As answered in the CB, use utf8::upgrade on the string.

The behaviour of some Perl ops currently depends on the internal encoding of the string. utf8::upgrade and utf8::downgrade alter the internal encoding of the string.

\u and \l are implemented in terms of uc and lc, which are susceptible to this limitation/bug.

For example,

$ perl -le'use open ":std", ":locale"; $_="\xE0 la plage"; utf8::downg
+rade($_); print "\u$_"'
à la plage

$ perl -le'use open ":std", ":locale"; $_="\xE0 la plage"; utf8::upgra
+de($_); print "\u$_"'
À la plage
[download]

I've got use UTF8;

I hope you mean use utf8;, which simply tells Perl the source code that contains it is encoded using UTF-8 (not iso-latin-1). It doesn't affect IO.

I have lots of name in a UTF8 text file.

Did you decode the contents back into character? One way:

open(my $fh, '<:encoding(UTF-8)', $qfn)
    or die("Can't open file $qfn: $!\n");
[download]

Don't forget to encoding on the way out.

Undefined subroutine &main::setlocale called

setlocale is from POSIX. Did you actually load the POSIX module and import setlocale from it?

Update: Added example.