This is mostly an autosummary. Knut.Hellebo_at_nho.hydro.com and
"Steven N. Hirsch" <shirsch_at_adelphia.net> sent some suggestions
and Steven sent a patch that appeared appropriate to Berkeley DB 1.86,
but I didn't hear from anyone who has tried to compile perl 5.003_03
on TU50. So I don't know if the problem is general or specific.
I'll describe what I've done in the event it is general.
The original problem was the db-hash test on a Tru64 Ver 5.0 (tu50)
system failed when compiling perl 5.003_03. The same compile
on DU40b was fine. (Details in the original message appended at the
end.) Working on the TU50 system (a DS20, latest firmware, etc.),
I obtained from 'www.sleepycat.com' various versions of BerkeleyDB.
In doing these compiles of perl, I moved /usr/shlib/libdb.so
to a holding directory so that it would not use the old shared
library version. (If I was better at this, I would have compiled
a new shared library version of libdb for /usr/local/lib, but
my understanding of how to do that is basically nil.)
Here's what I've found so far:
With a new db-185 installed, I can replicate the segmentation fault
that occurs with a plain old tru64 5.0 test.
With db-186, I there is no problem with seg faulting, but the speed
in testing lib/db-hash.t is abysmal (26 sec vs 3 sec on a DU40b system).
One location of the speed problem is in section:
--------- in t/lib/db-hash.t
# check cache overflow and numeric keys and contents
$ok = 1;
for ($i = 1; $i < 200; $i++) { $h{$i + 0} = $i + 0; }
for ($i = 1; $i < 200; $i++) { $ok = 0 unless $h{$i} == $i; }
ok(28, $ok );
---------
I applied some patches that helped "Steven N. Hirsch"
<shirsch_at_adelphia.net> on an AIX machine, but these did not
speed things up.
With db-277, there is no problem with seg faults and the speed
to test lib/db-hash.t is less than 1 second.
Compiling db-3055 had some problems and so I never did test
using that version. (Berkeley DB Ver 3.0.55 came out in Nov 99.)
So the solution I will probably use is to stick with Version 2.7.7
of Berkeley DB and now will try to make sure sendmail 8.9.3 compiles
with these also. If all that is successful, I'll need to move
my existing DU 4.0b and Ultrix 4.3a to use that version. All these
versions of BerkeleyDB (1.85, 1.86, 2.7.7) have _different_ database
structures, so one can't read one database created in one
version with another. (Older linux systems used 1.85; a linux
system built from Redhat 6.2 uses a database format compatible
with 2.7.7.)
My current plan is to leave /usr/shlib/libdb.so alone so that any
existing TU50 programs that need the shared library will find it.
During the compiles (and _only_ during the compiles), I'm moving
it aside so that the programs use the static library /usr/local/libdb.a
instead. After that libdb.so is moved back where it belongs.
I think that a smarter person would build a new shared library from the
Versions 2.7.7 Berkeley DB package using the --enable-compat185
configure switch. I think....
-mike
-----------------------------------------------------------------------------
Michael A. Crowley Director of Networking
mcrowley_at_mtholyoke.edu 216 Dwight Hall, Mount Holyoke College
413-538-2140 fax: 413-538-2331 South Hadley, MA 01075-6415
http://www.mtholyoke.edu/~mcrowley http://www.mtholyoke.edu/lits/network
-----------------------------------------------------------------------------
On Thu, 30 Dec 1999, Michael A Crowley wrote the original question:
>
>On a DS20 with TU50 (tru64 5.0), patch kit 1....
>Configuring using all the defaults, compiles fine.
>It fails the test of file: t/lib/db-hash.t
>
>It seg faults in that file after test 21 at the "delete" line
>listed below (added test prints to see where it was blowing up):
>......
>print "TEST1:\n";
>$h{'goner2'} = 'snork';
>print "TEST2:\n";
>delete $h{'goner2'}; # seg faults on this line.
>print "TEST3:\n";
>......
>
>This occurs on 5.005.03 and the latest development version 5.005.63.
>It does not occur when I've built on DU 4.0b.
>
>Is this just me and/or have I missed something or have others experienced
>this. Given the apparent lack of messages on this, I am hoping it is
>something I'm omitting.
>
>Screen dumps of the configuration, make, and test are at:
> http://www.mtholyoke.edu/~mcrowley/perl-tmp/
>
>thanks,
>-mike
>
>-----------------------------------------------------------------------------
> Michael A. Crowley Director of Networking
> mcrowley_at_mtholyoke.edu 216 Dwight Hall, Mount Holyoke College
> 413-538-2140 fax: 413-538-2331 South Hadley, MA 01075-6415
> http://www.mtholyoke.edu/~mcrowley http://www.mtholyoke.edu/lits/network
>-----------------------------------------------------------------------------
>
>
Received on Tue Jan 04 2000 - 00:30:52 NZDT