# Hints for the Power MAX OS operating system (formerly PowerUX - hence the # name) running on Concurrent (formerly Harris) NightHawk machines. Written # by Tom.Horsley@ccur.com # # This hint uses dynamic linking and the new Concurrent C compiler (based # on the Edison front end). This hint file was produced for a build of the # 5.7.3 development release of perl running on a PowerMAX_OS 5.1SR2 system # (but it should work on any Power MAX release using the newer "ec" (versus # "cc") compiler, and hopefully will also work for the upcoming 5.8 # development release of perl). # First find out where the root of the source tree is located. SRCROOT="" if [ -f ./INSTALL ] then SRCROOT="." else if [ -f ../INSTALL ] then SRCROOT=".." fi fi if [ -z "$SRCROOT" ] then echo "powerux hint file cannot locate root perl source!" 1>&2 exit 2 fi # We DO NOT want -lmalloc or -lPW, we DO need -lgen to follow -lnsl, so # fixup libswanted to reflect that desire (also need -lresolv if you want # DNS name lookup to work, which seems desirable :-). # libswanted=`echo ' '$libswanted' ' | sed -e 's/ malloc / /' -e 's/ PW / /' -e 's/ nsl / nsl gen resolv /'` # We DO NOT want /usr/ucblib in glibpth # glibpth=`echo ' '$glibpth' ' | sed -e 's@ /usr/ucblib @ @'` # Yes, csh exists, but doesn't work worth beans, if perl tries to use it, # the glob test fails, so just pretend it isn't there... # d_csh='undef' # Need to use Concurrent ec for most of these options to be meaningful (if you # want to get this to work with gcc, you're on your own :-). Passing # -Bexport to the linker when linking perl is important because it leaves # the interpreter internal symbols visible to the shared libs that will be # loaded on demand (and will try to reference those symbols). The -usys_nerr # drags in some stuff from libc that perl proper doesn't reference but # some dynamically linked extension will need to be in the static part # of perl (there are probably more of these that might be useful, but # for the extensions I build, this turned out to be enough). The -uldexp # makes sure the custom ldexp.o I add to archobjs actually gets pulled # into perl from libperl.a. The -unanosleep makes Timer::HiRes happy. # cc='/usr/ccs/bin/ec' cccdlflags='-Zpic' ccdlflags='-Zlink=dynamic -Wl,-usys_nerr -Wl,-uldexp -Wl,-unanosleep -Wl,-Bexport' lddlflags='-Zlink=so' # Sigh... Various versions of Power MAX went out with a broken ldexp runtime # routine in libc (it is fixed for sure in the upcoming SR4 release, but # that hasn't made it out the door yet). Since libc is linked dynamically, # and the perl you build might try to run on one of the broken systems, we # need to statically link a corrected copy of ldexp.o into perl. What the # following code does is determine if the ldexp.o on the current system # works right. If it does, it simply extracts the ldexp.o from the system C # library and uses that .o file. If the system .o is broken, the btoa # encoded copy of a correct ldexp.o file included in this hint file is used # (what a pain...) # if [ ! -f $SRCROOT/ldexp.o ] then echo Finding a correct copy of ldexp.o to link with... 1>&2 cat > $SRCROOT/UU/ldexptest.c <<'EOF' #include #include #include int main(int argc, char ** argv) { double result = pow(2.0, 38.0); char buf[100]; sprintf(buf, "%g", result); if (strncmp(buf, "inf", 3) == 0) { exit(2); } return 0; } EOF GOODLDEXP="no" $cc -v -Zlink=static -o $SRCROOT/UU/ldexptest $SRCROOT/UU/ldexptest.c -lm > $SRCROOT/UU/ldexptest.lo 2>&1 if [ $? -eq 0 ] then $SRCROOT/UU/ldexptest if [ $? -eq 0 ] then LDEXPLIB=`fgrep libc.a $SRCROOT/UU/ldexptest.lo | tail -1 | sed -e 's@^[^/]*@@'` if [ -s "$LDEXPLIB" ] then if [ -f "$LDEXPLIB" ] then GOODLDEXP="yes" fi fi fi fi if [ "$GOODLDEXP" = "yes" ] then echo Congratulations! The ldexp.o on this system looks good! 1>&2 echo Using ldexp.o from $LDEXPLIB 1>&2 ( cd $SRCROOT ; ar x $LDEXPLIB ldexp.o ) else echo Sorry, the ldexp.o on this system is busted. 