Subject: port-alpha/17926: gcc bug?
To: None <>
From: Ray Phillips <>
List: netbsd-bugs
Date: 08/13/2002 16:52:05
>Number:         17926
>Category:       port-alpha
>Synopsis:       perl 5.6.1 test generates core dump
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-alpha-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Aug 12 23:53:00 PDT 2002
>Originator:     Ray Phillips
>Release:        NetBSD 1.6_BETA5

System: NetBSD 1.6_BETA5 NetBSD 1.6_BETA5 
(GENERIC) #1: Thu Aug 1 16:00:09 MEST 2002 
Machine: alpha


When installing perl 5.6.1 from its source file, perl-5.6.1.tar.gz, 
sh Configure  and  make  run without problems but  make test 
produces an error:

lib/bigfltpm.........FAILED at test 351

Running  ./perl harness  from ./t to get "more granularity" gives:

         Test returned status 0 (wstat 136, 0x88)
         test program seems to have generated a core
DIED. FAILED tests 351-370
         Failed 20/370 tests, 94.59% okay

I've written to Jarkko Hietaniemi (one of the main Perl maintainers) 
who said gcc on the Alpha is "notoriously buggy" and suggested trying 
2.95.4--not a possibility since gcc needs some special tweaking 
before it will work with NetBSD doesn't it?

I've tried answering "No" to perl's Configure's

Try to use maximal 64-bit support, if available? [y]

question to see if that made a difference.  It didn't affect  make 
test's  output but a new error appeared when  ./perl harness  was run 
(as well as the bigfltpm one):

pragma/locale.......# The following locales
#       C POSIX zh_CN.eucCN
# tested okay.
# The following locales
#       zh_TW.eucTW
# had problems.
FAILED tests 105-106, 108-109, 111, 114
         Failed 6/116 tests, 94.83% okay

Jarkko said the fact that "64-bitness support affects locales is just 
so bizarre that I even more suspect something fishy going on either 
in gcc/alpha or in netbsd/alpha."

The strange thing is the very newest perl (5.8.0) compiles and 
installs fine under NetBSD/alpha BETA5; and neither 5.6.1 nor 5.8.0 
produce errors under NetBSD/i386 1.6BETA4.

Sorry if this report is a red herring.  I thought it might highlight 
problem lurking somewhere inside gcc.