hpux libperl.a linkage
Doug Maceachern
the hpux cc/ld, unless +Z is added to ccflags. i have no hints-fu to get
the change right, figure it'll be a sinch for somebody who does.
in the meantime, i've been building on hpux with -Accflags=+Z which works
fine. (also required for 5.6.1)
from the cc manpage...
+z,+Z Both of these options cause the compiler to
generate position independent code (PIC) for use in
building shared libraries. +Z is the default in
64-bit mode. The -G and -p options are ignored if
+z or +Z is used. Normally, +z should be used to
generate PIC; however, when certain limits are
exceeded, +Z is required to generate PIC. The ld
linker issues the error indicating when +Z is
required. If both +z and +Z are specified, only the
last one encountered applies.
passes all tests:
Summary of my perl5 (revision 5.0 version 7 subversion 3 patch 16749) configuration:
Platform:
osname=hpux, osvers=11.00, archname=PA-RISC2.0-thread-multi
uname='hp-ux horace b.11.00 a 9000785 2010607443 two-user license '
config_args='-des -Dprefix=/local0/tmp/dougm/perl -Accflags=+Z -Dusedevel -Doptimize=-g -Dusethreads'
hint=recommended, useposix=true, d_sigaction=define
usethreads=define use5005threads=undef useithreads=define usemultiplicity=define
useperlio=define d_sfio=undef uselargefiles=define usesocks=undef
use64bitint=undef use64bitall=undef uselongdouble=undef
usemymalloc=n, bincompat5005=undef
Compiler:
cc='cc', ccflags =' -D_POSIX_C_SOURCE=199506L -D_REENTRANT -Ae -D_HPUX_SOURCE -Wl,+vnocompatwarnings +Z -DDEBUGGING -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 ',
optimize='-g',
cppflags='-Aa -D__STDC_EXT__ -D_HPUX_SOURCE -D_POSIX_C_SOURCE=199506L -D_REENTRANT -Ae -D_HPUX_SOURCE -Wl,+vnocompatwarnings +Z -DDEBUGGING -I/usr/local/include'
ccversion='B.11.11.02', gccversion='', gccosandvers=''
intsize=4, longsize=4, ptrsize=4, doublesize=8, byteorder=4321
d_longlong=define, longlongsize=8, d_longdbl=define, longdblsize=16
ivtype='long', ivsize=4, nvtype='double', nvsize=8, Off_t='off_t', lseeksize=8
alignbytes=8, prototype=define
Linker and Libraries:
ld='/usr/bin/ld', ldflags =' -L/usr/local/lib'
libpth=/usr/local/lib /lib /usr/lib /usr/ccs/lib
libs=-lnsl -lnm -lndbm -lmalloc -ldld -lm -lpthread -lc -lndir -lcrypt -lsec
perllibs=-lnsl -lnm -lmalloc -ldld -lm -lpthread -lc -lndir -lcrypt -lsec
libc=/lib/libc.sl, so=sl, useshrplib=false, libperl=libperl.a
Dynamic Linking:
dlsrc=dl_hpux.xs, dlext=sl, d_dlsymun=undef, ccdlflags='-Wl,-E -Wl,-B,deferred '
cccdlflags='+Z', lddlflags='-b +vnocompatwarnings -L/usr/local/lib'
Characteristics of this binary (from libperl):
Compile-time options: DEBUGGING MULTIPLICITY USE_ITHREADS USE_LARGE_FILES PERL_IMPLICIT_CONTEXT
Locally applied patches:
DEVEL16734
Built under hpux
Compiled at May 22 2002 18:43:02
@INC:
/local0/tmp/dougm/perl/lib/5.7.3/PA-RISC2.0-thread-multi
/local0/tmp/dougm/perl/lib/5.7.3
/local0/tmp/dougm/perl/lib/site_perl/5.7.3/PA-RISC2.0-thread-multi
/local0/tmp/dougm/perl/lib/site_perl/5.7.3
/local0/tmp/dougm/perl/lib/site_perl
.
H.Merijn Brand
> mod_perl.so cannot link in libperl.a on hpux if perl was built with
> the hpux cc/ld, unless +Z is added to ccflags. i have no hints-fu to get
> the change right, figure it'll be a sinch for somebody who does.
But +Z *is* default when using HP's C compiler, and it is set in line 394
cccdlflags='+Z'
Though I personally would not object in promoting this flag to ccflags, since
it does not /harm/ static builds.
Any objections?
> in the meantime, i've been building on hpux with -Accflags=+Z which works
> fine. (also required for 5.6.1)
Does converting libperl.a to libperl.sl work?
