On Wed, Feb 14, 2007 at 09:18:19PM +0000, David Cantrell wrote:
> Perl does not build properly on NetBSD 2.1.0 for Alpha hardware, when
> using 'Configure -de'. The solitary failing test is ...
>
> $ ./perl ../ext/POSIX/t/sigaction.t
> ...
> not ok 5
>
> Perl was built thus:
>
> $ rm -f config.sh Policy.sh
> $ sh Configure -Dprefix=/arpa/af/d/drhyde/cpantesting/perl-5.8.8 -de
> $ make
> $ make test
>
> This BEGIN block in the test file looks like a good place to start
> investigating what's going on.
>
> BEGIN{
> # Don't do anything if POSIX is missing, or sigaction missing.
> use Config;
> eval 'use POSIX';
> if($@ || $^O eq 'MSWin32' || $^O eq 'NetWare' || $^O eq 'dos' ||
> $^O eq 'MacOS' || ($^O eq 'VMS' && !$Config{'d_sigaction'})) {
> print "1..0\n";
> exit 0;
> }
> }
This may be something particular to earlier NetBSD systems, since my
NetBSD 3.1 smokes seem to be running just fine.
Running verbose tests shows the following tests...
ok 1 - no warnings
ok 2 - DEFAULT
ok 3
ok 4
ok 5 - SIGUSR1 ismember MASK
Looking at the code, it seems to be handling some sigaction
related code corectly, but not all of it. Can you please try the
test again by going in the t/ directory and running
./perl harness -v ../ext/POSIX/t/sigaction.t
In the meantime, I'll look at making this test a little more vocal
in case of errors, since there are several system calls that should
set $!, but there are no checks for $!. Also, can you please check the
manpages for sigaction() and sigismember() for any possible clues
for this behavior where a portion of the sigsets are getting copied
fine?
Thanks much,
Steve Peters
[EMAIL PROTECTED]