1>&2 echo Using the ldexp.o from the powerux hint file 1>&2 atob > $SRCROOT/ldexp.o << 'EOF' xbtoa Begin Imm%#!7Ecb`!!%rA)G]Wp/%FC\ s(@fS,lAR]dp?YjFoAH3u00JG4;0JEJZF*VVE@:B4QA7^")/n4k]/hUsNAU&0$@rH4'?Zg7#FC/KgB 5)5`!%om?A7^")?Yj7aG]7#$DI``"/o5'0G]7#+A7^")?N:'+5\stBG]7#+Bl7KhF*(i2F9"RBA7^" )?YjFoARB"dA,nl2A7^")?YjFoARAnXB5)5`5\stBG]7#/Ec5c4B6@cmASu#Y5\stBG]7#/Ec5c4B6 @cm@V'1dD?'ZQA7^")!+0)TBQ@HkEcQ&9!+p7_G]3XiCh[?cG%G]8Bl@kh?XIJhB4YFn@;GorEb0&q /p(ZLF9!q6ASbd-FC\s(@fS-%ASbd-A7]4mB4#IhDIieJz3$J<@IAd4EOoYQ5HuL$L3Pb]og;*c.rk Jf$0+\*`g>N$VfHC6nOeDcBJaNL.[$5ct)LFr#l&\JY<(2JcPk%3$A9`IAd7F:4NFJY<(2JcPk%3$A9`IAd7F:4NF4eMcT^b#Nd5a26gb/"'sJY<(2JcPk%3$A9`IA d7F:4NFJcPk%3$A9`:4NFJcPk%3$A9`:4NNrmh6n.M)S$6';3>OC8,cI;FEfb2E>FD1mE>OF[C2aT2BpO3T0qs/VX)J4[;@g3<0$[I1UWg5car>8,rVj35>MFJY<(2JcPk%3$A9`IAd7F:4NFI11W[4obQ_a\Vs;D1mE>OF[CBa^G0WJcPk%r@:\mr&.(kb/"'s3$A9`:4NFJY<(2JcPk%3$A9`IAd7F:4NV!V!A:>Q=d*zz"98E)zzzzzzzzzz!!!!\zz"9AJl!!!!dzz!rr<&!!!!ezz!!!!# !!!"(zz!rr<'!!!")!!!)]z!!!!#!!!";!!!!Mz!!!!#!!!"Jzz!rr<(!!!"Kzz!rr<)!!!"Lzz!!! !&!!!"^!!!"Dz!!!!&!!!"n!!!!%z!!!!&!!!#+!!!!;z!!!!&!!!#?!!!!+z!!!!&!!!#Uzz!rr<* !!!#Vz!!!)]&c_n5!!!#\zz&-)\1!!!#hzz&-)\1!!!#nzz&-)\1!!!$&zz&-)\1!!!"$!!!Q!!!-H!!!!A!!!&F!!!-G!!!!9!!!&J!!!-H!!!!9!!!'[!! !Q!!!*Ezzzzzzzzzzz!!!!"!!!!$zz!!!!U!!!$Yzz!!!!"z!!!!*!!!!"!!!!' z!!!%=!!!)]zz!!!!1z!!!!0!!!!"!!!!#z!!!.(!!!!Qzz!!!!)z!!!!8!!!!"!!!!#z!!!.X!!!! Ezz!!!!%z!!!!?!!!!"zz!!!/'!!!"Dzz!!!!%z!!!!KJ,fQLzz!!!0J!!!!Ezz!!!!%z!!!!T!!!! #zz!!!0n!!!$b!!!!"!!!!0!!!!%!!!!1!!!$1!!!!%zz!!!4Z!!!$B!!!!(!!!!#!!!!%!!!!-!!! $!!!!-!!!!(!!!!&!!!!%!! !!- xbtoa End N 2436 984 E ad S 1bf43 R a7867666 EOF fi ( cd $SRCROOT/UU ; rm -f ldexptest* ) fi if [ -f $SRCROOT/ldexp.o ] then archobjs='ldexp.o' fi # Configure sometime finds what it believes to be ndbm header files on the # system and imagines that we have the NDBM library, but we really don't. # There is something there that once resembled ndbm, but it is purely # for internal use in some tool and has been hacked beyond recognition # (or even function :-) # i_ndbm='undef' # I have no clue what perl thinks it wants for, but if you # include it in a program in PowerMAX without first including # the code don't compile (apparently some other operating system has # something completely different in its sys/mode.h) # i_sysmode='undef' # There was a bug in memcmp (which was fixed a while ago) which sometimes # fails to provide the correct compare status (it is data dependant). I # don't want to figure out if you are building with the correct version or # not, so just pretend there is no memcmp (since perl has its own handy # substitute). # d_memcmp='undef' # Due to problems with dynamic linking (which I also hope will be fixed soon) # you can't build a libperl.so, the core has to be in the static part of the # perl executable. # useshrplib='false' # PowerMAX OS has support for a few different kinds of filesystems. The # newer "xfs" filesystem does *not* report a reasonable value in the # 'nlinks' field of stat() info for directories (in fact, it is always 1). # Since xfs is the only filesystem which supports partitions bigger than # 2gig and you can't hardly buy a disk that small anymore, xfs is coming in # to greater and greater use, so we pretty much have no choice but to # abandon all hope that number of links will mean anything. # dont_use_nlink=define # Configure comes up with the wrong type for these for some reason. The # pointers shouldn't have const in them. (And it looks like I have to # provide netdb_hlen_type as well because when I predefine the others it # comes up empty :-). # netdb_host_type='char *' netdb_name_type='char *' netdb_hlen_type='int' # Misc other flags that might be able to change, but I know these work right. # d_suidsafe='define' d_isascii='define' d_mymalloc='undef' usemymalloc='n' ssizetype='ssize_t' usevfork='false'