# mkdir tmp
# cd tmp
# ar x ../libperl.a
# ld -b -o libperl.sl *.o
# mv libperl.sl ..
# cd ..
# rm -rf tmp
/If/ all objects were created with +Z, this should work
--
H.Merijn Brand Amsterdam Perl Mongers (http://amsterdam.pm.org/)
using perl-5.6.1, 5.7.3 & 631 on HP-UX 10.20 & 11.00, AIX 4.2, AIX 4.3,
WinNT 4, Win2K pro & WinCE 2.11. Smoking perl CORE: smo...@perl.org
http://archives.develooper.com/daily...@perl.org/ per...@perl.org
send smoke reports to: smokers...@perl.org, QA: http://qa.perl.org
Jarkko Hietaniemi
> On Thu 23 May 2002 04:25, Doug MacEachern <do...@covalent.net> wrote:
> > mod_perl.so cannot link in libperl.a on hpux if perl was built with
> > the hpux cc/ld, unless +Z is added to ccflags. i have no hints-fu to get
> > the change right, figure it'll be a sinch for somebody who does.
>
> But +Z *is* default when using HP's C compiler, and it is set in line 394
>
> cccdlflags='+Z'
>
> Though I personally would not object in promoting this flag to ccflags, since
> it does not /harm/ static builds.
>
> Any objections?
Go ahead (with a patch).
--
$jhi++; # http://www.iki.fi/jhi/
# There is this special biologist word we use for 'stable'.
# It is 'dead'. -- Jack Cohen
H.Merijn Brand
> On Thu, May 23, 2002 at 09:07:47AM +0200, H.Merijn Brand wrote:
> > On Thu 23 May 2002 04:25, Doug MacEachern <do...@covalent.net> wrote:
> > > mod_perl.so cannot link in libperl.a on hpux if perl was built with
> > > the hpux cc/ld, unless +Z is added to ccflags. i have no hints-fu to get
> > > the change right, figure it'll be a sinch for somebody who does.
> >
> > But +Z *is* default when using HP's C compiler, and it is set in line 394
> >
> > cccdlflags='+Z'
> >
> > Though I personally would not object in promoting this flag to ccflags, since
> > it does not /harm/ static builds.
> >
> > Any objections?
>
> Go ahead (with a patch).
I *think* this would be OK, but I'm not sure about removing the +Z from ...
Any implications? I think ... is always used together with $ccflags, so I'm
safe here. Second pair of eyes please
All tests successful.
u=3.03 s=1.18 cu=241.85 cs=17.75 scripts=658 tests=64177
--- hints/hpux.sh 2002-05-16 16:51:26.000000000 +0200
+++ hints/hpux.sh 2002-05-23 13:39:28.000000000 +0200
@@ -137,7 +137,12 @@ case `$cc -v 2>&1`"" in
ccversion=`which cc | xargs what | awk '/Compiler/{print $2}'`
case "$ccflags" in
"-Ae "*) ;;
- *) ccflags="-Ae $cc_cppflags -Wl,+vnocompatwarnings" ;;
+ *) ccflags="-Ae $cc_cppflags -Wl,+vnocompatwarnings" ;;
+ esac
+ case "$ccflags" in
+ *"+Z"*) ;;
+ *"+z"*) ccflags=`echo "$ccflags" | sed -e 's/\+z/+Z/'` ;;
+ *) ccflags="$ccflags +Z" ;;
esac
# Needed because cpp does only support -Aa (not -Ae)
cpplast='-'
@@ -391,7 +396,7 @@ case "$ccisgcc" in
;;
esac
ld=/usr/bin/ld
- cccdlflags='+Z'
+ cccdlflags=''
lddlflags='-b +vnocompatwarnings'
;;
esac
Doug Maceachern
> But +Z *is* default when using HP's C compiler, and it is set in line 394
>
> cccdlflags='+Z'
that does not get applied to sources in libperl.a, maybe it should?
just noticed this comment in the main Makefile:
# Static targets are ordinarily built without CCCDLFLAGS. However,
# if building a shared libperl.so that might later be linked into
# another application, then it might be appropriate to also build static
# extensions (usually just DynaLoader) with relocatable code (e.g. -fPIC
# for GNU cc). This is handled by ext/util/make_ext.
libperl.a should be considered the same way as DynaLoader.a, that might
get linked into another app.
was about to see if -fPIC is applied on solaris, perl-current chokes here:
make: *** No rule to make target `wince.h', needed by `perl.h'. Stop.
puzzling.
Vadim Konovalov
really puzzling. Why did you noticed wince.h at all?
Even wince's elder brother Win32 should not depend on WinCE in any way.
Which line? Which extension?
Vadim.
