Some signals are more real than others
[p5sagit/p5-mst-13.2.git] / hints / dec_osf.sh
CommitLineData
a0d0e21e 1# hints/dec_osf.sh
8e964347 2
b971f6e4 3# * If you want to debug perl or want to send a
8e964347 4# stack trace for inclusion into an bug report, call
5# Configure with the additional argument -Doptimize=-g2
6# or uncomment this assignment to "optimize":
7#
8#optimize=-g2
9#
8a019ce7 10# If you want both to optimise and debug with the DEC cc
11# you must have -g3, e.g. "-O4 -g3", and (re)run Configure.
12#
b971f6e4 13# * gcc can always have both -g and optimisation on.
8a019ce7 14#
b971f6e4 15# * debugging optimised code, no matter what compiler
8a019ce7 16# one is using, can be surprising and confusing because of
17# the optimisation tricks like code motion, code removal,
18# loop unrolling, and inlining. The source code and the
19# executable code simply do not agree any more while in
20# mid-execution, the optimiser only cares about the results.
21#
b971f6e4 22# * Configure will automatically add the often quoted
f2c69087 23# -DDEBUGGING for you if the -g is specified.
24#
b971f6e4 25# * There is even more optimisation available in the new
f2c69087 26# (GEM) DEC cc: -O5 and -fast. "man cc" will tell more about them.
27# The jury is still out whether either or neither help for Perl
28# and how much. Based on very quick testing, -fast boosts
29# raw data copy by about 5-15% (-fast brings in, among other
30# things, inlined, ahem, fast memcpy()), while on the other
31# hand searching things (index, m//, s///), seems to get slower.
32# Your mileage will vary.
8e964347 33#
b971f6e4 34# * The -std is needed because the following compiled
e60a08f8 35# without the -std and linked with -lm
36#
37# #include <math.h>
38# #include <stdio.h>
39# int main(){short x=10,y=sqrt(x);printf("%d\n",y);}
40#
41# will in Digital UNIX 3.* and 4.0b print 0 -- and in Digital
42# UNIX 4.0{,a} dump core: Floating point exception in the printf(),
43# the y has become a signaling NaN.
44#
b971f6e4 45# * Compilation warnings like:
46#
47# "Undefined the ANSI standard macro ..."
48#
49# can be ignored, at least while compiling the POSIX extension
50# and especially if using the sfio (the latter is not a standard
51# part of Perl, never mind if it says little to you).
52#
8e964347 53
e60a08f8 54# If using the DEC compiler we must find out the DEC compiler style:
55# the style changed between Digital UNIX (aka DEC OSF/1) 3 and
56# Digital UNIX 4. The old compiler was originally from Ultrix and
57# the MIPS company, the new compiler is originally from the VAX world
58# and it is called GEM. Many of the options we are going to use depend
59# on the compiler style.
60
a68015de 61cc=${cc:-cc}
62
6c7aff00 63case "`$cc -v 2>&1 | grep cc`" in
64*gcc*) isgcc=gcc ;;
65esac
66
b691c02f 67# do NOT, I repeat, *NOT* take away the leading tabs
68# Configure Black Magic (TM)
b971f6e4 69 # reset
e60a08f8 70 _DEC_cc_style=
6c7aff00 71case "$isgcc" in
a4349bea 72gcc) if [ "X$gccversion" = "X" ]; then
73 # Done too late in Configure if hinted
74 gccversion=`$cc --version | sed 's/.*(GCC) *//'`
75 fi
993a793c 76 set $gccversion
77 if test "$1" -lt 2 -o \( "$1" -eq 2 -a \( "$2" -lt 95 -o \( "$2" -eq 95 -a "$3" -lt 3 \) \) \); then
2cae8c0d 78 cat >&4 <<EOF
79
993a793c 80*** Your cc seems to be gcc and its version ($gccversion) seems to be
81*** less than 2.95.3. This is not a good idea since old versions of gcc
61f70568 82*** are known to produce buggy code when compiling Perl (and no doubt for
83*** other programs, too).
