INN 2.4.0 dumps core on Solaris 9

[Nathan Coraor]

"Russ Allbery" said:
> 
> Nathan Coraor <nate at cse.psu.edu> writes:
> 
> >   Sun compiler, specifically, Sun ONE Studio 8.  However, I got the
> >   same results with gcc 3.3.  Looks like it actually may be an issue of
> >   an overview created with a non-largefile INN, but trying to be used
> >   by a largefile INN.
> 
> Oh!  Yes, that will break every time.  Need to find a way of detecting
> that properly....
> 

  Ok.  I removed everything except ~news/etc and installed it all
  again, recreated the cycbuffs, ran makedbz -i again...

  innd starts and runs, but as soon as my production server tries to
  initiate a feed, I log:

    innd-orig: news connected 37 streaming allowed
    innd-orig: news:37 NCmode "mode stream" received
    innfeed: ME source lost . Exiting
    innfeed: ME finishing at Thu Oct 23 16:45:30 2003

  At this point, innd dies.  A truss reveals:

    13072/1:        llseek(18, 1064960, SEEK_SET)                   = 1064960
    13072/1:        writev(18, 0x00638D10, 6)                       = 5283
    13072/1:            Incurred fault #6, FLTBOUNDS  %pc = 0x00065B2C
    13072/1:              siginfo: SIGSEGV SEGV_MAPERR addr=0x00000258
    13072/1:            Received signal #11, SIGSEGV [default]
    13072/1:              siginfo: SIGSEGV SEGV_MAPERR addr=0x00000258
    13079:  read(0, 0x0025A69C, 8192)                       = 0
    13079:  llseek(4, 0, SEEK_CUR)                          Err#29 ESPIPE
    13079:  close(4)                                        = 0

    (gdb) bt
    #0  0x00065b2c in ARTpost ()
    #1  0x0006f3b4 in NCpostit ()
    #2  0x000711a8 in NCproc ()
    #3  0x000719a4 in NCreader ()
    #4  0x0006c4c8 in CHANreadloop ()
    #5  0x0006eca0 in main ()

  Bummer!

  What might this be?

Thanks,
--nate