It would seem that in Linux 2.2 there is no way

we can do the right thing to make ps(1) see $0
our way.  Also amend the $0 documentation a bit.

p4raw-id: //depot/perl@19935

diff --git a/pod/perlvar.pod b/pod/perlvar.pod
index e00ec4f..3a2ed3f 100644 (file)
--- a/pod/perlvar.pod
+++ b/pod/perlvar.pod
@@ -838,18 +838,22 @@ and C<$)> can be swapped only on machines supporting setregid().
 
 =item $0
 
-Contains the name of the program being executed.  On some (read: not
-all) operating systems assigning to C<$0> modifies the argument area
-that the C<ps> program sees.  On some platforms you may have to use
-special C<ps> options or a different C<ps> to see the changes.
-Modifying the $0 is more useful as a way of indicating thecurrent
-program state than it is for hiding the program you're running.
-(Mnemonic: same as B<sh> and B<ksh>.)
+Contains the name of the program being executed.
+
+On some (read: not all) operating systems assigning to C<$0> modifies
+the argument area that the C<ps> program sees.  On some platforms you
+may have to use special C<ps> options or a different C<ps> to see the
+changes.  Modifying the $0 is more useful as a way of indicating the
+current program state than it is for hiding the program you're
+running.  (Mnemonic: same as B<sh> and B<ksh>.)
 
 Note that there are platform specific limitations on the the maximum
 length of C<$0>.  In the most extreme case it may be limited to the
 space occupied by the original C<$0>.
 
+In some platforms there may be arbitrary amount of padding, for
+example space characters, after the modified name as shown by C<ps>.
+
 Note for BSD users: setting C<$0> does not completely remove "perl"
 from the ps(1) output.  For example, setting C<$0> to C<"foobar"> may
 result in C<"perl: foobar (perl)"> (whether both the C<"perl: "> prefix
@@ -858,7 +862,9 @@ and version).  This is an operating system feature, Perl cannot help it.
 
 In multithreaded scripts Perl coordinates the threads so that any
 thread may modify its copy of the C<$0> and the change becomes visible
-to ps(1) (assuming the operating system plays along).
+to ps(1) (assuming the operating system plays along).  Note that the
+the view of C<$0> the other threads have will not change since they
+have their own copies of it.
 
 =item $[
 

diff --git a/t/op/magic.t b/t/op/magic.t
index 8f0db78..545d34b 100755 (executable)
--- a/t/op/magic.t
+++ b/t/op/magic.t
@@ -300,7 +300,13 @@ else {
             };
             my $ps = $mydollarzero->("x");
             ok(!$ps  # we allow that something goes wrong with the ps command
-               || $ps eq 'x'
+              # In Linux 2.4 we would get an exact match ($ps eq 'x') but
+              # in Linux 2.2 there seems to be something funny going on:
+              # it seems as if the original length of the argv[] would
+              # be stored in the proc struct and then used by ps(1),
+              # no matter what characters we use to pad the argv[].
+              # (And if we use \0:s, they are shown as spaces.)  Sigh.
+               || $ps =~ /^x\s*$/
               # FreeBSD cannot get rid of both the leading "perl :"
               # and the trailing " (perl)": some FreeBSD versions
               # can get rid of the first one.