3f8b8817 84***
61f70568 85*** Therefore, I strongly suggest upgrading your gcc. (Why don't you use
86*** the vendor cc is also a good question. It comes with the operating
3f8b8817 87*** system and produces good code.)
2cae8c0d 88
89Cannot continue, aborting.
90
91EOF
92 exit 1
93 fi
3f8b8817 94 if test "$1" -eq 2 -a "$2" -eq 95 -a "$3" -le 2; then
95 cat >&4 <<EOF
96
97*** Note that as of gcc 2.95.2 (19991024) and Perl 5.6.0 (March 2000)
98*** if the said Perl is compiled with the said gcc the lib/sdbm test
61f70568 99*** may dump core (meaning that the SDBM_File extension is unusable).
100*** As this core dump never happens with the vendor cc, this is most
101*** probably a lingering bug in gcc. Therefore unless you have a better
102*** gcc installation you are still better off using the vendor cc.
3f8b8817 103
104Since you explicitly chose gcc, I assume that you know what are doing.
105
106EOF
107 fi
2cae8c0d 108 ;;
e60a08f8 109*) # compile something small: taint.c is fine for this.
4de7e3a2 110 ccversion=`cc -V | awk '/(Compaq|DEC) C/ {print $3}' | grep '^V'`
e60a08f8 111 # the main point is the '-v' flag of 'cc'.
52c7d5b6 112 case "`cc -v -I. -c taint.c -o taint$$.o 2>&1`" in
e60a08f8 113 */gemc_cc*) # we have the new DEC GEM CC
dfa3a3d3 114 _DEC_cc_style=new
8a019ce7 115 ;;
e60a08f8 116 *) # we have the old MIPS CC
dfa3a3d3 117 _DEC_cc_style=old
8a019ce7 118 ;;
e60a08f8 119 esac
120 # cleanup
52c7d5b6 121 rm -f taint$$.o
e60a08f8 122 ;;
123esac
124
b971f6e4 125# be nauseatingly ANSI
6c7aff00 126case "$isgcc" in
127gcc) ccflags="$ccflags -ansi"
b971f6e4 128 ;;
24801a4b 129*) ccflags="$ccflags -std1"
b971f6e4 130 ;;
131esac
132
133# for gcc the Configure knows about the -fpic:
134# position-independent code for dynamic loading
135
e60a08f8 136# we want optimisation
137
138case "$optimize" in
6c7aff00 139'') case "$isgcc" in
140 gcc) optimize='-O3' ;;
e60a08f8 141 *) case "$_DEC_cc_style" in
6c7aff00 142 new) optimize='-O4' ;;
e60a08f8 143 old) optimize='-O2 -Olimit 3200' ;;
144 esac
b971f6e4 145 ccflags="$ccflags -D_INTRINSICS"
e60a08f8 146 ;;
8e964347 147 esac
148 ;;
1fc4cb55 149esac
28757baa 150
532eb838 151## Optimization limits
152case "$isgcc" in
153gcc) # gcc 3.2.1 wants a lot of memory for -O3'ing toke.c
154cat >try.c <<EOF
155#include <sys/resource.h>
156
157int main ()
158{
159 struct rlimit rl;
160 int i = getrlimit (RLIMIT_DATA, &rl);
161 printf ("%d\n", rl.rlim_cur / (1024 * 1024));
162 } /* main */
163EOF
164$cc -o try $ccflags $ldflags try.c
165 maxdsiz=`./try`
166rm -f try try.c core
167if [ $maxdsiz -lt 256 ]; then
168 # less than 256 MB is probably not enough to optimize toke.c with gcc -O3
169 cat <<EOM >&4
170
171Your process datasize is limited to $maxdsiz MB, which is (sadly) not
172always enough to fully optimize some source code files of Perl,
173at least 256 MB seems to be necessary as of Perl 5.8.0. I'll try to
174use a lower optimization level for those parts. You could either try
175using your shell's ulimit/limit/limits command to raise your datasize
176(assuming the system-wide hard resource limits allow you to go higher),
177or if you can't go higher and if you are a sysadmin, and you *do* want
178the full optimization, you can tune the 'max_per_proc_data_size'
179kernel parameter: see man sysconfigtab, and man sys_attrs_proc.
