ports/70922: x11/x3270 breaks palm/pose

[Eric P. Scott]

The following reply was made to PR ports/70922; it has been noted by GNATS.

From: eps+pbug0408 at ana.com (Eric P. Scott)
To: freebsd-gnats-submit at FreeBSD.org
Cc:  
Subject: Re: ports/70922: x11/x3270 breaks palm/pose
Date: Fri, 27 Aug 2004 15:36:55 -0700 (PDT)

 ktrace says it fails trying to process
 /usr/X11R6/lib/X11/fonts/local/3270.pcf.gz.
 
 I rebuilt pose with debugging enabled, and the backtrace then
 looked like:
 
 (gdb) bt
 #0  0x0 in ?? ()
 #1  0x833b01f in fill_inbuf (eof_ok=0) at ./../../SrcShared/Gzip/util.c:105
 #2  0x83393e2 in inflate_block (e=0xbfbfe70c)
     at ./../../SrcShared/Gzip/inflate.c:882
 #3  0x8339506 in inflate () at ./../../SrcShared/Gzip/inflate.c:931
 #4  0x28814014 in ft_gzip_file_fill_output ()
    from /usr/local/lib/libfreetype.so.9
    [...]
 
 Aha!
 
 This probably wasn't identified earlier because:
 
 1) Few users have "strange" fonts installed, so the offending
 code never got executed.
 
 2) The developers presumably use systems with a linker/loader
 that resolves undefined symbols left-to-right.
 
 You'll see pose links with this order:
 	Miscellaneous.o libposergzip.a -lfltk
 
 Miscellaneous.o calls inflate(), which is satisfied by
 libposergzip.a.  libfltk depends on [libXft, which depends
 on libfontconfig, which depends on] libfreetype, which calls
 an inflate() that should come from libz--but instead binds to
 the incompatible inflate() already brought in from
 libposergzip.a.  That one doesn't expect to be called, and trips
 over a NULL pointer from an uninitialized global.
 
 One solution is to add "palm/pose/files/patch-gzip-conflict" to
 rename the conflicting functions.
 
 --- ../SrcShared/Gzip/gzip.h.orig	Fri Mar 29 05:11:06 2002
 +++ ../SrcShared/Gzip/gzip.h	Thu Aug 26 16:12:35 2004
 @@ -320,4 +320,6 @@
  //extern voidp xmalloc      OF((unsigned int size));	// POSER
  
  	/* in inflate.c */
 +#define inflate POSE_inflate
 +#define inflate_codes POSE_inflate_codes
  extern int inflate OF((void));
 
 As a rule, I don't suggest using CONFLICTS except for mutually-
 exclusive packages.  x3270 wasn't impacted; it just exposed the
 underlying issue.  (I wonder how many other ports are similarly
 affected by multiply-defined symbols?)
 
 					-=EPS=-