Re: Wireshark won't compile on PPC64 boxes

[Gerasimos Dimitriadis <gedimitr () gmail com>]

Almost all of the string constants are used for initializing data
structures, so an extra problem I think is that the contents of a
strings array cannot be directly used for initializing e.g. the
hf_register_info array, since constants are needed for that. A
solution for that would be to initialize the hf_register_info array
with all string pointers set to NULL like this:
    { &hf_rrc_DL_DCCH_Message_PDU,
      { NULL, NULL,
        FT_NONE, BASE_NONE, NULL, 0,
        NULL, HFILL }},
....

and then before calling proto_register_field_array set the correct
pointers in runtime, from arrays where all the strings have been
collected:

for(i = 0; i < array_length(hf); i++)
{
   hf[i].hfinfo.name = bigNameArray[i];
   hf[i].hfinfo.abbrev = bigAbbrevArray[i];
   hf[i].hfinfo.blurb = bigBlurbArray[i];
}

Do you think that such a thing could solve the problem? I will try to
create a script for automating this process and see what comes from
it.

/Gerasimos

2010/1/27 didier <dgautheron () magic fr>:
> Hi,
> Le mercredi 27 janvier 2010 à 14:56 +0100, Radek Vokál a écrit :
> > On 01/27/2010 09:32 AM, didier wrote:
> > > Hi,
> > > Le mercredi 27 janvier 2010 à 00:54 -0700, Stephen Fisher a écrit :
> > > > On Tue, Jan 26, 2010 at 06:40:38PM +0100, Radek Vokál wrote:
> > > >
> > > > > The wireshark package cannot be build on ppc64 due to ppc64 ABI
> > > > > limitation and RRC dissector source file which contains too many
> > > > > string constants.
> > > >
> > > > That's not good :(.  I didn't realize until now that packet-rrc.c is
> > > > about 7MB of text (code) with over 150,000 lines!  Only the similar
> > > > dissector packet-nbap.c is even close at almost 70,000 lines at about
> > > > 3.5MB.  I assume that packet-nbap.c isn't causing compilation problems
> > > > on ppc64?  My first thought is to just pull packet-rrc.c from the
> > > > makefile, but if you're building packages for others, then leaving out a
> > > > dissector could be bad.  I wonder if it would make a difference if we
> > > > converted it to be a plug-in dissector that ships with Wireshark?
> > > >
> > > > Lacking that, we may be able to make the ASN.1 dissector generator spit
> > > > out a file that works on ppc64.  Sure seems like a lot of extra work to
> > > > support one platform, but then again, maybe it would be easier to manage
> > > > anyway having all of those strings handled differently.
> > > There's a lot of stuff like:
> > >   { "midambleAllocationMode", "rrc.midambleAllocationMode", ...
> > > Is it possible to compress it to the pseudo code:
> > >
> > > const foo = "rrc.midambleAllocationMode"
> > > {&foo +4, foo, ....
> > >
> > > ?
> > > It should be faster on all platforms.
> > >
> > > Don't know if it's enough.
> > >
> > > Didier
> Never mind it's about the number of toc entries, my idea doesn't work in
> this case.
>
> > The other solution would be to move all string literals to an const char
> > array and access them from that array.
> Yes but with xxx_sequence_t the not const char, may blow latter though
> this stuff is really big.
>
> Didier
>
>
> ___________________________________________________________________________
> Sent via:    Wireshark-dev mailing list <wireshark-dev () wireshark org>
> Archives:    http://www.wireshark.org/lists/wireshark-dev
> Unsubscribe: https://wireshark.org/mailman/options/wireshark-dev
>             mailto:wireshark-dev-request () wireshark org?subject=unsubscribe
___________________________________________________________________________
Sent via:    Wireshark-dev mailing list <wireshark-dev () wireshark org>
Archives:    http://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://wireshark.org/mailman/options/wireshark-dev
             mailto:wireshark-dev-request () wireshark org?subject=unsubscribe