180
181EOM
182toke_cflags='optimize=-O2'
183 fi
184;;
185esac
186
61ba5045 187# The patch 23787
188# http://public.activestate.com/cgi-bin/perlbrowse?patch=23787
0fae7d62 189# broke things for gcc (at least gcc 3.3) so that many of the pack()
190# checksum tests for formats L, j, J, especially when combined
191# with the < and > specifiers, started to fail if compiled with plain -O3.
61ba5045 192case "$isgcc" in
193gcc)
0fae7d62 194pp_pack_cflags='optimize="-O3 -fno-cse-skip-blocks"'
61ba5045 195;;
196esac
197
6c7aff00 198# we want dynamic fp rounding mode, and we want ieee exception semantics
199case "$isgcc" in
a59bce4b 200gcc) ;;
201*) case "$_DEC_cc_style" in
6c7aff00 202 new) ccflags="$ccflags -fprm d -ieee" ;;
203 esac
204 ;;
205esac
206
313489a2 207# Make glibpth agree with the compiler suite. Note that /shlib
208# is not here. That's on purpose. Even though that's where libc
209# really lives from V4.0 on, the linker (and /sbin/loader) won't
210# look there by default. The sharable /sbin utilities were all
211# built with "-Wl,-rpath,/shlib" to get around that. This makes
212# no attempt to figure out the additional location(s) searched by
213# gcc, since not all versions of gcc are easily coerced into
214# revealing that information.
b0362887 215glibpth="/usr/shlib /usr/ccs/lib /usr/lib/cmplrs/cc"
216glibpth="$glibpth /usr/lib /usr/local/lib /var/shlib"
313489a2 217
8e964347 218# dlopen() is in libc
219libswanted="`echo $libswanted | sed -e 's/ dl / /'`"
220
b971f6e4 221# libPW contains nothing useful for perl
8a019ce7 222libswanted="`echo $libswanted | sed -e 's/ PW / /'`"
223
723e14d4 224# libnet contains nothing useful for perl here, and doesn't work
225libswanted="`echo $libswanted | sed -e 's/ net / /'`"
226
b971f6e4 227# libbsd contains nothing used by perl that is not already in libc
8a019ce7 228libswanted="`echo $libswanted | sed -e 's/ bsd / /'`"
229
b971f6e4 230# libc need not be separately listed
8a019ce7 231libswanted="`echo $libswanted | sed -e 's/ c / /'`"
232
b971f6e4 233# ndbm is already in libc
234libswanted="`echo $libswanted | sed -e 's/ ndbm / /'`"
8a019ce7 235
236# the basic lddlflags used always
237lddlflags='-shared -expect_unresolved "*"'
238
144df5e1 239# Intentional leading tab.
8cfab5ff 240 myosvers="`/usr/sbin/sizer -v 2>/dev/null || uname -r`"
144df5e1 241
b971f6e4 242# Fancy compiler suites use optimising linker as well as compiler.
243# <spider@Orb.Nashua.NH.US>
b8f0c030 244case "`uname -r`" in
b971f6e4 245*[123].*) # old loader
246 lddlflags="$lddlflags -O3"
247 ;;
a07564da 248*) if $test "X$optimize" = "X$undef"; then
249 lddlflags="$lddlflags -msym"
250 else
144df5e1 251 case "$myosvers" in
e3159d07 252 *4.0D*)
253 # QAR 56761: -O4 + .so may produce broken code,
254 # fixed in 4.0E or better.
255 ;;
256 *)
257 lddlflags="$lddlflags $optimize"
258 ;;
259 esac
260 # -msym: If using a sufficiently recent /sbin/loader,
261 # keep the module symbols with the modules.
24801a4b 262 lddlflags="$lddlflags -msym -std1"
a07564da 263 fi
b971f6e4 264 ;;
265esac
266# Yes, the above loses if gcc does not use the system linker.
