#!./perl -w BEGIN { chdir 't' if -d 't'; @INC = '../lib'; require './test.pl'; } # This is truth in an if statement, and could be a skip message my $no_endianness = $] > 5.009 ? '' : "Endianness pack modifiers not available on this perl"; my $no_signedness = $] > 5.009 ? '' : "Signed/unsigned pack modifiers not available on this perl"; plan tests => 14696; use strict; use warnings qw(FATAL all); use Config; my $Is_EBCDIC = (defined $Config{ebcdic} && $Config{ebcdic} eq 'define'); my $Perl = which_perl(); my @valid_errors = (qr/^Invalid type '\w'/); my $ByteOrder = 'unknown'; my $maybe_not_avail = '(?:hto[bl]e|[bl]etoh)'; if ($no_endianness) { push @valid_errors, qr/^Invalid type '[<>]'/; } elsif ($Config{byteorder} =~ /^1234(?:5678)?$/) { $ByteOrder = 'little'; $maybe_not_avail = '(?:htobe|betoh)'; } elsif ($Config{byteorder} =~ /^(?:8765)?4321$/) { $ByteOrder = 'big'; $maybe_not_avail = '(?:htole|letoh)'; } else { push @valid_errors, qr/^Can't (?:un)?pack (?:big|little)-endian .*? on this platform/; } if ($no_signedness) { push @valid_errors, qr/^'!' allowed only after types sSiIlLxX in (?:un)?pack/; } for my $size ( 16, 32, 64 ) { if (defined $Config{"u${size}size"} and $Config{"u${size}size"} != ($size >> 3)) { push @valid_errors, qr/^Perl_my_$maybe_not_avail$size\(\) not available/; } } my $IsTwosComplement = pack('i', -1) eq "\xFF" x $Config{intsize}; print "# \$IsTwosComplement = $IsTwosComplement\n"; sub is_valid_error { my $err = shift; for my $e (@valid_errors) { $err =~ $e and return 1; } return 0; } sub encode_list { my @result = map {_qq($_)} @_; if (@result == 1) { return @result; } return '(' . join (', ', @result) . ')'; } sub list_eq ($$) { my ($l, $r) = @_; return 0 unless @$l == @$r; for my $i (0..$#$l) { if (defined $l->[$i]) { return 0 unless defined ($r->[$i]) && $l->[$i] eq $r->[$i]; } else { return 0 if defined $r->[$i] } } return 1; } ############################################################################## # # Here starteth the tests # { my $format = "c2 x5 C C x s d i l a6"; # Need the expression in here to force ary[5] to be numeric. This avoids # test2 failing because ary2 goes str->numeric->str and ary doesn't. my @ary = (1,-100,127,128,32767,987.654321098 / 100.0,12345,123456, "abcdef"); my $foo = pack($format,@ary); my @ary2 = unpack($format,$foo); is($#ary, $#ary2); my $out1=join(':',@ary); my $out2=join(':',@ary2); # Using long double NVs may introduce greater accuracy than wanted. $out1 =~ s/:9\.87654321097999\d*:/:9.87654321098:/; $out2 =~ s/:9\.87654321097999\d*:/:9.87654321098:/; is($out1, $out2); like($foo, qr/def/); } # How about counting bits? { my $x; is( ($x = unpack("%32B*", "\001\002\004\010\020\040\100\200\377")), 16 ); is( ($x = unpack("%32b69", "\001\002\004\010\020\040\100\200\017")), 12 ); is( ($x = unpack("%32B69", "\001\002\004\010\020\040\100\200\017")), 9 ); } { my $sum = 129; # ASCII $sum = 103 if $Is_EBCDIC; my $x; is( ($x = unpack("%32B*", "Now is the time for all good blurfl")), $sum ); my $foo; open(BIN, $Perl) || die "Can't open $Perl: $!\n"; binmode BIN; sysread BIN, $foo, 8192; close BIN; $sum = unpack("%32b*", $foo); my $longway = unpack("b*", $foo); is( $sum, $longway =~ tr/1/1/ ); } { my $x; is( ($x = unpack("I",pack("I", 0xFFFFFFFF))), 0xFFFFFFFF ); } { # check 'w' my @x = (5,130,256,560,32000,3097152,268435455,1073741844, 2**33, '4503599627365785','23728385234614992549757750638446'); my $x = pack('w*', @x); my $y = pack 'H*', '0581028200843081fa0081bd8440ffffff7f8480808014A0808'. '0800087ffffffffffdb19caefe8e1eeeea0c2e1e3e8ede1ee6e'; is($x, $y); my @y = unpack('w*', $y); my $a; while ($a = pop @x) { my $b = pop @y; is($a, $b); } @y = unpack('w2', $x); is(scalar(@y), 2); is($y[1], 130); $x = pack('w*', 5000000000); $y = ''; eval { use Math::BigInt; $y = pack('w*', Math::BigInt::->new(5000000000)); }; is($x, $y); $x = pack 'w', ~0; $y = pack 'w', (~0).''; is($x, $y); is(unpack ('w',$x), ~0); is(unpack ('w',$y), ~0); $x = pack 'w', ~0 - 1; $y = pack 'w', (~0) - 2; if (~0 - 1 == (~0) - 2) { is($x, $y, "NV arithmetic"); } else { isnt($x, $y, "IV/NV arithmetic"); } cmp_ok(unpack ('w',$x), '==', ~0 - 1); cmp_ok(unpack ('w',$y), '==', ~0 - 2); # These should spot that pack 'w' is using NV, not double, on platforms # where IVs are smaller than doubles, and harmlessly pass elsewhere. # (tests for change 16861) my $x0 = 2**54+3; my $y0 = 2**54-2; $x = pack 'w', $x0; $y = pack 'w', $y0; if ($x0 == $y0) { is($x, $y, "NV arithmetic"); } else { isnt($x, $y, "IV/NV arithmetic"); } cmp_ok(unpack ('w',$x), '==', $x0); cmp_ok(unpack ('w',$y), '==', $y0); } { print "# test exceptions\n"; my $x; eval { $x = unpack 'w', pack 'C*', 0xff, 0xff}; like($@, qr/^Unterminated compressed integer/); eval { $x = unpack 'w', pack 'C*', 0xff, 0xff, 0xff, 0xff}; like($@, qr/^Unterminated compressed integer/); eval { $x = unpack 'w', pack 'C*', 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; like($@, qr/^Unterminated compressed integer/); eval { $x = pack 'w', -1 }; like ($@, qr/^Cannot compress negative numbers/); eval { $x = pack 'w', '1'x(1 + length ~0) . 'e0' }; like ($@, qr/^Can only compress unsigned integers/); # Check that the warning behaviour on the modifiers !, < and > is as we # expect it for this perl. my $can_endian = $no_endianness ? '' : 'sSiIlLqQjJfFdDpP'; my $can_shriek = 'sSiIlL'; $can_shriek .= 'nNvV' unless $no_signedness; # h and H can't do either, so act as sanity checks in blead foreach my $base (split '', 'hHsSiIlLqQjJfFdDpPnNvV') { foreach my $mod ('', '<', '>', '!', '!', '!<', '!>') { SKIP: { # Avoid void context warnings. my $a = eval {pack "$base$mod"}; skip "pack can't $base", 1 if $@ =~ /^Invalid type '\w'/; # Which error you get when 2 would be possible seems to be emergent # behaviour of pack's format parser. my $fails_shriek = $mod =~ /!/ && index ($can_shriek, $base) == -1; my $fails_endian = $mod =~ /[<>]/ && index ($can_endian, $base) == -1; my $shriek_first = $mod =~ /^!