add infrastructure of test for exotic sub names

This is scaffolding to sanity-check our assumptions about perl: that
anything we *can* eval is named by the interpreter as expected. This
test passes on 5.6.1+ on both 32 and 64 bit perls.

diff --git a/t/exotic_names.t b/t/exotic_names.t
new file mode 100644 (file)
index 0000000..00a8b46
--- /dev/null
+++ b/t/exotic_names.t
@@ -0,0 +1,91 @@
+use strict;
+use warnings;
+
+use Test::More;
+use B 'svref_2object';
+
+# This is a mess. The stash can supposedly handle Unicode but the behavior
+# is literally undefined before 5.16 (with crashes beyond the basic plane),
+# and remains unclear past 5.16 with evalbytes and feature unicode_eval
+# In any case - Sub::Name needs to *somehow* work with this, so we will do
+# a heuristic with ambiguous eval and looking for octets in the stash
+use if $] >= 5.016, feature => 'unicode_eval';
+
+sub compile_named_sub {
+    my ( $fullname, $body ) = @_;
+    my $sub = eval "sub $fullname { $body }" . '\\&{$fullname}';
+    return $sub if $sub;
+    my $e = $@;
+    require Carp;
+    Carp::croak $e;
+}
+
+sub caller3_ok {
+    my ( $sub, $expected, $type, $ord ) = @_;
+
+    local $Test::Builder::Level = $Test::Builder::Level + 1;
+
+    my $for_what = sprintf "when it contains \\x%s ( %s )", (
+        ( ($ord > 255)
+            ? sprintf "{%X}", $ord
+            : sprintf "%02X", $ord
+        ),
+        (
+            $ord > 255                    ? unpack('H*', pack 'C0U', $ord )
+            : ($ord > 0x1f and $ord < 0x7f) ? sprintf "%c", $ord
+            :                                 sprintf '\%o', $ord
+        ),
+    );
+
+    $expected =~ s/'/::/g;
+
+    # this is apparently how things worked before 5.16
+    utf8::encode($expected) if $] < 5.016 and $ord > 255;
+
+    my $stash_name = join '::', map { $_->STASH->NAME, $_->NAME } svref_2object($sub)->GV;
+
+    is $stash_name, $expected, "stash name for $type is correct $for_what";
+    is $sub->(), $expected, "caller() in $type returns correct name $for_what";
+}
+
+#######################################################################
+
+my @ordinal = ( 1 .. 255 );
+
+# 5.14 is the first perl to start properly handling \0 in identifiers
+unshift @ordinal, 0
+    unless $] < 5.014;
+
+# Unicode in 5.6 is not sane (crashes etc)
+push @ordinal,
+    0x100,    # LATIN CAPITAL LETTER A WITH MACRON
+    0x498,    # CYRILLIC CAPITAL LETTER ZE WITH DESCENDER
+    0x2122,   # TRADE MARK SIGN
+    0x1f4a9,  # PILE OF POO
+    unless $] < 5.008;
+
+plan tests => @ordinal * 2;
+
+my $legal_ident_char = "A-Z_a-z0-9'";
+$legal_ident_char .= join '', map chr, 0x100, 0x498
+    unless $] < 5.008;
+
+for my $ord (@ordinal) {
+    my $sub;
+    my $pkg      = sprintf 'test::SOME_%c_STASH', $ord;
+    my $subname  = sprintf 'SOME_%c_NAME', $ord;
+    my $fullname = join '::', $pkg, $subname;
+
+    # test that we can *always* compile at least within the correct package
+    my $expected;
+    if ( chr($ord) =~ m/^[$legal_ident_char]$/o ) { # compile directly
+        $expected = $fullname;
+        $sub = compile_named_sub $fullname => '(caller(0))[3]';
+    }
+    else { # not a legal identifier but at least test the package name by aliasing
+        $expected = "${pkg}::foo";
+        { no strict 'refs'; *palatable:: = *{"${pkg}::"} } # now palatable:: literally means ${pkg}::
+        $sub = compile_named_sub 'palatable::foo' => '(caller(0))[3]';
+    }
+    caller3_ok $sub, $expected, 'natively compiled sub', $ord;
+}