267# If that happens, let me know about it. <jhi@iki.fi>
268
8cfab5ff 269# Because there is no other handy way to recognize 3.X.
270case "`uname -r`" in
271*3.*) ccflags="$ccflags -DDEC_OSF1_3_X" ;;
272esac
8a019ce7 273
b971f6e4 274# If debugging or (old systems and doing shared)
275# then do not strip the lib, otherwise, strip.
276# As noted above the -DDEBUGGING is added automagically by Configure if -g.
8a019ce7 277case "$optimize" in
278 *-g*) ;; # left intentionally blank
b8f0c030 279*) case "`uname -r`" in
b971f6e4 280 *[123].*)
281 case "$useshrplib" in
282 false|undef|'') lddlflags="$lddlflags -s" ;;
283 esac
284 ;;
8a019ce7 285 *) lddlflags="$lddlflags -s"
b971f6e4 286 ;;
287 esac
288 ;;
8a019ce7 289esac
8e964347 290
291#
313489a2 292# Make embedding in things like INN and Apache more memory friendly.
293# Keep it overridable on the Configure command line, though, so that
294# "-Uuseshrplib" prevents this default.
295#
296
2bf2710f 297case "$_DEC_cc_style.$useshrplib" in
298 new.) useshrplib="$define" ;;
299esac
313489a2 300
85ab1d1d 301# The EFF_ONLY_OK from <sys/access.h> is present but dysfunctional for
302# [RWX]_OK as of Digital UNIX 4.0[A-D]?. If and when this gets fixed,
303# please adjust this appropriately. See also pp_sys.c just before the
304# emulate_eaccess().
5ff3f7a4 305
baa8820a 306# Fixed in V5.0A.
144df5e1 307case "$myosvers" in
f60a5d46 308*5.0[A-Z]*|*5.[1-9]*|*[6-9].[0-9]*)
baa8820a 309 : ok
310 ;;
311*)
312# V5.0 or previous
5ff3f7a4 313pp_sys_cflags='ccflags="$ccflags -DNO_EFF_ONLY_OK"'
baa8820a 314 ;;
315esac
5ff3f7a4 316
6b8eaf93 317# The off_t is already 8 bytes, so we do have largefileness.
318
f60a5d46 319cat > UU/usethreads.cbu <<'EOCBU'
104d25b7 320# This script UU/usethreads.cbu will get 'called-back' by Configure
321# after it has prompted the user for whether to use threads.
104d25b7 322case "$usethreads" in
323$define|true|[yY]*)
700a71f5 324 # In Tru64 V5 (at least V5.1A, V5.1B) gcc (at least 3.2.2)
325 # cannot be used to compile a threaded Perl.
326 cat > pthread.c <<EOF
327#include <pthread.h>
328extern int foo;
329EOF
330 $cc -c pthread.c 2> pthread.err
331 if grep -q "unrecognized compiler" pthread.err; then
332 cat >&4 <<EOF
333***
334*** I'm sorry but your C compiler ($cc) cannot be used to
335*** compile Perl with threads. The system C compiler should work.
336***
337
338Cannot continue, aborting.
339
340EOF
341 rm -f pthread.*
342 exit 1
343 fi
344 rm -f pthread.*
75d72f2c 345 # Threads interfaces changed with V4.0.
6c7aff00 346 case "$isgcc" in
700a71f5 347 gcc)
348 ccflags="-D_REENTRANT $ccflags"
349 ;;
75d72f2c 350 *) case "`uname -r`" in
351 *[123].*) ccflags="-threads $ccflags" ;;
352 *) ccflags="-pthread $ccflags" ;;
353 esac
104d25b7 354 ;;
75d72f2c 355 esac
356 case "`uname -r`" in
357 *[123].*) libswanted="$libswanted pthreads mach exc c_r" ;;
358 *) libswanted="$libswanted pthread exc" ;;
359 esac
e883c329 360
baa8820a 361 case "$usemymalloc" in
24130e51 362 '')
363 usemymalloc='n'
baa8820a 364 ;;
365 esac
a48ec845 366 # These symbols are renamed in <time.h> so
367 # that the Configure hasproto doesn't see them.
