Mailing List Archive

>
> > It appears that E55 never made it into 0.5.1. This weeks B70 is
> > dependent on that patch being applied. Anyone opposed to me
> > pulling it into the for_Apache_0.5.1 directory?
>
> I was going to put E55 into 0.6 if there were enough votes.
> It's got a 1 from Cliff and a 0.5 from me so far (at least).
>
>

This needs to be in Apache. I have done some minimal testing
and will lobby strongly with my +1 to see it in 0.6.

> >I was going to put E55 into 0.6 if there were enough votes.
> >It's got a 1 from Cliff and a 0.5 from me so far (at least).
> It will only get a -0.5 from me.
> I can't test it.
> It doesn't allow you to provide IP addresses instead of hostnames
> (every other UNIX utility does)
> I don't see the need for the MULTI_ADDRESS_BIND #ifdef in the Makefile.
> (its confusing for the user to have to set _two_ defines for no readily
> apparent reason)

I agree with this David. Let's get it in the code base and fix
the blemishes in the coming week.

> It appears that E55 never made it into 0.5.1. This weeks B70 is
> dependent on that patch being applied. Anyone opposed to me
> pulling it into the for_Apache_0.5.1 directory?

I was going to put E55 into 0.6 if there were enough votes.
It's got a 1 from Cliff and a 0.5 from me so far (at least).

>
> >I agree with this David. Let's get it in the code base and fix
> >the blemishes in the coming week.
> Which is why I didn't veto it. If enough of the rest of you want it... (i.e.
> 3.5 people)

in that case...

Cliff +1
Robh +0.5
Brian +1
Randy +1
David -0.5
----
3

If nobody else -1's it. It'll be in 0.6

I also vote +1 for B70 which cleans up E55 and fixes some
missing ifdefs in a a header file.

Dave,

You can user the MULTI_ADDRESS_BIND without the VIRTUAL_HOST, two
different features, two defines. I can easily combine them if you
wish. File the non IP-address as a bug and it will be fixed, but lets
get the feature in.

Cliff

I jsut tested it, and indeed the code does allow you to use an IP address
instead of a name. Indeed if your name maps to more than one IP
address, you must use an interface specific name or a number. What
platform did you have problems on? (I tested this on BSDI)

Cliff

On Apr 14, 5:38pm, David Robinson wrote:
} Subject: Re: This weeks patches
} >I was going to put E55 into 0.6 if there were enough votes.
} >It's got a 1 from Cliff and a 0.5 from me so far (at least).
} It will only get a -0.5 from me.
} I can't test it.
} It doesn't allow you to provide IP addresses instead of hostnames
} (every other UNIX utility does)
} I don't see the need for the MULTI_ADDRESS_BIND #ifdef in the Makefile.
} (its confusing for the user to have to set _two_ defines for no readily
} apparent reason)
}
} David.
}-- End of excerpt from David Robinson

>I was going to put E55 into 0.6 if there were enough votes.
>It's got a 1 from Cliff and a 0.5 from me so far (at least).
It will only get a -0.5 from me.
I can't test it.
It doesn't allow you to provide IP addresses instead of hostnames
(every other UNIX utility does)
I don't see the need for the MULTI_ADDRESS_BIND #ifdef in the Makefile.
(its confusing for the user to have to set _two_ defines for no readily
apparent reason)

David.

>I agree with this David. Let's get it in the code base and fix
>the blemishes in the coming week.
Which is why I didn't veto it. If enough of the rest of you want it... (i.e.
3.5 people)

On Apr 15, 6:46pm, David Robinson wrote:
} You pass a string to gethostbyname [sic], so I presume it must be a hostname.
} Your example that I saw before used a hostname. Yet you claim that
} a numeric IP address should work? I don't think there is any reason to expect
} this, although I could imagine BSDI is being 'nice' and simply returning the
} IP address you gave it.
}
} This certainly would not work under OSF/1. (Though it does work under Solaris,
} to my surprise.)

This is a bsd thing, to my surprise it does not work under
OSF :-). Anyway, I changed the code to use inet_aton() now so that
is not an issue (B73 comming to a server near you after I test it).
I also only use the "VIRTUAL_HOST" define now, so your objections
should be all fixed.

The whole background for these bits of code was I got the basis
from the net long time ago and never really did bother to look
at it too much. I just added a couple features. The lack of
examination is why I did I did not hand it over to the apache group
earlier. Then it came up that we needed it fast, so I ent it in
as-is.

There is still one perfomance problem with the patches in that it
opens a log file every access (ugh) if you are accessing a virtual_host,
but I'll fix that in the patch for B73 also.

Cliff

Cliff wrote:
>You can user the MULTI_ADDRESS_BIND without the VIRTUAL_HOST, two
>different features, two defines. I can easily combine them if you
>wish...

But why have MULTI_ADDRESS_BIND #define at all? I don't see any gain in having
this compile-time option; all it does is make the package more confusing for
the user to build.

David.

>I jsut tested it, and indeed the code does allow you to use an IP address
>instead of a name. Indeed if your name maps to more than one IP
>address, you must use an interface specific name or a number. What
>platform did you have problems on? (I tested this on BSDI)

Ok, so I can't test it, and E55 doesn't seem to be available for Apache 0.5.1
(so I can't compile it), so all I can do is look at your code.

For the BindAddress:
if (!strcmp(w, "*")) {
bind_address.s_addr = htonl(INADDR_ANY);
} else {
hep = gethostbyname(w);
if (hep && hep->h_addrtype == AF_INET &&
hep->h_addr_list[0] && !hep->h_addr_list[1])
{
memcpy(&bind_address, hep->h_addr_list[0],
sizeof(struct in_addr));
} else {
fprintf(errors,"Syntax error on line %d of %s:\n",n,server_confname);
fprintf(errors,"BindAddress must be \"*\", or a numeric IP address,\n");
fprintf(errors,"or a name that maps to exactly one address.\n");
exit(1);
}
}
}
#endif /* APB_BIND_ADDRESS */

You pass a string to gethostbyname [sic], so I presume it must be a hostname.
Your example that I saw before used a hostname. Yet you claim that
a numeric IP address should work? I don't think there is any reason to expect
this, although I could imagine BSDI is being 'nice' and simply returning the
IP address you gave it.

This certainly would not work under OSF/1. (Though it does work under Solaris,
to my surprise.)


David.