# hints/linux.sh # Original version by rsanders # Additional support by Kenneth Albanowski # # ELF support by H.J. Lu # Additional info from Nigel Head # and Kenneth Albanowski # # Consolidated by Andy Dougherty # # Updated Thu Feb 8 11:56:10 EST 1996 # Updated Thu May 30 10:50:22 EDT 1996 by # Updated Fri Jun 21 11:07:54 EDT 1996 # NDBM support for ELF renabled by # No version of Linux supports setuid scripts. d_suidsafe='undef' # Debian and Red Hat, and perhaps other vendors, provide both runtime and # development packages for some libraries. The runtime packages contain shared # libraries with version information in their names (e.g., libgdbm.so.1.7.3); # the development packages supplement this with versionless shared libraries # (e.g., libgdbm.so). # # If you want to link against such a library, you must install the development # version of the package. # # These packages use a -dev naming convention in both Debian and Red Hat: # libgdbmg1 (non-development version of GNU libc 2-linked GDBM library) # libgdbmg1-dev (development version of GNU libc 2-linked GDBM library) # So make sure that for any libraries you wish to link Perl with under # Debian or Red Hat you have the -dev packages installed. # # Some operating systems (e.g., Solaris 2.6) will link to a versioned shared # library implicitly. For example, on Solaris, `ld foo.o -lgdbm' will find an # appropriate version of libgdbm, if one is available; Linux, however, doesn't # do the implicit mapping. ignore_versioned_solibs='y' # BSD compatibility library no longer needed # 'kaffe' has a /usr/lib/libnet.so which is not at all relevant for perl. # bind causes issues with several reentrant functions set `echo X "$libswanted "| sed -e 's/ bsd / /' -e 's/ net / /' -e 's/ bind / /'` shift libswanted="$*" # If you have glibc, then report the version for ./myconfig bug reporting. # (Configure doesn't need to know the specific version since it just uses # gcc to load the library for all tests.) # We don't use __GLIBC__ and __GLIBC_MINOR__ because they # are insufficiently precise to distinguish things like # libc-2.0.6 and libc-2.0.7. if test -L /lib/libc.so.6; then libc=`ls -l /lib/libc.so.6 | awk '{print $NF}'` libc=/lib/$libc fi # Configure may fail to find lstat() since it's a static/inline # function in . d_lstat=define # malloc wrap works case "$usemallocwrap" in '') usemallocwrap='define' ;; esac # The system malloc() is about as fast and as frugal as perl's. # Since the system malloc() has been the default since at least # 5.001, we might as well leave it that way. --AD 10 Jan 2002 case "$usemymalloc" in '') usemymalloc='n' ;; esac # Check if we're about to use Intel's ICC compiler case "`${cc:-cc} -V 2>&1`" in *"Intel(R) C++ Compiler"*) # This is needed for Configure's prototype checks to work correctly ccflags="-we147 $ccflags" # If we're using ICC, we usually want the best performance case "$optimize" in '') optimize='-O3' ;; esac ;; esac case "$optimize" in # use -O2 by default ; -O3 doesn't seem to bring significant benefits with gcc '') optimize='-O2' case "`uname -m`" in ppc*) # on ppc, it seems that gcc (at least gcc 3.3.2) isn't happy # with -O2 ; so downgrade to -O1. optimize='-O1' ;; esac ;; esac # Are we using ELF? Thanks to Kenneth Albanowski # for this test. cat >try.c <<'EOM' /* Test for whether ELF binaries are produced */ #include #include main() { char buffer[4]; int i=open("a.out",O_RDONLY); if(i==-1) exit(1); /* fail */ if(read(i,&buffer[0],4)<4) exit(1); /* fail */ if(buffer[0] != 127 || buffer[1] != 'E' || buffer[2] != 'L' || buffer[3] != 'F') exit(1); /* fail */ exit(0); /* succeed (yes, it's ELF) */ } EOM if ${cc:-gcc} try.c >/dev/null 2>&1 && $run ./a.out; then cat <<'EOM' >&4 You appear to have ELF support. I'll try to use it for dynamic loading. If dynamic loading doesn't work, read hints/linux.sh for further information. EOM else cat <<'EOM' >&4 You don't have an ELF gcc. I will use dld if possible. If you are using a version of DLD earlier than 3.2.6, or don't have it at all, you should probably upgrade. If you are forced to use 3.2.4, you should uncomment a couple of lines in hints/linux.sh and restart Configure so that shared libraries will be disallowed. EOM lddlflags="-r $lddlflags" # These empty values are so that Configure doesn't put in the # Linux ELF values. ccdlflags=' ' cccdlflags=' ' ccflags="-DOVR_DBL_DIG=14 $ccflags" so='sa' dlext='o' nm_so_opt=' ' ## If you are using DLD 3.2.4 which does not support shared libs, ## uncomment the next two lines: #ldflags="-static" #so='none' # In addition, on some systems there is a problem with perl and NDBM # which causes AnyDBM and NDBM_File to lock up. This is evidenced # in the tests as AnyDBM just freezing. Apparently, this only # happens on a.out systems, so we disable NDBM for all a.out linux # systems. If someone can suggest a more robust test # that would be appreciated. # # More info: # Date: Wed, 7 Feb 1996 03:21:04 +0900 # From: Jeffrey Friedl # # I tried compiling with DBM support and sure enough things locked up # just as advertised. Checking into it, I found that the lockup was # during the call to dbm_open. Not *in* dbm_open -- but between the call # to and the jump into. # # To make a long story short, making sure that the *.a and *.sa pairs of # /usr/lib/lib{m,db,gdbm}.