MATSUURA> gcc: nocombreloc: No such file or directory Yeah, this is a bug in PPC gcc, which doesn't understand the -z flag for some reason. Probably we should force use of ld directly on that platform. MATSUURA> By adding --pdump in configure option, I think we should default it that way, anyway. -- 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.

Malcolm> In this case I think that the real problem is that you're Malcolm> passing "-b i486-linuxaout" to the test. I don't think Malcolm> that there are many X86 cross-compilers found on PPC Malcolm> machines :-) You may be right, however the evidence is very strong that that isn't the problem (gcc is known not to handle -z, the error message mentions nocombreloc, it was executing cc1 not ld, and --pdump worked). It may be _a_ problem, and I'll take a look at the configure code, though. We should probably get rid of that -b flag. Stephen> Probably we should force use of ld directly on that Stephen> platform. "Probably." I really don't understand why we currently insist on using gcc as a linker instead of ld itself, as we go to some trouble to find out what ld is anyway. Until I understand more, I'm not going to do anything that doesn't fix a manifest bug. However, I've already applied one patch that _does_ fix a bug that manifests because gcc DTWT in some cases. Malcolm> I don't think that's necessary. I've has no problems Malcolm> building bleeding edge XEmacs with "configure Malcolm> --prefix=/usr/local/gcc3-world --with-gnome" on a Mac for Malcolm> months using gcc 2.95 and 3.0.4. And we don't use any -z flags except nocombreloc on Linux and ignore on Solaris. If we ever do use a -z flag on Linux/PPC, we'll get the same problem again. No? Simple test: configure with LDFLAGS='-z nosuchkeyword' in the environment. Betcha configure don't like that much. -- 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.

Kaoru> Gcc doesn't have -z option which is linker option as ld. Only on PPC. On other platforms (at least all other Linux available to the Debian XEmacs maintainer and Solaris) where ld supports -z, gcc does pass it through to ld, although this is not documented. Eg, a random test program I have lying around on Debian Linux x86: bash-2.05a$ gcc --help | fgrep -e -z # no documentation bash-2.05a$ gcc -z nocombreloc -o test-z test.c # but it works bash-2.05a$ ./test-z uc << 14 = 4177920. bash-2.05a$ Unfortunately, it's possible that even if gcc is available to use as the linker, the linker may get called directly. So using -Wl, or -Xlinker is risky too. :-( -- 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.

Malcolm> There's a macro (XE_PROTECT_LINKER_FLAGS) that protects Malcolm> other linker variables, but not ld_switch_site. I'll have to take a look at it. The question is, does that macro really know how to tell the difference between gcc and a real linker? Malcolm> A quick hack didn't fix the problem. Any suggestions? Unfortunately not. I'm not a configure wizard. -- 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.

Stephen> Unfortunately, it's possible that even if gcc is Stephen> available to use as the linker, the linker may get called Stephen> directly. So using -Wl, or -Xlinker is risky too. :-( Malcolm> A look through the configuration data makes me think that Malcolm> at worst this would only be a problem for Decstations and Malcolm> IRIX 5.1 boxes (and then only if GNU ld is in the path) Malcolm> because everything else uses some variation of $(CC) as Malcolm> the linker. Right. The situation I was thinking of is apparently only relevant to the special compiler driver used for DSOs (ellcc), and that is due to a patch I'm now thinking should be reverted[1], since not only does it go against your argument above, but it also goes against the design notes in aclocal.m4. Malcolm> This works for me here and a patch is forthcoming. Relying on "this works for me" is a great way for a maintainer to get himself burned at the stake. Try to avoid that argument (except as one data point, ie, proof that you've tested the patch) -- it colors my ability to evaluate the patch objectively. :-) Footnotes: [1] Not before I talk to you, Andrew, if you're reading this. -- 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.