/; if ($no_endianness and ($mod eq '!')) { # The ! isn't seem as part of $base. Instead it's seen as a modifier # on > or < $fails_shriek = 1; undef $fails_endian; } elsif ($fails_shriek and $fails_endian) { if ($shriek_first) { undef $fails_endian; } } if ($fails_endian) { if ($no_endianness) { # < and > are seen as pattern letters, not modifiers like ($@, qr/^Invalid type '[<>]'/, "pack can't $base$mod"); } else { like ($@, qr/^'[<>]' allowed only after types/, "pack can't $base$mod"); } } elsif ($fails_shriek) { like ($@, qr/^'!' allowed only after types/, "pack can't $base$mod"); } else { is ($@, '', "pack can $base$mod"); } } } } SKIP: { skip $no_endianness, 2*3 + 2*8 if $no_endianness; for my $mod (qw( ! < > )) { eval { $x = pack "a$mod", 42 }; like ($@, qr/^'$mod' allowed only after types \S+ in pack/); eval { $x = unpack "a$mod", 'x'x8 }; like ($@, qr/^'$mod' allowed only after types \S+ in unpack/); } for my $mod (qw( <> >< !<> !>< >!< <>! >' after type 'I' in pack/); eval { $x = unpack "sI${mod}s", 'x'x16 }; like ($@, qr/^Can't use both '<' and '>' after type 'I' in unpack/); } } SKIP: { # Is this a stupid thing to do on VMS, VOS and other unusual platforms? skip("-- the IEEE infinity model is unavailable in this configuration.", 1) if (($^O eq 'VMS') && !defined($Config{useieee})); skip("-- $^O has serious fp indigestion on w-packed infinities", 1) if ( ($^O eq 'mpeix') || ($^O eq 'ultrix') || ($^O =~ /^svr4/ && -f "/etc/issue" && -f "/etc/.relid") # NCR MP-RAS ); my $inf = eval '2**1000000'; skip("Couldn't generate infinity - got error '$@'", 1) unless defined $inf and $inf == $inf / 2 and $inf + 1 == $inf; local our $TODO; $TODO = "VOS needs a fix for posix-1022 to pass this test." if ($^O eq 'vos'); eval { $x = pack 'w', $inf }; like ($@, qr/^Cannot compress integer/, "Cannot compress integer"); } SKIP: { skip("-- the full range of an IEEE double may not be available in this configuration.", 3) if (($^O eq 'VMS') && !defined($Config{useieee})); skip("-- $^O does not like 2**1023", 3) if (($^O eq 'ultrix')); # This should be about the biggest thing possible on an IEEE double my $big = eval '2**1023'; skip("Couldn't generate 2**1023 - got error '$@'", 3) unless defined $big and $big != $big / 2; eval { $x = pack 'w', $big }; is ($@, '', "Should be able to pack 'w', $big # 2**1023"); my $y = eval {unpack 'w', $x}; is ($@, '', "Should be able to unpack 'w' the result of pack 'w', $big # 2**1023"); # I'm getting about 1e-16 on FreeBSD my $quotient = int (100 * ($y - $big) / $big); ok($quotient < 2 && $quotient > -2, "Round trip pack, unpack 'w' of $big is within 1% ($quotient%)"); } } print "# test the 'p' template\n"; # literals is(unpack("p",pack("p","foo")), "foo"); SKIP: { skip $no_endianness, 2 if $no_endianness; is(unpack("p<",pack("p<","foo")), "foo"); is(unpack("p>",pack("p>","foo")), "foo"); } # scalars is(unpack("p",pack("p",239)), 239); SKIP: { skip $no_endianness, 2 if $no_endianness; is(unpack("p<",pack("p<",239)), 239); is(unpack("p>",pack("p>",239)), 239); } # temps sub foo { my $a = "a"; return $a . $a++ . $a++ } { use warnings qw(NONFATAL all);; my $warning; local $SIG{__WARN__} = sub { $warning = $_[0]; }; my $junk = pack("p", &foo); like($warning, qr/temporary val/); } # undef should give null pointer like(pack("p", undef), qr/^\0+$/); SKIP: { skip $no_endianness, 2 if $no_endianness; like(pack("p<", undef), qr/^\0+$/); like(pack("p>", undef), qr/^\0+$/); } # Check for optimizer bug (e.g. Digital Unix GEM cc with -O4 on DU V4.0B gives # 4294967295 instead of -1) # see #ifdef __osf__ in pp.c pp_unpack is((unpack("i",pack("i",-1))), -1); print "# test the pack lengths of s S i I l L n N v V + modifiers\n"; my @lengths = ( qw(s 2 S 2 i -4 I -4 l 4 L 4 n 2 N 4 v 2 V 4 n! 2 N! 4 v! 2 V! 4), 's!' => $Config{shortsize}, 'S!' => $Config{shortsize}, 'i!' => $Config{intsize}, 'I!' => $Config{intsize}, 'l!' => $Config{longsize}, 'L!' => $Config{longsize}, ); while (my ($base, $expect) = splice @lengths, 0, 2) { my @formats = ($base); $base =~ /^[nv]/i or push @formats, "$base>", "$base<"; for my $format (@formats) { SKIP: { skip $no_endianness, 1 if $no_endianness && $format =~ m/[<>]/; skip $no_signedness, 1 if $no_signedness && $format =~ /[nNvV]!/; my $len = length(pack($format, 0)); if ($expect > 0) { is($expect, $len, "format '$format'"); } else { $expect = -$expect; ok ($len >= $expect, "format '$format'") || print "# format '$format' has length $len, expected >= $expect\n"; } } } } print "# test unpack-pack lengths\n"; my @templates = qw(c C W i I s S l L n N v V f d q Q); foreach my $base (@templates) { my @tmpl = ($base); $base =~ /^[cwnv]/i or push @tmpl, "$base>", "$base<"; foreach my $t (@tmpl) { SKIP: { my @t = eval { unpack("$t*", pack("$t*", 12, 34)) }; skip "cannot pack '$t' on this perl", 4 if is_valid_error($@); is( $@, '', "Template $t works"); is(scalar @t, 2); is($t[0], 12); is($t[1], 34); } } } { # uuencode/decode # Note that first uuencoding known 'text' data and then checking the # binary values of the uuencoded version would not be portable between # character sets. Uuencoding is meant for encoding binary data, not # text data. my $in = pack 'C*', 0 .. 255; # just to be anal, we do some random tr/`/ / my $uu = <<'EOUU'; M` $"`P0%!@<("0H+# T.#Q`1$A,4%187&!D:&QP='A\@(2(C)"4F)R@I*BLL M+2XO,#$R,S0U-C'EZ>WQ]?G^`@8*#A(6& MAXB)BHN,C8Z/D)&2DY25EI>8F9J;G)V>GZ"AHJ.DI::GJ*FJJZRMKJ^PL;*S MM+6VM[BYNKN\O;Z_P,'"P\3%QL?(R+CY.7FY^CIZNOL[>[O\/'R\_3U]O?X^?K[_/W^_P ` EOUU $_ = $uu; tr/ /`/; is(pack('u', $in), $_); is(unpack('u', $uu), $in); $in = "\x1f\x8b\x08\x08\x58\xdc\xc4\x35\x02\x03\x4a\x41\x50\x55\x00\xf3\x2a\x2d\x2e\x51\x48\xcc\xcb\x2f\xc9\x48\x2d\x52\x08\x48\x2d\xca\x51\x28\x2d\x4d\xce\x4f\x49\x2d\xe2\x02\x00\x64\x66\x60\x5c\x1a\x00\x00\x00"; $uu = <<'EOUU'; M'XL("%C("`&1F &8%P:```` EOUU is(unpack('u', $uu), $in); # This is identical to the above except that backquotes have been # changed to spaces $uu = <<'EOUU'; M'XL("%C(" &1F &8%P: EOUU # ' # Grr is(unpack('u', $uu), $in); } # test the ascii template types (A, a, Z) foreach ( ['p', 'A*', "foo\0bar\0 ", "foo\0bar\0 "], ['p', 'A11', "foo\0bar\0 ", "foo\0bar\0 "], ['u', 'A*', "foo\0bar \0", "foo\0bar"], ['u', 'A8', "foo\0bar \0", "foo\0bar"], ['p', 'a*', "foo\0bar\0 ", "foo\0bar\0 "], ['p', 'a11', "foo\0bar\0 ", "foo\0bar\0 \0\0"], ['u', 'a*', "foo\0bar \0", "foo\0bar \0"], ['u', 'a8', "foo\0bar \0", "foo\0bar "], ['p', 'Z*', "foo\0bar\0 ", "foo\0bar\0 \0"], ['p', 'Z11', "foo\0bar\0 ", "foo\0bar\0 \0\0"], ['p', 'Z3', "foo", "fo\0"], ['u', 'Z*', "foo\0bar \0", "foo"], ['u', 'Z8', "foo\0bar \0", "foo"], ) { my ($what, $template, $in, $out) = @$_; my $got = $what eq 'u' ? (unpack $template, $in) : (pack $template, $in); unless (is($got, $out)) { my $un = $what eq 'u' ? 