did you mean "encode() unless is_utf8()"?

No. is_utf8 returns true if the string is stored using the upgraded/wide/UTF8=1 internal storage format. So the sub encodes the string using UTF-8 if it's stored in that format.

utf8::downgrade( my $d = "\xE3" );
printf "%vX\n", $d;                          # E3
printf "%d\n", utf8::is_utf8( $d ) ? 1 : 0;  # 0
printf "%vX\n", transform( $d );             # E3

utf8::upgrade(  my $u = "\xE3" );
printf "%vX\n", $u;                          # E3
printf "%d\n", utf8::is_utf8( $u ) ? 1 : 0;  # 1
printf "%vX\n", transform( $u );             # C3.A3

utf8::upgrade( my $w = "\x{2660}" );
printf "%vX\n", $w;                          # 2660
printf "%d\n", utf8::is_utf8( $w ) ? 1 : 0;  # 1
printf "%vX\n", transform( $w );             # E2.99.A0
[download]

IO::File->new( chr (0xbb), q(>))

chr(0xbb) returns a string using the downgraded/8bit/UTF8=0 internal storage format.

Outside of Windows, that will create file whose name consists of the byte BB.

In Windows, that will create file whose name is the result of decode( "cp".Win32::GetACP(), chr( 0xBB ) ).

So what should I do -- always encode args to open()?

File names in unix are just a sequence of a bytes (which may not contain bytes 00 and 2F). It's best if they're text encoded using the current locale, but you could obtain file names which aren't.

For creating a file? encode returns a string using the downgraded/8bit/UTF8=0 internal storage format, so it won't be mangled by open.