bind 4.8.x and AIX 1.2 ?

[Jack F. Vogel]

In article <26b7929a.68e3 at petunia.CalPoly.EDU> booga at polyslo.CalPoly.EDU (Steve Jankowski [cut]) writes:

[...stuff about the nameserver being back-level deleted...]

>We have the same problem here and I've started a port of named from
>4.8.1 to AIX 1.2 on a PS/2.  The problem with AIX (especially the
>PS/2) is their (ahem) crappy C compiler.  named uses bit fields to pick
>apart request packets.  This is a less than portable solution, but
>it's ok since most C compilers honor the spirit of bit fields and
>pack the bits into the smallest space possible.  But, the AIX
>compiler just makes every bit field 32-bits and leaves it at that.
 
Yup, ran into this problem in the new port I mentioned in a previous
posting. Interestingly enough, the 370 Metaware compiler does
things correctly! However there isn't any option to the 386 compiler
to keep it from doing this, oh well :-{.

>If you look in arpa/nameser.h you'll notice that they've put in
>two unsigned chars and a bunch of macros to replace the bit fields.
>(this is in the structure that represents a request packet)
>This means lots of manual labor to change each reference to the
>structure.  I started the work, but stopped when I heard that
>we would be getting AIX 1.2 which reportedly has named from
>4.8.1.  But I just string'ed the binary and I'm not impressed;
>the dates in the file are all from 1986.
 
The new BIND is in what I believe is called the July preventative
update as I said previously, I don't know if you have to request
it or if you get it automatically. The easiest way to tell if it
is installed is to see if you have a new binary, /etc/named-xfer.
This is the asyncronous zone transfer facility, if its there you
have the new nameserver. Now a little bit on how the port was done...

In my first attempt at the code I went ahead as you described and
used the macros, and as you noticed this is a fair amount of work
and change to the code. Then a friend working in our Boston office
(Thanks Brad!) came up with a creative way around this problem and
this is what I actually implemented in the code that's shipping.
The header structure as shipped by Berkeley contains an unsigned
short id variable and then 2 unsigned chars carved up as the bitfields.
What I did was make one unsigned int as a 32-bit bitfield and just
declare the first 16 bits as the id. Doing this the PS/2 compiler
leaves things alone. You can take a look at this change after installing
the update. The important thing to observe if you use this structure
is that you must define BIT_ZERO_ON_LEFT for the 370 and BIT_ZERO_ON_RIGHT
for the PS/2.

Also, just as a tip for those out there who might want to build nslookup
or other related tools, be sure you use -DBSD=4.3 and -D_BSD in your
compiler defines in addition to those mentioned above, then in your
link command be sure and include -lbsd. With the new nameser.h and these
proper defines the BSD source should build and work without modification.

>If I actually go through with the port and get it working, I'll
>let people know.
 
Hope you don't waste too much time before you see this :-}. I suggest
you call your IBM representative and request the latest preventative
update for this fix. As I mentioned in a previous followup, I also
finished the sendmail 5.61 port and it is also in this update.

Disclaimer: I get paid to solve problems, not to speak for LCC or IBM.

-- 
Jack F. Vogel			jackv at locus.com
AIX370 Technical Support	       - or -
Locus Computing Corp.		jackv at turnkey.TCC.COM