Andy Dougherty
> On Thu, 23 May 2002, H.Merijn Brand wrote:
>
> > But +Z *is* default when using HP's C compiler, and it is set in line 394
> >
> > cccdlflags='+Z'
>
> that does not get applied to sources in libperl.a, maybe it should?
>
> just noticed this comment in the main Makefile:
> # Static targets are ordinarily built without CCCDLFLAGS. However,
> # if building a shared libperl.so that might later be linked into
> # another application, then it might be appropriate to also build static
> # extensions (usually just DynaLoader) with relocatable code (e.g. -fPIC
> # for GNU cc). This is handled by ext/util/make_ext.
>
> libperl.a should be considered the same way as DynaLoader.a, that might
> get linked into another app.
But libperl.a is not a shared library (it would be libperl.sl on HP/UX, I
think).
There is (or at least there was when I tested it some years ago) a
potentially significant performance penalty involved in building libperl.a
with PIC code. If you weren't going to link libperl into another app,
then it seemed inappropriate to pay the penalty.
I think that -fPIC (or +Z or whatever) ought to be invoked only if needed.
Or, alternatively, it ought not to be invoked when not needed.
Phrased differently, I would object on performance grounds to adding in
-fPIC all the time even when I definitely don't want it. However, a patch
to make building a shared libperl.sl the default (overrideable by
-Uuseshrplib or some such) might be reasonable.
> was about to see if -fPIC is applied on solaris, perl-current chokes here:
> make: *** No rule to make target `wince.h', needed by `perl.h'. Stop.
I've seen that before. It's usually some sort of half-complete build or
timestamp weirdness (esp. on NFS with drifting clocks). Restarting with a
clean distribution (or fixing up time stamps) probably will make it go
away.
--
Andy Dougherty doug...@lafayette.edu
Dept. of Physics
Lafayette College, Easton PA 18042
Doug Maceachern
> But libperl.a is not a shared library (it would be libperl.sl on HP/UX, I
> think).
DynaLoader.a is not a shared library either, but the comment seemed to
imply cccdlflags are applied to it. i see that is not the case now, at
least i don't see it happening on hpux.
> There is (or at least there was when I tested it some years ago) a
> potentially significant performance penalty involved in building libperl.a
> with PIC code. If you weren't going to link libperl into another app,
> then it seemed inappropriate to pay the penalty.
>
> I think that -fPIC (or +Z or whatever) ought to be invoked only if needed.
> Or, alternatively, it ought not to be invoked when not needed.
fair enough. modperl's Makefile.PL has a check tells the use to rebuild
with -Accflags=+Z if needed. would be interesting to see modern day
benchmarks with and without +Z.
> I've seen that before. It's usually some sort of half-complete build or
> timestamp weirdness (esp. on NFS with drifting clocks). Restarting with a
> clean distribution (or fixing up time stamps) probably will make it go
> away.
that did the trick, thanks.
H.Merijn Brand
> On Thu, 23 May 2002, Doug Maceachern wrote:
>
> > On Thu, 23 May 2002, H.Merijn Brand wrote:
> >
> > > But +Z *is* default when using HP's C compiler, and it is set in line 394
> > >
> > > cccdlflags='+Z'
> >
> > that does not get applied to sources in libperl.a, maybe it should?
> >
> > just noticed this comment in the main Makefile:
> > # Static targets are ordinarily built without CCCDLFLAGS. However,
> > # if building a shared libperl.so that might later be linked into
> > # another application, then it might be appropriate to also build static
> > # extensions (usually just DynaLoader) with relocatable code (e.g. -fPIC
> > # for GNU cc). This is handled by ext/util/make_ext.
> >
> > libperl.a should be considered the same way as DynaLoader.a, that might
> > get linked into another app.
>
> But libperl.a is not a shared library (it would be libperl.sl on HP/UX, I
> think).
>
> There is (or at least there was when I tested it some years ago) a
> potentially significant performance penalty involved in building libperl.a
> with PIC code. If you weren't going to link libperl into another app,
> then it seemed inappropriate to pay the penalty.
HP 9000/A500 (rp2450) / 1Gb / HP-UX 11.00 / HPc B.11.11.04
All objects +Z (+Z in ccflags):
All tests successful.
u=3.03 s=1.18 cu=241.85 cs=17.75 scripts=658 tests=64177
Previous (current) situation:
All tests successful.
u=2.79 s=1.04 cu=232.08 cs=16.62 scripts=658 tests=64177
speed loss of 3-9%
Jarkko: Don't apply
> I think that -fPIC (or +Z or whatever) ought to be invoked only if needed.
> Or, alternatively, it ought not to be invoked when not needed.