368 d_asctime_r_proto="$define"
369 d_ctime_r_proto="$define"
370 d_gmtime_r_proto="$define"
371 d_localtime_r_proto="$define"
104d25b7 372 ;;
373esac
374EOCBU
375
da0b61dd 376# malloc wrap works
377case "$usemallocwrap" in
378'') usemallocwrap='define' ;;
379esac
380
f60a5d46 381cat > UU/uselongdouble.cbu <<'EOCBU'
382# This script UU/uselongdouble.cbu will get 'called-back' by Configure
383# after it has prompted the user for whether to use long doubles.
384case "$uselongdouble" in
bef5f079 385$define|true|[yY]*)
144df5e1 386 case "$myosvers" in
bef5f079 387 *[1-4].0*) cat >&4 <<EOF
388
389***
390*** Sorry, you cannot use long doubles in pre-V5.0 releases of Tru64.
391***
392
393Cannot continue, aborting.
394
395EOF
396 exit 1
397 ;;
fa17d112 398 *)
399 # Test whether libc's been fixed yet.
400 cat >try.c <<\TRY
401#include <stdio.h>
402int main(int argc, char **argv)
403{
404 unsigned long uvmax = ~0UL;
405 long double ld = uvmax + 0.0L;
406 char buf1[30], buf2[30];
407
408 (void) sprintf(buf1, "%lu", uvmax);
409 (void) sprintf(buf2, "%.0Lf", ld);
410 return strcmp(buf1, buf2) != 0;
411}
412TRY
413 # Don't bother trying to work with Configure's idea of
414 # cc and the various flags. This might not work as-is
415 # with gcc -- but we're testing libc, not the compiler.
24801a4b 416 if cc -o try -std1 try.c && ./try
fa17d112 417 then
418 : ok
419 else
420 cat <<\UGLY >&4
421!
422Warning! Your libc has not yet been patched so that its "%Lf" format for
423printing long doubles shows all the significant digits. You will get errors
424in the t/op/numconvert test because of this. (The data is still good
425internally, and the "%e" format of printf() or sprintf() in perl will still
426produce valid results.) See README.tru64 for additional details.
427
428Continuing anyway.
429!
430UGLY
431 fi
432 $rm -f try try.c
bef5f079 433 esac
bef5f079 434 ;;
f60a5d46 435esac
436EOCBU
437
144df5e1 438case "$myosvers" in
bef5f079 439*[1-4].0*) d_modfl=undef ;; # must wait till 5.0
1cade9fc 440esac
441
cfc8a802 442# Keep that leading tab.
a4ccfa76 443 old_LD_LIBRARY_PATH=$LD_LIBRARY_PATH
444for p in $loclibpth
445do
a4ccfa76 446 if test -d $p; then
447 echo "Appending $p to LD_LIBRARY_PATH." >& 4
448 case "$LD_LIBRARY_PATH" in
449 '') LD_LIBRARY_PATH=$p ;;
450 *) LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$p ;;
451 esac
452 fi
453done
454case "$LD_LIBRARY_PATH" in
455"$old_LD_LIBRARY_PATH") ;;
456*) echo "LD_LIBRARY_PATH is now $LD_LIBRARY_PATH." >& 4 ;;
c93bd003 457esac
5460b889 458case "$LD_LIBRARY_PATH" in
459'') ;;
460* ) export LD_LIBRARY_PATH ;;
461esac
c93bd003 462
313489a2 463#
e60a08f8 464# Unset temporary variables no more needed.
465#
466
467unset _DEC_cc_style
468
469#
8e964347 470# History:
471#
85ab1d1d 472# perl5.005_51:
473#
474# September-1998 Jarkko Hietaniemi <jhi@iki.fi>
475#
476# * Added the -DNO_EFF_ONLY_OK flag ('use filetest;' support).
477#
313489a2 478# perl5.004_57:
479#
480# 19-Dec-1997 Spider Boardman <spider@Orb.Nashua.NH.US>
481#
85ab1d1d 482# * Newer Digital UNIX compilers enforce signaling for NaN without
313489a2 483# -ieee. Added -fprm d at the same time since it's friendlier for
484# embedding.