'un' : ''; print "# ${un}pack ('$template', "._qq($in).') gave '._qq($out). ' not '._qq($got)."\n"; } } print "# packing native shorts/ints/longs\n"; is(length(pack("s!", 0)), $Config{shortsize}); is(length(pack("i!", 0)), $Config{intsize}); is(length(pack("l!", 0)), $Config{longsize}); ok(length(pack("s!", 0)) <= length(pack("i!", 0))); ok(length(pack("i!", 0)) <= length(pack("l!", 0))); is(length(pack("i!", 0)), length(pack("i", 0))); sub numbers { my $base = shift; my @formats = ($base); $base =~ /^[silqjfdp]/i and push @formats, "$base>", "$base<"; for my $format (@formats) { numbers_with_total ($format, undef, @_); } } sub numbers_with_total { my $format = shift; my $total = shift; if (!defined $total) { foreach (@_) { $total += $_; } } print "# numbers test for $format\n"; foreach (@_) { SKIP: { my $out = eval {unpack($format, pack($format, $_))}; skip "cannot pack '$format' on this perl", 2 if is_valid_error($@); is($@, '', "no error"); is($out, $_, "unpack pack $format $_"); } } my $skip_if_longer_than = ~0; # "Infinity" if (~0 - 1 == ~0) { # If we're running with -DNO_PERLPRESERVE_IVUV and NVs don't preserve all # UVs (in which case ~0 is NV, ~0-1 will be the same NV) then we can't # correctly in perl calculate UV totals for long checksums, as pp_unpack # is using UV maths, and we've only got NVs. $skip_if_longer_than = $Config{nv_preserves_uv_bits}; } foreach ('', 1, 2, 3, 15, 16, 17, 31, 32, 33, 53, 54, 63, 64, 65) { SKIP: { my $sum = eval {unpack "%$_$format*", pack "$format*", @_}; skip "cannot pack '$format' on this perl", 3 if is_valid_error($@); is($@, '', "no error"); ok(defined $sum, "sum bits $_, format $format defined"); my $len = $_; # Copy, so that we can reassign '' $len = 16 unless length $len; SKIP: { skip "cannot test checksums over $skip_if_longer_than bits", 1 if $len > $skip_if_longer_than; # Our problem with testing this portably is that the checksum code in # pp_unpack is able to cast signed to unsigned, and do modulo 2**n # arithmetic in unsigned ints, which perl has no operators to do. # (use integer; does signed ints, which won't wrap on UTS, which is just # fine with ANSI, but not with most people's assumptions. # This is why we need to supply the totals for 'Q' as there's no way in # perl to calculate them, short of unpack '%0Q' (is that documented?) # ** returns NVs; make sure it's IV. my $max = 1 + 2 * (int (2 ** ($len-1))-1); # The max possible checksum my $max_p1 = $max + 1; my ($max_is_integer, $max_p1_is_integer); $max_p1_is_integer = 1 unless $max_p1 + 1 == $max_p1; $max_is_integer = 1 if $max - 1 < ~0; my $calc_sum; if (ref $total) { $calc_sum = &$total($len); } else { $calc_sum = $total; # Shift into range by some multiple of the total my $mult = $max_p1 ? int ($total / $max_p1) : undef; # Need this to make sure that -1 + (~0+1) is ~0 (ie still integer) $calc_sum = $total - $mult; $calc_sum -= $mult * $max; if ($calc_sum < 0) { $calc_sum += 1; $calc_sum += $max; } } if ($calc_sum == $calc_sum - 1 && $calc_sum == $max_p1) { # we're into floating point (either by getting out of the range of # UV arithmetic, or because we're doing a floating point checksum) # and our calculation of the checksum has become rounded up to # max_checksum + 1 $calc_sum = 0; } if ($calc_sum == $sum) { # HAS to be ==, not eq (so no is()). pass ("unpack '%$_$format' gave $sum"); } else { my $delta = 1.000001; if ($format =~ tr /dDfF// && ($calc_sum <= $sum * $delta && $calc_sum >= $sum / $delta)) { pass ("unpack '%$_$format' gave $sum, expected $calc_sum"); } else { my $text = ref $total ? &$total($len) : $total; fail; print "# For list (" . join (", ", @_) . ") (total $text)" . " packed with $format unpack '%$_$format' gave $sum," . " expected $calc_sum\n"; } } } } } } numbers ('c', -128, -1, 0, 1, 127); numbers ('C', 0, 1, 127, 128, 255); numbers ('W', 0, 1, 127, 128, 255, 256, 0x7ff, 0x800, 0xfffd); numbers ('s', -32768, -1, 0, 1, 32767); numbers ('S', 0, 1, 32767, 32768, 65535); numbers ('i', -2147483648, -1, 0, 1, 2147483647); numbers ('I', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('l', -2147483648, -1, 0, 1, 2147483647); numbers ('L', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('s!', -32768, -1, 0, 1, 32767); numbers ('S!', 0, 1, 32767, 32768, 65535); numbers ('i!', -2147483648, -1, 0, 1, 2147483647); numbers ('I!', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('l!', -2147483648, -1, 0, 1, 2147483647); numbers ('L!', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('n', 0, 1, 32767, 32768, 65535); numbers ('v', 0, 1, 32767, 32768, 65535); numbers ('N', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('V', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('n!', -32768, -1, 0, 1, 32767); numbers ('v!', -32768, -1, 0, 1, 32767); numbers ('N!', -2147483648, -1, 0, 1, 2147483647); numbers ('V!', -2147483648, -1, 0, 1, 2147483647); # All these should have exact binary representations: numbers ('f', -1, 0, 0.5, 42, 2**34); numbers ('d', -(2**34), -1, 0, 1, 2**34); ## These don't, but 'd' is NV. XXX wrong, it's double #numbers ('d', -1, 0, 1, 1-exp(-1), -exp(1)); numbers_with_total ('q', -1, -9223372036854775808, -1, 0, 1,9223372036854775807); # This total is icky, but the true total is 2**65-1, and need a way to generate # the epxected checksum on any system including those where NVs can preserve # 65 bits. (long double is 128 bits on sparc, so they certainly can) # or where rounding is down not up on binary conversion (crays) numbers_with_total ('Q', sub { my $len = shift; $len = 65 if $len > 65; # unmasked total is 2**65-1 here my $total = 1 + 2 * (int (2**($len - 1)) - 1); return 0 if $total == $total - 1; # Overflowed integers return $total; # NVs still accurate to nearest integer }, 0, 1,9223372036854775807, 9223372036854775808, 18446744073709551615); print "# pack nvNV byteorders\n"; is(pack("n", 0xdead), "\xde\xad"); is(pack("v", 0xdead), "\xad\xde"); is(pack("N", 0xdeadbeef), "\xde\xad\xbe\xef"); is(pack("V", 0xdeadbeef), "\xef\xbe\xad\xde"); SKIP: { skip $no_signedness, 4 if $no_signedness; is(pack("n!", 0xdead), "\xde\xad"); is(pack("v!", 0xdead), "\xad\xde"); is(pack("N!", 0xdeadbeef), "\xde\xad\xbe\xef"); is(pack("V!", 0xdeadbeef), "\xef\xbe\xad\xde"); } print "# test big-/little-endian conversion\n"; sub byteorder { my $format = shift; print "# byteorder test for $format\n"; for my $value (@_) { SKIP: { my($nat,$be,$le) = eval { map { pack $format.