This is a valid argument.
> Phrased differently, I would object on performance grounds to adding in
convinced
> -fPIC all the time even when I definitely don't want it. However, a patch
> to make building a shared libperl.sl the default (overrideable by
> -Uuseshrplib or some such) might be reasonable.
Sounds OK. But if performance *is/was* an issue, I'd make the static build the
default.
Doug Maceachern
> really puzzling. Why did you noticed wince.h at all?
> Even wince's elder brother Win32 should not depend on WinCE in any way.
>
> Which line? Which extension?
top-level 'make', that's all i get:
make: *** No rule to make target `wince.h', needed by `perl.h'. Stop.
nothing else. starting over with a fresh tree as andy suggests cured the
problem.
Andy Dougherty
> On Thu, 23 May 2002, Andy Dougherty wrote:
>
> > But libperl.a is not a shared library (it would be libperl.sl on HP/UX, I
> > think).
>
> DynaLoader.a is not a shared library either, but the comment seemed to
> imply cccdlflags are applied to it. i see that is not the case now, at
> least i don't see it happening on hpux.
cccdlflags are applied to DynaLoader.a only if the user is building a
shared libperl. (That is, static_target=static_pic only if $useshrplib
is true. This is handled earlier in Makefile.SH.)
And yes, it's an obscure inelegant way to set (or not set) -fPIC (or
equivalent) but I stared at it off and on for 2 days and didn't think of
anything more elegant. :-(
Andy Dougherty
> fair enough. modperl's Makefile.PL has a check tells the use to rebuild
> with -Accflags=+Z if needed. would be interesting to see modern day
> benchmarks with and without +Z.
"Modern day"? The same hardware I did those old benchmarks on is still my
main system!
--
Andy Dougherty doug...@lafayette.edu
Gurusamy Sarathy
>On Thu, 23 May 2002, Doug MacEachern wrote:
>
>> On Thu, 23 May 2002, Andy Dougherty wrote:
>>
>> > But libperl.a is not a shared library (it would be libperl.sl on HP/UX, I
>> > think).
>>
>> DynaLoader.a is not a shared library either, but the comment seemed to
>> imply cccdlflags are applied to it. i see that is not the case now, at
>> least i don't see it happening on hpux.
>
>cccdlflags are applied to DynaLoader.a only if the user is building a
>shared libperl. (That is, static_target=static_pic only if $useshrplib
>is true. This is handled earlier in Makefile.SH.)
>
>And yes, it's an obscure inelegant way to set (or not set) -fPIC (or
>equivalent) but I stared at it off and on for 2 days and didn't think of
>anything more elegant. :-(
I seem to have been using the attached patch for a while now.
HTH,
Sarathy
gs...@ActiveState.com
-----------------------------------8<-----------------------------------
--- perl/ext/util/make_ext Thu Apr 5 22:25:53 2001
+++ ./ext/util/make_ext Mon Jun 4 17:49:32 2001
@@ -88,7 +88,12 @@
# check link type and do any preliminaries
case "$target" in
# convert 'static' or 'dynamic' into 'all LINKTYPE=XXX'
-static) makeargs="LINKTYPE=static CCCDLFLAGS="
+static) case "$mname" in
+ # For Apache, DynaLoader needs the CCCDLFLAGS variable
+ # (+z/+Z/-fpic/-fPIC) to stick around
+ *DynaLoader*) makeargs="LINKTYPE=static" ;;
+ *) makeargs="LINKTYPE=static CCCDLFLAGS=" ;;
+ esac
target=all
;;
dynamic) makeargs="LINKTYPE=dynamic";
End of Patch.
Andy Dougherty
> I seem to have been using the attached patch for a while now.
[omitted]
That's almost always fine, but my patch [16356] avoids adding the -fPIC if
you're not building a shared libperl (in which case DynaLoader is probably
the least of your problems -- hence the origin of this thread) and also
correctly adds the -fPIC to all static extensions (though usually that's
just DynaLoader).
Gurusamy Sarathy
>On Thu, 23 May 2002, Gurusamy Sarathy wrote:
>
>> I seem to have been using the attached patch for a while now.
>[omitted]
>
>That's almost always fine, but my patch [16356] avoids adding the -fPIC if
>you're not building a shared libperl
As I recall, I needed the patch to make mod_perl build its *.so's
properly (because mod_perl links with libperl.a and DynaLoader.a?).
I was not building a shared libperl per se.
>(in which case DynaLoader is probably
>the least of your problems -- hence the origin of this thread) and also
>correctly adds the -fPIC to all static extensions (though usually that's
>just DynaLoader).
Sarathy
gs...@ActiveState.com