485#
486# * Fixed the library search path to match cc, ld, and /sbin/loader.
487#
488# * Default to building -Duseshrplib on newer systems. -Uuseshrplib
489# still overrides.
490#
491# * Fix -pthread additions for useshrplib. ld has no -pthread option.
492#
493#
723e14d4 494# perl5.004_04:
495#
496# 19-Sep-1997 Spider Boardman <spider@Orb.Nashua.NH.US>
497#
498# * libnet on Digital UNIX is for JAVA, not for sockets.
499#
500#
b971f6e4 501# perl5.003_28:
502#
503# 22-Feb-1997 Jarkko Hietaniemi <jhi@iki.fi>
504#
505# * Restructuring Spider's suggestions.
506#
507# * Older Digital UNIXes cannot handle -Olimit ... for $lddlflags.
508#
509# * ld -s cannot be used in older Digital UNIXes when doing shared.
510#
511#
512# 21-Feb-1997 Spider Boardman <spider@Orb.Nashua.NH.US>
513#
514# * -hidden removed.
515#
516# * -DSTANDARD_C removed.
517#
518# * -D_INTRINSICS added. (that -fast does not seem to buy much confirmed)
519#
520# * odbm not in libc, only ndbm. Therefore dbm back to $libswanted.
521#
522# * -msym for the newer runtime loaders.
523#
524# * $optimize also in $lddflags.
525#
526#
e60a08f8 527# perl5.003_27:
528#
529# 18-Feb-1997 Jarkko Hietaniemi <jhi@iki.fi>
530#
531# * unset _DEC_cc_style and more commentary on -std.
532#
533#
dfa3a3d3 534# perl5.003_26:
535#
536# 15-Feb-1997 Jarkko Hietaniemi <jhi@iki.fi>
537#
e60a08f8 538# * -std and -ansi.
dfa3a3d3 539#
540#
f2c69087 541# perl5.003_24:
542#
543# 30-Jan-1997 Jarkko Hietaniemi <jhi@iki.fi>
544#
545# * Fixing the note on -DDEBUGGING.
546#
547# * Note on -O5 -fast.
548#
549#
8a019ce7 550# perl5.003_23:
551#
552# 26-Jan-1997 Jarkko Hietaniemi <jhi@iki.fi>
553#
554# * Notes on how to do both optimisation and debugging.
555#
556#
557# 25-Jan-1997 Jarkko Hietaniemi <jhi@iki.fi>
558#
559# * Remove unneeded libraries from $libswanted: PW, bsd, c, dbm
560#
561# * Restructure the $lddlflags build.
562#
563# * $optimize based on which compiler we have.
564#
565#
8e964347 566# perl5.003_22:
567#
568# 23-Jan-1997 Achim Bohnet <ach@rosat.mpe-garching.mpg.de>
569#
570# * Added comments 'how to create a debugging version of perl'
571#
572# * Fixed logic of this script to prevent stripping of shared
573# objects by the loader (see ld man page for -s) is debugging
574# is set via the -g switch.
575#
576#
577# 21-Jan-1997 Achim Bohnet <ach@rosat.mpe-garching.mpg.de>
578#
579# * now 'dl' is always removed from libswanted. Not only if
580# optimize is an empty string.
581#
582#
583# 17-Jan-1997 Achim Bohnet <ach@rosat.mpe-garching.mpg.de>
584#
585# * Removed 'dl' from libswanted: When the FreePort binary
586# translator for Sun binaries is installed Configure concludes
587# that it should use libdl.x.yz.fpx.so :-(
588# Because the dlopen, dlclose,... calls are in the
589# C library it not necessary at all to check for the
590# dl library. Therefore dl is removed from libswanted.
591#
592#
593# 1-Jan-1997 Achim Bohnet <ach@rosat.mpe-garching.mpg.de>
594#
595# * Set -Olimit to 3200 because perl_yylex.c got too big
596# for the optimizer.
597#