$_, $value } '', '>', '<' }; skip "cannot pack '$format' on this perl", 5 if is_valid_error($@); { use warnings qw(NONFATAL utf8); print "# [$value][$nat][$be][$le][$@]\n"; } SKIP: { skip "cannot compare native byteorder with big-/little-endian", 1 if $ByteOrder eq 'unknown'; is($nat, $ByteOrder eq 'big' ? $be : $le); } is($be, reverse($le)); my @x = eval { unpack "$format$format>$format<", $nat.$be.$le }; print "# [$value][", join('][', @x), "][$@]\n"; is($@, ''); is($x[0], $x[1]); is($x[0], $x[2]); } } } byteorder('s', -32768, -1, 0, 1, 32767); byteorder('S', 0, 1, 32767, 32768, 65535); byteorder('i', -2147483648, -1, 0, 1, 2147483647); byteorder('I', 0, 1, 2147483647, 2147483648, 4294967295); byteorder('l', -2147483648, -1, 0, 1, 2147483647); byteorder('L', 0, 1, 2147483647, 2147483648, 4294967295); byteorder('j', -2147483648, -1, 0, 1, 2147483647); byteorder('J', 0, 1, 2147483647, 2147483648, 4294967295); byteorder('s!', -32768, -1, 0, 1, 32767); byteorder('S!', 0, 1, 32767, 32768, 65535); byteorder('i!', -2147483648, -1, 0, 1, 2147483647); byteorder('I!', 0, 1, 2147483647, 2147483648, 4294967295); byteorder('l!', -2147483648, -1, 0, 1, 2147483647); byteorder('L!', 0, 1, 2147483647, 2147483648, 4294967295); byteorder('q', -9223372036854775808, -1, 0, 1, 9223372036854775807); byteorder('Q', 0, 1, 9223372036854775807, 9223372036854775808, 18446744073709551615); byteorder('f', -1, 0, 0.5, 42, 2**34); byteorder('F', -1, 0, 0.5, 42, 2**34); byteorder('d', -(2**34), -1, 0, 1, 2**34); byteorder('D', -(2**34), -1, 0, 1, 2**34); print "# test negative numbers\n"; SKIP: { skip "platform is not using two's complement for negative integers", 120 unless $IsTwosComplement; for my $format (qw(s i l j s! i! l! q)) { SKIP: { my($nat,$be,$le) = eval { map { pack $format.$_, -1 } '', '>', '<' }; skip "cannot pack '$format' on this perl", 15 if is_valid_error($@); my $len = length $nat; is($_, "\xFF"x$len) for $nat, $be, $le; my(@val,@ref); if ($len >= 8) { @val = (-2, -81985529216486896, -9223372036854775808); @ref = ("\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFE", "\xFE\xDC\xBA\x98\x76\x54\x32\x10", "\x80\x00\x00\x00\x00\x00\x00\x00"); } elsif ($len >= 4) { @val = (-2, -19088744, -2147483648); @ref = ("\xFF\xFF\xFF\xFE", "\xFE\xDC\xBA\x98", "\x80\x00\x00\x00"); } else { @val = (-2, -292, -32768); @ref = ("\xFF\xFE", "\xFE\xDC", "\x80\x00"); } for my $x (@ref) { if ($len > length $x) { $x = $x . "\xFF" x ($len - length $x); } } for my $i (0 .. $#val) { my($nat,$be,$le) = eval { map { pack $format.$_, $val[$i] } '', '>', '<' }; is($@, ''); SKIP: { skip "cannot compare native byteorder with big-/little-endian", 1 if $ByteOrder eq 'unknown'; is($nat, $ByteOrder eq 'big' ? $be : $le); } is($be, $ref[$i]); is($be, reverse($le)); } } } } { # / my ($x, $y, $z); eval { ($x) = unpack '/a*','hello' }; like($@, qr!'/' must follow a numeric type!); undef $x; eval { $x = unpack '/a*','hello' }; like($@, qr!'/' must follow a numeric type!); undef $x; eval { ($z,$x,$y) = unpack 'a3/A C/a* C/Z', "003ok \003yes\004z\000abc" }; is($@, ''); is($z, 'ok'); is($x, 'yes'); is($y, 'z'); undef $z; eval { $z = unpack 'a3/A C/a* C/Z', "003ok \003yes\004z\000abc" }; is($@, ''); is($z, 'ok'); undef $x; eval { ($x) = pack '/a*','hello' }; like($@, qr!Invalid type '/'!); undef $x; eval { $x = pack '/a*','hello' }; like($@, qr!Invalid type '/'!); $z = pack 'n/a* N/Z* w/A*','string','hi there ','etc'; my $expect = "\000\006string\0\0\0\012hi there \000\003etc"; is($z, $expect); undef $x; $expect = 'hello world'; eval { ($x) = unpack ("w/a", chr (11) . "hello world!")}; is($x, $expect); is($@, ''); undef $x; # Doing this in scalar context used to fail. eval { $x = unpack ("w/a", chr (11) . "hello world!")}; is($@, ''); is($x, $expect); foreach ( ['a/a*/a*', '212ab345678901234567','ab3456789012'], ['a/a*/a*', '3012ab345678901234567', 'ab3456789012'], ['a/a*/b*', '212ab', $Is_EBCDIC ? '100000010100' : '100001100100'], ) { my ($pat, $in, $expect) = @$_; undef $x; eval { ($x) = unpack $pat, $in }; is($@, ''); is($x, $expect) || printf "# list unpack ('$pat', '$in') gave %s, expected '$expect'\n", encode_list ($x); undef $x; eval { $x = unpack $pat, $in }; is($@, ''); is($x, $expect) || printf "# scalar unpack ('$pat', '$in') gave %s, expected '$expect'\n", encode_list ($x); } # / with # my $pattern = <<'EOU'; a3/A # Count in ASCII C/a* # Count in a C char C/Z # Count in a C char but skip after \0 EOU $x = $y = $z =undef; eval { ($z,$x,$y) = unpack $pattern, "003ok \003yes\004z\000abc" }; is($@, ''); is($z, 'ok'); is($x, 'yes'); is($y, 'z'); undef $x; eval { $z = unpack $pattern, "003ok \003yes\004z\000abc" }; is($@, ''); is($z, 'ok'); $pattern = <<'EOP'; n/a* # Count as network short w/A* # Count a BER integer EOP $expect = "\000\006string\003etc"; $z = pack $pattern,'string','etc'; is($z, $expect); } SKIP: { skip("(EBCDIC and) version strings are bad idea", 2) if $Is_EBCDIC; is("1.20.300.4000", sprintf "%vd", pack("U*",1,20,300,4000)); is("1.20.300.4000", sprintf "%vd", pack(" U*",1,20,300,4000)); } isnt(v1.20.300.4000, sprintf "%vd", pack("C0U*",1,20,300,4000)); my $rslt = $Is_EBCDIC ? "156 67" : "199 162"; is(join(" ", unpack("U0 C*", chr(0x1e2))), $rslt); # does pack U create Unicode? is(ord(pack('U', 300)), 300); # does unpack U deref Unicode? is((unpack('U', chr(300)))[0], 300); # is unpack U the reverse of pack U for Unicode string? is("@{[unpack('U*', pack('U*', 100, 200, 300))]}", "100 200 300"); # is unpack U the reverse of pack U for byte string? is("@{[unpack('U*', pack('U*', 100, 200))]}", "100 200"); SKIP: { skip "Not for EBCDIC", 4 if $Is_EBCDIC; # does pack U0C create Unicode? is("@{[pack('U0C*', 100, 195, 136)]}", v100.v200); # does pack C0U create characters? is("@{[pack('C0U*', 100, 200)]}", pack("C*", 100, 195, 136)); # does unpack U0U on byte data warn? { use warnings qw(NONFATAL all);; my $bad = pack("U0C", 255); local $SIG{__WARN__} = sub { $@ = "@_" }; my @null = unpack('U0U', $bad); like($@, qr/^Malformed UTF-8 character /); } } { my $p = pack 'i*', -2147483648, ~0, 0, 1, 2147483647; my (@a); # bug - % had to be at the start of the pattern, no leading whitespace or # comments. %i! didn't work at all. foreach my $pat ('%32i*', ' %32i*', "# Muhahahaha\n%32i*", '%32i* ', '%32i!