, where ortentry is wiped entirely, and net-snmp
takes rtentry, we've got the bug accidentially fixed.

I'm now running universe build with r263203 merged. If it succeeds,
I will merge r263203 removing declaration of ortentry.

-- 
Totus tuus, Glebius.


More information about the freebsd-ports mailing list