Is "Modification of a read-only value attempted" really a fatal error?

[jkeenan]

Consider this file, which is derived from ext/PerlIO-scalar/t/scalar.t in the Perl 5 core distribution:

$ cat xscalar.t 
#!./perl

BEGIN {
    unless (find PerlIO::Layer 'perlio') {
	print "1..0 # Skip: not perlio\n";
	exit 0;
    }
    require Config;
    if (($Config::Config{'extensions'} !~ m!\bPerlIO/scalar\b!) ){
        print "1..0 # Skip -- Perl configured without PerlIO::scalar module\n";
        exit 0;
    }
}

use Fcntl qw(SEEK_SET SEEK_CUR SEEK_END); # Not 0, 1, 2 everywhere.
use Errno qw(EACCES);

$| = 1;

use Test::More tests =>   5;

# [perl #40267] PerlIO::scalar doesn't respect readonly-ness
{
    ok(!(defined open(F, '>', \undef)), "[perl #40267] - $!");
    close F;

    my $ro = \43;
    ok(!(defined open(F, '>', $ro)), $!);
    is($!+0, EACCES, "check we get a read-onlyish error code");
    close F;
    # but we can read from it
    ok(open(F, '<', $ro), $!);
    is(<F>, 43);
    close F;
}

Let's run it:

$ prove -v xscalar.t 
xscalar.t .. 
1..5
ok 1 - [perl \#40267] - Permission denied
ok 2 - Permission denied
ok 3 - check we get a read-onlyish error code
ok 4 - Bad file descriptor
ok 5
ok
All tests successful.
Files=1, Tests=5,  0 wallclock secs ( 0.02 usr  0.00 sys +  0.03 cusr  0.01 csys =  0.06 CPU)
Result: PASS

Unremarkable. But now let's modify it to run it with warnings.

$ cat yscalar.t 
#!./perl -w

BEGIN {
    unless (find PerlIO::Layer 'perlio') {
	print "1..0 # Skip: not perlio\n";
	exit 0;
    }
...
# As before

Let's run it.

$ prove -v yscalar.t 
yscalar.t .. 
1..5
Modification of a read-only value attempted at yscalar.t line 24.
ok 1 - [perl \#40267] - Permission denied
ok 2 - Permission denied
ok 3 - check we get a read-onlyish error code
ok 4 - Bad file descriptor
ok 5
Modification of a read-only value attempted at yscalar.t line 28.
ok
All tests successful.
Files=1, Tests=5,  0 wallclock secs ( 0.02 usr  0.00 sys +  0.05 cusr  0.00 csys =  0.07 CPU)
Result: PASS

Two warnings are emitted.

Now I might want to suppress these with a no warnings 'XXXXX; in scope. However, when I go to perldoc perldiag to learn what category to use to replace XXXXX, I see that this warning is classified F as in fatal.

Modification of a read-only value attempted
    (F) You tried, directly or indirectly, to change the value of a
    constant. You didn't, of course, try "2 = 1", because the compiler
    catches that. 

For Category F, we have:

(F) A fatal error (trappable).

Problem: The error wasn't fatal (for an ordinary definition of "fatal"). The program kept on running. Nor did we attempt to trap the error.

So, is pod/perldiag.pod wrong with respect to categorizing "Modification of a read-only value attempted"?

Thank you very much.
Jim Keenan

[xenu]

It is a fatal error but PerlIO::scalar happens to emit a warning with exactly the same text as perl's fatal error.

You can silence it with no warnings 'layer'.

[jkeenan]

On 7/15/20 10:11 PM, xenu wrote: It is a fatal error but PerlIO::scalar happens <https://github.com/Perl/perl5/blob/b9965e1496efe3cb6116e74d50aa83152c70e877/ext/PerlIO-scalar/scalar.xs#L45> to emit a warning with exactly the same text as perl's fatal error. You can silence it with |no warnings 'layer'|.

Could you propose a documentation patch? Thanks.

[Leont]

You can silence it with no warnings 'layer'.

Or add a testcase that verifies we get that warning :-)

[jkeenan]

You can silence it with no warnings 'layer'.

Or add a testcase that verifies we get that warning :-)

I know I can silence the warning, because that's what I've done in the p7 branch.

However, my main concern is this ticket is the documentation. We need to clarify what we claim should be happening before we can write a test case.

So, again, does someone want to propose a patch?

Thank you very much.
Jim Keenan