*', ' %32i!*', "\n#\n#\n\r \t\f%32i!*", '%32i!*#') { @a = unpack $pat, $p; is($a[0], 0xFFFFFFFF) || print "# $pat\n"; @a = scalar unpack $pat, $p; is($a[0], 0xFFFFFFFF) || print "# $pat\n"; } $p = pack 'I*', 42, 12; # Multiline patterns in scalar context failed. foreach my $pat ('I', < 32 bits with floating # point, so a pathologically long pattern would wrap at 32 bits. my $pat = "\xff\xff"x65538; # Start with it long, to save any copying. foreach (4,3,2,1,0) { my $len = 65534 + $_; is(unpack ("%33n$len", $pat), 65535 * $len); } } # pack x X @ foreach ( ['x', "N", "\0"], ['x4', "N", "\0"x4], ['xX', "N", ""], ['xXa*', "Nick", "Nick"], ['a5Xa5', "cameL", "llama", "camellama"], ['@4', 'N', "\0"x4], ['a*@8a*', 'Camel', 'Dromedary', "Camel\0\0\0Dromedary"], ['a*@4a', 'Perl rules', '!', 'Perl!'], ) { my ($template, @in) = @$_; my $out = pop @in; my $got = eval {pack $template, @in}; is($@, ''); is($out, $got) || printf "# pack ('$template', %s) gave %s expected %s\n", encode_list (@in), encode_list ($got), encode_list ($out); } # unpack x X @ foreach ( ['x', "N"], ['xX', "N"], ['xXa*', "Nick", "Nick"], ['a5Xa5', "camellama", "camel", "llama"], ['@3', "ice"], ['@2a2', "water", "te"], ['a*@1a3', "steam", "steam", "tea"], ) { my ($template, $in, @out) = @$_; my @got = eval {unpack $template, $in}; is($@, ''); ok (list_eq (\@got, \@out)) || printf "# list unpack ('$template', %s) gave %s expected %s\n", _qq($in), encode_list (@got), encode_list (@out); my $got = eval {unpack $template, $in}; is($@, ''); @out ? is( $got, $out[0] ) # 1 or more items; should get first : ok( !defined $got ) # 0 items; should get undef or printf "# scalar unpack ('$template', %s) gave %s expected %s\n", _qq($in), encode_list ($got), encode_list ($out[0]); } { my $t = 'Z*Z*'; my ($u, $v) = qw(foo xyzzy); my $p = pack($t, $u, $v); my @u = unpack($t, $p); is(scalar @u, 2); is($u[0], $u); is($u[1], $v); } { is((unpack("w/a*", "\x02abc"))[0], "ab"); # "w/a*" should be seen as one unit is(scalar unpack("w/a*", "\x02abc"), "ab"); } SKIP: { print "# group modifiers\n"; skip $no_endianness, 3 * 2 + 3 * 2 + 1 if $no_endianness; for my $t (qw{ (s<)< (sl>s)> (s(l(sl) (sl>s)< (s(l(sl) }) { print "# testing pattern '$t'\n"; eval { ($_) = unpack($t, 'x'x18); }; like($@, qr/Can't use '[<>]' in a group with different byte-order in unpack/); eval { $_ = pack($t, (0)x6); }; like($@, qr/Can't use '[<>]' in a group with different byte-order in pack/); } is(pack('L', (0x12345678)x2), pack('(((L1)1)<)(((L)1)1)>1', (0x12345678)x2)); } { sub compress_template { my $t = shift; for my $mod (qw( < > )) { $t =~ s/((?:(?:[SILQJFDP]!?$mod|[^SILQJFDP\W]!?)(?:\d+|\*|\[(?:[^]]+)\])?\/?){2,})/ my $x = $1; $x =~ s!$mod!!g ? "($x)$mod" : $x /ieg; } return $t; } my %templates = ( 's<' => [-42], 's [-42, -11, 12, 4711], '(i [-11, -22, -33, 1000000, 1100, 2201, 3302, -1000000, 32767, -32768, 1, -123456789 ], '(I!<4(J<2L<)3)5' => [1 .. 65], 'q [-50000000005, 60000000006], 'f [3.14159, 111.11, 2222.22], 'D [1e42, -128, 255, 1e-42], 'n/a*' => ['/usr/bin/perl'], 'C/a*S [qw(Just another Perl hacker)], ); for my $tle (sort keys %templates) { my @d = @{$templates{$tle}}; my $tbe = $tle; $tbe =~ y//; for my $t ($tbe, $tle) { my $c = compress_template($t); print "# '$t' -> '$c'\n"; SKIP: { my $p1 = eval { pack $t, @d }; skip "cannot pack '$t' on this perl", 5 if is_valid_error($@); my $p2 = eval { pack $c, @d }; is($@, ''); is($p1, $p2); s!(/[aAZ])\*!$1!g for $t, $c; my @u1 = eval { unpack $t, $p1 }; is($@, ''); my @u2 = eval { unpack $c, $p2 }; is($@, ''); is(join('!', @u1), join('!', @u2)); } } } } { # from Wolfgang Laun: fix in change #13163 my $s = 'ABC' x 10; my $t = '*'; my $x = ord($t); my $buf = pack( 'Z*/A* C', $s, $x ); my $y; my $h = $buf; $h =~ s/[^[:print:]]/./g; ( $s, $y ) = unpack( "Z*/A* C", $buf ); is($h, "30.ABCABCABCABCABCABCABCABCABCABC$t"); is(length $buf, 34); is($s, "ABCABCABCABCABCABCABCABCABCABC"); is($y, $x); } { # from Wolfgang Laun: fix in change #13288 eval { my $t=unpack("P*", "abc") }; like($@, qr/'P' must have an explicit size/); } { # Grouping constructs my (@a, @b); @a = unpack '(SL)', pack 'SLSLSL', 67..90; is("@a", "67 68"); @a = unpack '(SL)3', pack 'SLSLSL', 67..90; @b = (67..72); is("@a", "@b"); @a = unpack '(SL)3', pack 'SLSLSLSL', 67..90; is("@a", "@b"); @a = unpack '(SL)[3]', pack 'SLSLSLSL', 67..90; is("@a", "@b"); @a = unpack '(SL)[2] SL', pack 'SLSLSLSL', 67..90; is("@a", "@b"); @a = unpack 'A/(SL)', pack 'ASLSLSLSL', 3, 67..90; is("@a", "@b"); @a = unpack 'A/(SL)SL', pack 'ASLSLSLSL', 2, 67..90; is("@a", "@b"); @a = unpack '(SL)*', pack 'SLSLSLSL', 67..90; @b = (67..74); is("@a", "@b"); @a = unpack '(SL)*SL', pack 'SLSLSLSL', 67..90; is("@a", "@b"); eval { @a = unpack '(*SL)', '' }; like($@, qr/\(\)-group starts with a count/); eval { @a = unpack '(3SL)', '' }; like($@, qr/\(\)-group starts with a count/); eval { @a = unpack '([3]SL)', '' }; like($@, qr/\(\)-group starts with a count/); eval { @a = pack '(*SL)' }; like($@, qr/\(\)-group starts with a count/); @a = unpack '(SL)3 SL', pack '(SL)4', 67..74; is("@a", "@b"); @a = unpack '(SL)3 SL', pack '(SL)[4]', 67..74; is("@a", "@b"); @a = unpack '(SL)3 SL', pack '(SL)*', 67..74; is("@a", "@b"); } { # more on grouping (W.Laun) # @ absolute within ()-group my $badc = pack( '(a)*', unpack( '(@1a @0a @2)*', 'abcd' ) ); is( $badc, 'badc' ); my @b = ( 1, 2, 3 ); my $buf = pack( '(@1c)((@2C)@3c)', @b ); is( $buf, "\0\1\0\0\2\3" ); my @a = unpack( '(@1c)((@2c)@3c)', $buf ); is( "@a", "@b" ); # various unpack count/code scenarios my @Env = ( a => 'AAA', b => 'BBB' ); my $env = pack( 'S(S/A*S/A*)*', @Env/2, @Env ); # unpack full length - ok my @pup = unpack( 'S/(S/A* S/A*)', $env ); is( "@pup", "@Env" ); # warn when count/code goes beyond end of string # \0002 \0001 a \0003 AAA \0001 b \0003 BBB # 2 4 5 7 10 1213 eval { @pup = unpack( 'S/(S/A* S/A*)', substr( $env, 0, 13 ) ) }; like( $@, qr{length/code after end of string} ); # postfix repeat count $env = pack( '(S/A* S/A*)' . @Env/2, @Env ); # warn when count/code goes beyond end of string # \0001 a \0003 AAA \0001 b \0003 BBB # 2 3c 5 8 10 11 13 16 eval { @pup = unpack( '(S/A* S/A*)' . @Env/2, substr( $env, 0, 11 ) ) }; like( $@, qr{length/code after end of string} ); # catch stack overflow/segfault eval { $_ = pack( ('(' x 105) . 