IO::Handle::clearerr leaves EOF state latched after file growth #1922

Summary

IO::Handle::clearerr clears $! but does not clear the EOF/read-ahead state held by a file channel. If a file grows after a read reaches EOF, a subsequent read after clearerr still returns undef in PerlOnJava.

This caused the Log::Procmail 0.14 CPAN regression in run 20261029-112411: t/20abstract.t dies at line 121 after reading a line appended to a log file. The archived system Perl 5.42.2 run passed all 281 tests. PerlOnJava failed the same assertion on both JVM and interpreter backends. The last recorded passing CPAN run was 2026-10-07 at 66b868f89.

Minimal reproducer

use File::Temp qw(tempfile);
use IO::File;
my ($writer, $path) = tempfile();
print {$writer} "first\n";
close $writer;
my $reader = IO::File->new($path, 'r') or die $!;
<$reader>;
<$reader>; # reach EOF
open $writer, '>>', $path or die $!;
print {$writer} "second\n";
close $writer;
$reader->clearerr;
my $line = <$reader>;
print defined($line) ? $line : "undef\n";

Standard Perl prints second. PerlOnJava prints undef on both backends.

Expected behavior

After clearerr, a read from a file that has grown since EOF should be able to consume the newly appended data. clearerr should clear the relevant underlying EOF/read-ahead state as well as $!.

Environment and evidence

• Archived CPAN run: 20261029-112411, distribution Log-Procmail-0.14
• Failing test: t/20abstract.t, line 121; 1/9 test programs failed, 0/261 reported subtests failed
• System Perl oracle: Perl 5.42.2, all 9 programs and 281 tests passed
• PerlOnJava: both JVM and interpreter fail at the same test
• Reproduced on macOS 26.7.1 with OpenJDK 24.0.2; current checkout source commit a5bb374b6
• Root cause evidence: IO::Handle::_clearerr only clears $!; CustomFileChannel retains physicalEof/isEOF and buffered read state