{a,sa} # were perfectly in sync took care of it. # # This will generate a harmless Whoa There! message case "$d_dbm_open" in '') cat <<'EOM' >&4 Disabling ndbm. This will generate a Whoa There message in Configure. Read hints/linux.sh for further information. EOM # You can override this with Configure -Dd_dbm_open d_dbm_open=undef ;; esac fi rm -f try.c a.out if /bin/sh -c exit; then echo '' echo 'You appear to have a working bash. Good.' else cat << 'EOM' >&4 *********************** Warning! ********************* It would appear you have a defective bash shell installed. This is likely to give you a failure of op/exec test #5 during the test phase of the build, Upgrading to a recent version (1.14.4 or later) should fix the problem. ****************************************************** EOM fi # On SPARClinux, # The following csh consistently coredumped in the test directory # "/home/mikedlr/perl5.003_94/t", though not most other directories. #Name : csh Distribution: Red Hat Linux (Rembrandt) #Version : 5.2.6 Vendor: Red Hat Software #Release : 3 Build Date: Fri May 24 19:42:14 1996 #Install date: Thu Jul 11 16:20:14 1996 Build Host: itchy.redhat.com #Group : Shells Source RPM: csh-5.2.6-3.src.rpm #Size : 184417 #Description : BSD c-shell # For this reason I suggest using the much bug-fixed tcsh for globbing # where available. # November 2001: That warning's pretty old now and probably not so # relevant, especially since perl now uses File::Glob for globbing. # We'll still look for tcsh, but tone down the warnings. # Andy Dougherty, Nov. 6, 2001 if $csh -c 'echo $version' >/dev/null 2>&1; then echo 'Your csh is really tcsh. Good.' else if xxx=`./UU/loc tcsh blurfl $pth`; $test -f "$xxx"; then echo "Found tcsh. I'll use it for globbing." # We can't change Configure's setting of $csh, due to the way # Configure handles $d_portable and commands found in $loclist. # We can set the value for CSH in config.h by setting full_csh. full_csh=$xxx elif [ -f "$csh" ]; then echo "Couldn't find tcsh. Csh-based globbing might be broken." fi fi # Shimpei Yamashita # Message-Id: <33EF1634.B36B6500@pobox.com> # # The DR2 of MkLinux (osname=linux,archname=ppc-linux) may need # special flags passed in order for dynamic loading to work. # instead of the recommended: # # ccdlflags='-rdynamic' # # it should be: # ccdlflags='-Wl,-E' # # So if your DR2 (DR3 came out summer 1998, consider upgrading) # has problems with dynamic loading, uncomment the # following three lines, make distclean, and re-Configure: #case "`uname -r | sed 's/^[0-9.-]*//'``arch`" in #'osfmach3ppc') ccdlflags='-Wl,-E' ;; #esac case "`uname -m`" in sparc*) case "$cccdlflags" in *-fpic*) cccdlflags="`echo $cccdlflags|sed 's/-fpic/-fPIC/'`" ;; *) cccdlflags="$cccdlflags -fPIC" ;; esac ;; esac # SuSE8.2 has /usr/lib/libndbm* which are ld scripts rather than # true libraries. The scripts cause binding against static # version of -lgdbm which is a bad idea. So if we have 'nm' # make sure it can read the file # NI-S 2003/08/07 if [ -r /usr/lib/libndbm.so -a -x /usr/bin/nm ] ; then if /usr/bin/nm /usr/lib/libndbm.so >/dev/null 2>&1 ; then echo 'Your shared -lndbm seems to be a real library.' else echo 'Your shared -lndbm is not a real library.' set `echo X "$libswanted "| sed -e 's/ ndbm / /'` shift libswanted="$*" fi fi # This script UU/usethreads.cbu will get 'called-back' by Configure # after it has prompted the user for whether to use threads. cat > UU/usethreads.cbu <<'EOCBU' case "$usethreads" in $define|true|[yY]*) ccflags="-D_REENTRANT -D_GNU_SOURCE -DTHREADS_HAVE_PIDS $ccflags" set `echo X "$libswanted "| sed -e 's/ c / pthread c /'` shift libswanted="$*" # Somehow at least in Debian 2.2 these manage to escape # the #define forest of and so that # the hasproto macro of Configure doesn't see these protos, # even with the -D_GNU_SOURCE. d_asctime_r_proto="$define" d_crypt_r_proto="$define" d_ctime_r_proto="$define" d_gmtime_r_proto="$define" d_localtime_r_proto="$define" d_random_r_proto="$define" ;; esac EOCBU cat > UU/uselargefiles.cbu <<'EOCBU' # This script UU/uselargefiles.cbu will get 'called-back' by Configure # after it has prompted the user for whether to use large files. case "$uselargefiles" in ''|$define|true|[yY]*) # Keep this in the left margin. ccflags_uselargefiles="-D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64" ccflags="$ccflags $ccflags_uselargefiles" ;; esac EOCBU