'A' . (')' x 105) ); }; like( $@, qr{Too deeply nested \(\)-groups} ); } { # syntax checks (W.Laun) use warnings qw(NONFATAL all);; my @warning; local $SIG{__WARN__} = sub { push( @warning, $_[0] ); }; eval { my $s = pack( 'Ax![4c]A', 1..5 ); }; like( $@, qr{Malformed integer in \[\]} ); eval { my $buf = pack( '(c/*a*)', 'AAA', 'BB' ); }; like( $@, qr{'/' does not take a repeat count} ); eval { my @inf = unpack( 'c/1a', "\x03AAA\x02BB" ); }; like( $@, qr{'/' does not take a repeat count} ); eval { my @inf = unpack( 'c/*a', "\x03AAA\x02BB" ); }; like( $@, qr{'/' does not take a repeat count} ); # white space where possible my @Env = ( a => 'AAA', b => 'BBB' ); my $env = pack( ' S ( S / A* S / A* )* ', @Env/2, @Env ); my @pup = unpack( ' S / ( S / A* S / A* ) ', $env ); is( "@pup", "@Env" ); # white space in 4 wrong places for my $temp ( 'A ![4]', 'A [4]', 'A *', 'A 4' ){ eval { my $s = pack( $temp, 'B' ); }; like( $@, qr{Invalid type } ); } # warning for commas @warning = (); my $x = pack( 'I,A', 4, 'X' ); like( $warning[0], qr{Invalid type ','} ); # comma warning only once @warning = (); $x = pack( 'C(C,C)C,C', 65..71 ); like( scalar @warning, 1 ); # forbidden code in [] eval { my $x = pack( 'A[@4]', 'XXXX' ); }; like( $@, qr{Within \[\]-length '\@' not allowed} ); # @ repeat default 1 my $s = pack( 'AA@A', 'A', 'B', 'C' ); my @c = unpack( 'AA@A', $s ); is( $s, 'AC' ); is( "@c", "A C C" ); # no unpack code after / eval { my @a = unpack( "C/", "\3" ); }; like( $@, qr{Code missing after '/'} ); SKIP: { skip $no_endianness, 6 if $no_endianness; # modifier warnings @warning = (); $x = pack "I>>s!!", 47, 11; ($x) = unpack "I<!>", 'x'x20; is(scalar @warning, 5); like($warning[0], qr/Duplicate modifier '>' after 'I' in pack/); like($warning[1], qr/Duplicate modifier '!' after 's' in pack/); like($warning[2], qr/Duplicate modifier '<' after 'I' in unpack/); like($warning[3], qr/Duplicate modifier '!' after 'l' in unpack/); like($warning[4], qr/Duplicate modifier '>' after 'l' in unpack/); } } { # Repeat count [SUBEXPR] my @codes = qw( x A Z a c C W B b H h s v n S i I l V N L p P f F d s! S! i! I! l! L! j J); my $G; if (eval { pack 'q', 1 } ) { push @codes, qw(q Q); } else { push @codes, qw(s S); # Keep the count the same } if (eval { pack 'D', 1 } ) { push @codes, 'D'; } else { push @codes, 'd'; # Keep the count the same } push @codes, map { /^[silqjfdp]/i ? ("$_<", "$_>") : () } @codes; my %val; @val{@codes} = map { / [Xx] (?{ undef }) | [AZa] (?{ 'something' }) | C (?{ 214 }) | W (?{ 8188 }) | c (?{ 114 }) | [Bb] (?{ '101' }) | [Hh] (?{ 'b8' }) | [svnSiIlVNLqQjJ] (?{ 10111 }) | [FfDd] (?{ 1.36514538e67 }) | [pP] (?{ "try this buffer" }) /x; $^R } @codes; my @end = (0x12345678, 0x23456781, 0x35465768, 0x15263748); my $end = "N4"; for my $type (@codes) { my @list = $val{$type}; @list = () unless defined $list[0]; for my $count ('', '3', '[11]') { my $c = 1; $c = $1 if $count =~ /(\d+)/; my @list1 = @list; @list1 = (@list1) x $c unless $type =~ /[XxAaZBbHhP]/; for my $groupend ('', ')2', ')[8]') { my $groupbegin = ($groupend ? '(' : ''); $c = 1; $c = $1 if $groupend =~ /(\d+)/; my @list2 = (@list1) x $c; SKIP: { my $junk1 = "$groupbegin $type$count $groupend"; # print "# junk1=$junk1\n"; my $p = eval { pack $junk1, @list2 }; skip "cannot pack '$type' on this perl", 12 if is_valid_error($@); die "pack $junk1 failed: $@" if $@; my $half = int( (length $p)/2 ); for my $move ('', "X$half", "X!$half", 'x1', 'x!8', "x$half") { my $junk = "$junk1 $move"; # print "# junk='$junk', list=(@list2)\n"; $p = pack "$junk $end", @list2, @end; my @l = unpack "x[$junk] $end", $p; is(scalar @l, scalar @end); is("@l", "@end", "skipping x[$junk]"); } } } } } } # / is recognized after spaces in scalar context # XXXX no spaces are allowed in pack... In pack only before the slash... is(scalar unpack('A /A Z20', pack 'A/A* Z20', 'bcde', 'xxxxx'), 'bcde'); is(scalar unpack('A /A /A Z20', '3004bcde'), 'bcde'); { # X! and x! my $t = 'C[3] x!8 C[2]'; my @a = (0x73..0x77); my $p = pack($t, @a); is($p, "\x73\x74\x75\0\0\0\0\0\x76\x77"); my @b = unpack $t, $p; is(scalar @b, scalar @a); is("@b", "@a", 'x!8'); $t = 'x[5] C[6] X!8 C[2]'; @a = (0x73..0x7a); $p = pack($t, @a); is($p, "\0\0\0\0\0\x73\x74\x75\x79\x7a"); @b = unpack $t, $p; @a = (0x73..0x75, 0x79, 0x7a, 0x79, 0x7a); is(scalar @b, scalar @a); is("@b", "@a"); } { # struct {char c1; double d; char cc[2];} my $t = 'C x![d] d C[2]'; my @a = (173, 1.283476517e-45, 42, 215); my $p = pack $t, @a; ok( length $p); my @b = unpack "$t X[$t] $t", $p; # Extract, step back, extract again is(scalar @b, 2 * scalar @a); $b = "@b"; $b =~ s/(?:17000+|16999+)\d+(e-45) /17$1 /gi; # stringification is gamble is($b, "@a @a"); use warnings qw(NONFATAL all);; my $warning; local $SIG{__WARN__} = sub { $warning = $_[0]; }; @b = unpack "x[C] x[$t] X[$t] X[C] $t", "$p\0"; is($warning, undef); is(scalar @b, scalar @a); $b = "@b"; $b =~ s/(?:17000+|16999+)\d+(e-45) /17$1 /gi; # stringification is gamble is($b, "@a"); } is(length(pack("j", 0)), $Config{ivsize}); is(length(pack("J", 0)), $Config{uvsize}); is(length(pack("F", 0)), $Config{nvsize}); numbers ('j', -2147483648, -1, 0, 1, 2147483647); numbers ('J', 0, 1, 2147483647, 2147483648, 4294967295); numbers ('F', -(2**34), -1, 0, 1, 2**34); SKIP: { my $t = eval { unpack("D*", pack("D", 12.34)) }; skip "Long doubles not in use", 166 if $@ =~ /Invalid type/; is(length(pack("D", 0)), $Config{longdblsize}); numbers ('D', -(2**34), -1, 0, 1, 2**34); } # Maybe this knowledge needs to be "global" for all of pack.t # Or a "can checksum" which would effectively be all the number types" my %cant_checksum = map {$_=> 1} qw(A Z u w); # not a b B h H foreach my $template (qw(A Z c C s S i I l L n N v V q Q j J f d F D u U w)) { SKIP: { my $packed = eval {pack "${template}4", 1, 4, 9, 16}; if ($@) { die unless $@ =~ /Invalid type '$template'/; skip ("$template not supported on this perl", $cant_checksum{$template} ? 