On 12/01/11 23:31, Kinkie wrote:
On Wed, Jan 12, 2011 at 11:09 AM, Amos Jeffries<[email protected]>  wrote:
As Andrew Beverley found:
"
I also found a bug in the current configure.ac. You get the message
"SQUID_DEFINE_BOOL: unrecognized value for USE_LIBNETFILTERCONNTRACK:
'auto'" if you haven't explicitly set with-netfilter-conntrack. This
patch fixes that.
"

I think this is a bug in our implementation of SQUID_DEFINE_BOOL as matches
our tri-state logics.

With our default "everything on unless impossible" policy I believe our
'auto' should be accepted as a value for 'yes'.

Agreed? Opposed?

Hi Amos,
   what is the error-case? It's not clear to me.
What I find as a potential bug is the fact that
USE_LIBNETFILTERCONNTRACK has some wierd dependency on enable_zph_qos.

Blindly accepting "auto" for "yes" may cause issues, as "yes"
indicates that the required build infrastructure has been tested for
existence. So if anything, I'd suggest reviewing how the interaction
of the netfilter_conntrack build-options is handled, possibly
simplifying it (e.g. making zph_qos implicit in conntrack code).


Hm, okay. When writing out the use case in full to respond to that I convince myself tat you are right. I was (again) considering a --with in the same light as an --enable. Which was wrong. Applied the better fix for the bug.

As for the interactions. --enable-zph-qos is our current name for the overall QoS feature set. This conntrack is only one part of that set. IMO the dependency interaction here for them is correct, what wasn't was the "auto" getting past the conntrack sub-sub-dependency checks.

Amos
--
Please be using
  Current Stable Squid 2.7.STABLE9 or 3.1.10
  Beta testers wanted for 3.2.0.4

Reply via email to