Thomas> I have littel problem to start an self-compiled Thomas> xemacs. Without '--ldflags=-s -znocombreloc' Xemacs crashs on Thomas> start. Stephen> This is no problem, it's a feature. Of GNU ld. Stephen> ld now rearranges reloc sections of (ie, trashes) the dumped Stephen> xemacs by default. -znocombreloc is necessary for all Stephen> unexec'd applications (XEmacs and GNU Emacs are the only Stephen> ones I know of). You can also use --pdump. Stephen> I don't know what the -s does, but the -znocombreloc is Stephen> autoconfiscated. You should not see this behavior in Stephen> 21.4.6, nor in Stephen> Please send your Installation file and config.log output so Stephen> we can figure out what went wrong. The autoconf stuff doesn't seem to work on powerpc. It's compiler/binutils suite stuff still seems to be different than the went. Also included is a patch that fixes a problem with athena widget sensing. Not sure why this doesn't work by default (this is all on debian) 2002-01-07 James LewisMoss <dres(a)debian.org&gt; * configure.in (XE_COMPUTE_RUNPATH): Change nocombeloc arg from "-z nocombreloc" to "-Xlinker -znocombreloc" --- xemacs21-21.4.6.orig/configure.in +++ xemacs21-21.4.6/configure.in ＠＠ -348,7 +348,7 ＠＠ with_site_modules='yes' with_menubars='' with_scrollbars='' -with_widgets='' +with_widgets='athena' with_dialogs='' with_file_coding='' cpp='' cppflags='' libs='' ldflags='' ＠＠ -2526,10 +2526,10 ＠＠ dnl #### Should make this Solaris-friendly. dnl Link with -z nocombreloc for now. if test "$pdump" != "yes"; then - AC_MSG_CHECKING(for \"-z nocombreloc\" linker flag) + AC_MSG_CHECKING(for \"-Xlinker -znocombreloc\" linker flag) case "`ld --help 2>&1`" in *-z\ nocombreloc* ) AC_MSG_RESULT(yes) - XE_PREPEND(-z nocombreloc, ld_switch_site) ;; + XE_PREPEND(-Xlinker -znocombreloc, ld_switch_site) ;; *) AC_MSG_RESULT(no) ;; esac fi ＠＠ -4859,7 +4859,7 ＠＠ echo " Relocating allocator for buffers: $rel_alloc" echo " GNU version of malloc: ${GNU_MALLOC}${GNU_MALLOC_reason}" case "$ld_switch_site" in - *nocombreloc*) echo " Linking with \`-z nocombreloc'. + *nocombreloc*) echo " Linking with \`-Xlinker -znocombreloc'. - Consider configuring with --pdump." ;; esac -- ＠James LewisMoss <dres(a)debian.org&gt; | Blessed Be! ＠ http://jimdres.home.mindspring.com | Linux is kewl! ＠"Argue for your limitations and sure enough, they're yours." Bach

Stephen> VETO 21.4 21.5 The -z combreloc stuff needs to be completely Stephen> redone, with this much variation the simple approach Stephen> probably just doesn't cut it, and the athena sensing is Stephen> addressed by a separate patch by Jan Vroonhof. Dres> The autoconf stuff doesn't seem to work on powerpc. It's Dres> compiler/binutils suite stuff still seems to be different than Dres> the went. Stephen> Arrgh. Your -z combreloc patch is not going to work if ld Stephen> is called explicitly instead of gcc, right? Have you tested Stephen> this on other platforms (at least i386-linux)? Yes. As far as I know this works on debian linux (i386, powerpc, ia64, and probably a few others. I've gotten no bug reports (that I remember)). Jim -- ＠James LewisMoss <dres(a)debian.org&gt; | Blessed Be! ＠ http://jimdres.home.mindspring.com | Linux is kewl! ＠"Argue for your limitations and sure enough, they're yours." Bach

Jan> "Stephen J. Turnbull" <stephen(a)xemacs.org&gt; writes: Jan> We really should find another term for "Further discussion Jan> required". Jan> Note that James is the Debian XEmacs package maintainer. He's Jan> more likely to cross platform test than any of us. As far as I know the only platform that the original way doesn't work on is powerpc linux. And as far as I know this change works on *-linux. (all debian unstable. I can't say anything about the other distros.) Jan> That is a pity. On the other hand, in fact normally we almost Jan> always use gcc as the linker, and so the current version relies Jan> on the fact that some versions of gcc pass -z on to the Jan> linker. Which AFAICS is documented nowhere. I am tempted to Jan> agree with James that working on more platforms overrides Jan> cleanliness issues. Well it's ugly that the test consists of parsing the ld --help output to begin with. Could always try to compile with "gcc -z nocombreloc" then "gcc -Xlinker -znocombreloc". I think this might be more complicated though. Odd error return from powerpc ("no file nocombreloc" or thereabouts iirc). Jim -- ＠James LewisMoss <dres(a)debian.org&gt; | Blessed Be! ＠ http://jimdres.home.mindspring.com | Linux is kewl! ＠"Argue for your limitations and sure enough, they're yours." Bach

Dres> As far as I know the only platform that the original way Dres> doesn't work on is powerpc linux. And as far as I know this Dres> change works on *-linux. (all debian unstable. I can't say Dres> anything about the other distros.) Good. If it doesn't work, actually, I'd say that's a bug in gcc. I think that the real problem is that we trust gcc to be a drop-in replacement for ld, and it cannot be. We really need to replace `gcc $ld_switch_site' with `gcc -Xlinker "$ld_switch_site"' or thereabouts. Dres> Well it's ugly that the test consists of parsing the ld Dres> --help output to begin with. Hey, autoconf is just ugly. I did that because Sun ld also supports -z combreloc. I wanted to have half a prayer of detecting it, and any other lds that might pick it up. But it turns out that there are versions of GNU ld that don't report the combreloc facility even though they do it by default (aw, you guessed it, distributed by Red Hat). So we're going to have to do something more careful anyway. Dres> Could always try to compile with "gcc -z nocombreloc" then Dres> "gcc -Xlinker -znocombreloc". I think this might be more Dres> complicated though. I think that this would be solved by defining LD correctly, if that's possible. Dres> Odd error return from powerpc ("no file nocombreloc" or Dres> thereabouts iirc). You mean from gcc? That's not surprising if the PowerPC gcc is swallowing the '-z' for compatibility with something or other. -- Institute of Policy and Planning Sciences http://turnbull.sk.tsukuba.ac.jp University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN Don't ask how you can "do" free software business; ask what your business can "do for" free software.

Sorry for the multiple followups, but I wanted to redirect the review policy discussion to xemacs-review and put this in the wrong place at first. Jan> That is a pity. On the other hand, in fact normally we almost Jan> always use gcc as the linker, That part of configure has been changed because it is known to screw some people (partly the issue of linker flags); we now use the linker reported by gcc. 2001-05-17 Andrew Begel <abegel(a)eecs.berkeley.edu&gt; * aclocal.m4 (LTLD): Don't use "gcc", use the ld reported by gcc. Maybe this would be solved by some form of my suggestion about LD=gcc -Xlinker $ld_switch_site -- Institute of Policy and Planning Sciences http://turnbull.sk.tsukuba.ac.jp University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN Don't ask how you can "do" free software business; ask what your business can "do for" free software.