4 : 8); } my @unpack4 = unpack "${template}4", $packed; my @unpack = unpack "${template}*", $packed; my @unpack1 = unpack "${template}", $packed; my @unpack1s = scalar unpack "${template}", $packed; my @unpack4s = scalar unpack "${template}4", $packed; my @unpacks = scalar unpack "${template}*", $packed; my @tests = ( ["${template}4 vs ${template}*", \@unpack4, \@unpack], ["scalar ${template} ${template}", \@unpack1s, \@unpack1], ["scalar ${template}4 vs ${template}", \@unpack4s, \@unpack1], ["scalar ${template}* vs ${template}", \@unpacks, \@unpack1], ); unless ($cant_checksum{$template}) { my @unpack4_c = unpack "\%${template}4", $packed; my @unpack_c = unpack "\%${template}*", $packed; my @unpack1_c = unpack "\%${template}", $packed; my @unpack1s_c = scalar unpack "\%${template}", $packed; my @unpack4s_c = scalar unpack "\%${template}4", $packed; my @unpacks_c = scalar unpack "\%${template}*", $packed; push @tests, ( ["% ${template}4 vs ${template}*", \@unpack4_c, \@unpack_c], ["% scalar ${template} ${template}", \@unpack1s_c, \@unpack1_c], ["% scalar ${template}4 vs ${template}*", \@unpack4s_c, \@unpack_c], ["% scalar ${template}* vs ${template}*", \@unpacks_c, \@unpack_c], ); } foreach my $test (@tests) { ok (list_eq ($test->[1], $test->[2]), $test->[0]) || printf "# unpack gave %s expected %s\n", encode_list (@{$test->[1]}), encode_list (@{$test->[2]}); } } } ok(pack('u2', 'AA'), "[perl #8026]"); # used to hang and eat RAM in perl 5.7.2 $_ = pack('c', 65); # 'A' would not be EBCDIC-friendly is(unpack('c'), 65, "one-arg unpack (change #18751)"); # defaulting to $_ { my $a = "X\x0901234567\n" x 100; # \t would not be EBCDIC TAB my @a = unpack("(a1 c/a)*", $a); is(scalar @a, 200, "[perl #15288]"); is($a[-1], "01234567\n", "[perl #15288]"); is($a[-2], "X", "[perl #15288]"); } { use warnings qw(NONFATAL all);; my $warning; local $SIG{__WARN__} = sub { $warning = $_[0]; }; my $out = pack("u99", "foo" x 99); like($warning, qr/Field too wide in 'u' format in pack at /, "Warn about too wide uuencode"); is($out, ("_" . "9F]O" x 21 . "\n") x 4 . "M" . "9F]O" x 15 . "\n", "Use max width in case of too wide uuencode"); } # checksums { # verify that unpack advances correctly wrt a checksum my (@x) = unpack("b10a", "abcd"); my (@y) = unpack("%b10a", "abcd"); is($x[1], $y[1], "checksum advance ok"); # verify that the checksum is not overflowed with C0 if (ord('A') == 193) { is(unpack("C0%128U", "/bcd"), unpack("U0%128U", "abcd"), "checksum not overflowed"); } else { is(unpack("C0%128U", "abcd"), unpack("U0%128U", "abcd"), "checksum not overflowed"); } } { # U0 and C0 must be scoped my (@x) = unpack("a(U0)U", "b\341\277\274"); is($x[0], 'b', 'before scope'); is($x[1], 8188, 'after scope'); is(pack("a(U0)U", "b", 8188), "b\341\277\274"); } { # counted length prefixes shouldn't change C0/U0 mode # (note the length is actually 0 in this test) if (ord('A') == 193) { is(join(',', unpack("aU0C/UU", "b\0\341\277\274")), 'b,0'); is(join(',', unpack("aU0C/CU", "b\0\341\277\274")), 'b,0'); } else { is(join(',', unpack("aC/UU", "b\0\341\277\274")), 'b,8188'); is(join(',', unpack("aC/CU", "b\0\341\277\274")), 'b,8188'); is(join(',', unpack("aU0C/UU", "b\0\341\277\274")), 'b,225'); is(join(',', unpack("aU0C/CU", "b\0\341\277\274")), 'b,225'); } } { # "Z0" (bug #34062) my (@x) = unpack("C*", pack("CZ0", 1, "b")); is(join(',', @x), '1', 'pack Z0 doesn\'t destroy the character before'); } { # Encoding neutrality # String we will pull apart and rebuild in several ways: my $down = "\xf8\xf9\xfa\xfb\xfc\xfd\xfe\xff\x05\x06"; my $up = $down; utf8::upgrade($up); my %expect = # [expected result, # how many chars it should progress, # (optional) expected result of pack] (a5 => ["\xf8\xf9\xfa\xfb\xfc", 5], A5 => ["\xf8\xf9\xfa\xfb\xfc", 5], Z5 => ["\xf8\xf9\xfa\xfb\xfc", 5, "\xf8\xf9\xfa\xfb\x00\xfd"], b21 => ["000111111001111101011", 3, "\xf8\xf9\x1a\xfb"], B21 => ["111110001111100111111", 3, "\xf8\xf9\xf8\xfb"], H5 => ["f8f9f", 3, "\xf8\xf9\xf0\xfb"], h5 => ["8f9fa", 3, "\xf8\xf9\x0a\xfb"], "s<" => [-1544, 2], "s>" => [-1799, 2], "S<" => [0xf9f8, 2], "S>" => [0xf8f9, 2], "l<" => [-67438088, 4], "l>" => [-117835013, 4], "L>" => [0xf8f9fafb, 4], "L<" => [0xfbfaf9f8, 4], n => [0xf8f9, 2], N => [0xf8f9fafb, 4], v => [63992, 2], V => [0xfbfaf9f8, 4], c => [-8, 1], U0U => [0xf8, 1], w => ["8715569050387726213", 9], q => ["-283686952306184", 8], Q => ["18446460386757245432", 8], ); for my $string ($down, $up) { for my $format (sort {lc($a) cmp lc($b) || $a cmp $b } keys %expect) { SKIP: { my $expect = $expect{$format}; # unpack upgraded and downgraded string my @result = eval { unpack("$format C0 W", $string) }; skip "cannot pack/unpack '$format C0 W' on this perl", 5 if $@ && is_valid_error($@); is(@result, 2, "Two results from unpack $format C0 W"); # pack to downgraded my $new = pack("$format C0 W", @result); is(length($new), $expect->[1]+1, "pack $format C0 W should give $expect->[1]+1 chars"); is($new, $expect->[2] || substr($string, 0, length $new), "pack $format C0 W returns expected value"); # pack to upgraded $new = pack("a0 $format C0 W", chr(256), @result); is(length($new), $expect->[1]+1, "pack a0 $format C0 W should give $expect->[1]+1 chars"); is($new, $expect->[2] || substr($string, 0, length $new), "pack a0 $format C0 W returns expected value"); } } } } { # Encoding neutrality, numbers my $val = -2.68; for my $format (qw(s S i I l L j J f d F D q Q s! S! i! I! l! L! n! N! v! V!)) { SKIP: { my $down = eval { pack($format, $val) }; skip "cannot pack/unpack $format on this perl", 9 if $@ && is_valid_error($@); ok(!utf8::is_utf8($down), "Simple $format pack doesn't get upgraded"); my $up = pack("a0 $format", chr(256), $val); ok(utf8::is_utf8($up), "a0 $format with high char leads to upgrade"); is($down, $up, "$format generated strings are equal though"); my @down_expanded = unpack("$format W", $down . chr(0xce)); is(@down_expanded, 2, "Expand to two values"); is($down_expanded[1], 0xce, "unpack $format left us at the expected position"); my @up_expanded = unpack("$format W", $up . chr(0xce)); is(@up_expanded, 2, "Expand to two values"); is($up_expanded[1], 0xce, "unpack $format left us at the expected position"); is($down_expanded[0], $up_expanded[0], "$format unpack was neutral"); is(pack($format, $down_expanded[0]), $down, "Pack $format undoes unpack $format"); } } } { # C *is* neutral my $down = "\xf8\xf9\xfa\xfb\xfc\xfd\xfe\xff\x05\x06"; my $up = $down; utf8::upgrade($up); my @down = unpack("C*", $down); my @expect_down = (0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x05, 0x06); is("@down", "@expect_down", "byte expand"); is(pack("C*", @down), $down, "byte join"); my @up = unpack("C*", $up); my @expect_up = (0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x05, 0x06); is("@up", "@expect_up", "UTF-8 expand"); is(pack("U0C0C*", @up), $up, "UTF-8 join"); } { # Harder cases for the neutrality test # u format my $down = "\xf8\xf9\xfa\xfb\xfc\xfd\xfe\xff\x05\x06"; my $up = $down; utf8::upgrade($up); is(pack("u", $down), pack("u", $up), "u pack is neutral"); is(unpack("u", pack("u", $down)), $down, "u unpack to downgraded works"); is(unpack("U0C0u", pack("u", $down)), $up, "u unpack to upgraded works"); # p/P format # This actually only tests something if the address contains a byte >= 0x80 my $str = "abc\xa5\x00\xfede"; $down = pack("p", $str); is(pack("P", $str), $down); is(pack("U0C0p", $str), $down); is(pack("U0C0P", $str), $down); is(unpack("p", $down), "abc\xa5", "unpack p downgraded"); $up = $down; utf8::upgrade($up); is(unpack("p", $up), "abc\xa5", "unpack p upgraded"); is(unpack("P7", $down), "abc\xa5\x00\xfed", "unpack P downgraded"); is(unpack("P7", $up), "abc\xa5\x00\xfed", "unpack P upgraded"); # x, X and @ $down = "\xf8\xf9\xfa\xfb\xfc\xfd\xfe\xff\x05\x06"; $up = $down; utf8::upgrade($up); is(unpack('@4W', $down), 0xfc, "\@positioning on downgraded string"); is(unpack('@4W', $up), 0xfc, "\@positioning on upgraded string"); is(unpack('@4x2W', $down), 0xfe, "x moving on downgraded string"); is(unpack('@4x2W', $up), 0xfe, "x moving on upgraded string"); is(unpack('@4x!4W', $down), 0xfc, "x! moving on downgraded string"); is(unpack('@4x!4W', $up), 0xfc, "x! moving on upgraded string"); is(unpack('@5x!4W', $down), 0x05, "x! moving on downgraded string"); is(unpack('@5x!4W', $up), 0x05, "x! moving on upgraded string"); is(unpack('@4X2W', $down), 0xfa, "X moving on downgraded string"); is(unpack('@4X2W', $up), 0xfa, "X moving on upgraded string"); is(unpack('@4X!4W', $down), 0xfc, "X! moving on downgraded string"); is(unpack('@4X!4W', $up), 0xfc, "X! moving on upgraded string"); is(unpack('@5X!4W', $down), 0xfc, "X! moving on downgraded string"); is(unpack('@5X!4W', $up), 0xfc, "X! moving on upgraded string"); is(unpack('@5X!8W', $down), 0xf8, "X! moving on downgraded string"); is(unpack('@5X!8W', $up), 0xf8, "X! moving on upgraded string"); is(pack("W2x", 0xfa, 0xe3), "\xfa\xe3\x00", "x on downgraded string"); is(pack("W2x!4", 0xfa, 0xe3), "\xfa\xe3\x00\x00", "x! on downgraded string"); is(pack("W2x!2", 0xfa, 0xe3), "\xfa\xe3", "x! on downgraded string"); is(pack("U0C0W2x", 0xfa, 0xe3), "\xfa\xe3\x00", "x on upgraded string"); is(pack("U0C0W2x!4", 0xfa, 0xe3), "\xfa\xe3\x00\x00", "x! on upgraded string"); is(pack("U0C0W2x!2", 0xfa, 0xe3), "\xfa\xe3", "x! on upgraded string"); is(pack("W2X", 0xfa, 0xe3), "\xfa", "X on downgraded string"); is(pack("U0C0W2X", 0xfa, 0xe3), "\xfa", "X on upgraded string"); is(pack("W2X!2", 0xfa, 0xe3), "\xfa\xe3", "X! on downgraded string"); is(pack("U0C0W2X!2", 0xfa, 0xe3), "\xfa\xe3", "X! on upgraded string"); is(pack("W3X!2", 0xfa, 0xe3, 0xa6), "\xfa\xe3", "X! on downgraded string"); is(pack("U0C0W3X!2", 0xfa, 0xe3, 0xa6), "\xfa\xe3", "X! on upgraded string"); # backward eating through a ( moves the group starting point backwards is(pack("a*(Xa)", "abc", "q"), "abq", "eating before strbeg moves it back"); is(pack("a*(Xa)", "ab" . chr(512), "q"), "abq", "eating before strbeg moves it back"); # Check marked_upgrade is(pack('W(W(Wa@3W)@6W)@9W', 0xa1, 0xa2, 0xa3, "a", 0xa4, 0xa5, 0xa6), "\xa1\xa2\xa3a\x00\xa4\x00\xa5\x00\xa6"); $up = "a"; utf8::upgrade($up); is(pack('W(W(Wa@3W)@6W)@9W', 0xa1, 0xa2, 0xa3, $up, 0xa4, 0xa5, 0xa6), "\xa1\xa2\xa3a\x00\xa4\x00\xa5\x00\xa6", "marked upgrade caused by a"); is(pack('W(W(WW@3W)@6W)@9W', 0xa1, 0xa2, 0xa3, 256, 0xa4, 0xa5, 0xa6), "\xa1\xa2\xa3\x{100}\x00\xa4\x00\xa5\x00\xa6", "marked upgrade caused by W"); is(pack('W(W(WU0aC0@3W)@6W)@9W', 0xa1, 0xa2, 0xa3, "a", 0xa4, 0xa5, 0xa6), "\xa1\xa2\xa3a\x00\xa4\x00\xa5\x00\xa6", "marked upgrade caused by U0"); # a, A and Z $down = "\xa4\xa6\xa7"; $up = $down; utf8::upgrade($up); utf8::upgrade(my $high = "\xfeb"); for my $format ("a0", "A0", "Z0", "U0a0C0", "U0A0C0", "U0Z0C0") { is(pack("a* $format a*", "ab", $down, "cd"), "abcd", "$format format on plain string"); is(pack("a* $format a*", "ab", $up, "cd"), "abcd", "$format format on upgraded string"); is(pack("a* $format a*", $high, $down, "cd"), "\xfebcd", "$format format on plain string"); is(pack("a* $format a*", $high, $up, "cd"), "\xfebcd", "$format format on upgraded string"); my @down = unpack("a1 $format a*", "\xfeb"); is("@down", "\xfe b", "unpack $format"); my @up = unpack("a1 $format a*", $high); is("@up", "\xfe b", "unpack $format"); } is(pack("a1", $high), "\xfe"); is(pack("A1", $high), "\xfe"); is(pack("Z1", $high), "\x00"); is(pack("a2", $high), "\xfeb"); is(pack("A2", $high), "\xfeb"); is(pack("Z2", $high), "\xfe\x00"); is(pack("a5", $high), "\xfeb\x00\x00\x00"); is(pack("A5", $high), "\xfeb "); is(pack("Z5", $high), "\xfeb\x00\x00\x00"); is(pack("a*", $high), "\xfeb"); is(pack("A*", $high), "\xfeb"); is(pack("Z*", $high), "\xfeb\x00"); utf8::upgrade($high = "\xc3\xbeb"); is(pack("U0a2", $high), "\xfe"); is(pack("U0A2", $high), "\xfe"); is(pack("U0Z1", $high), "\x00"); is(pack("U0a3", $high), "\xfeb"); is(pack("U0A3", $high), "\xfeb"); is(pack("U0Z3", $high), "\xfe\x00"); is(pack("U0a6", $high), "\xfeb\x00\x00\x00"); is(pack("U0A6", $high), "\xfeb "); is(pack("U0Z6", $high), "\xfeb\x00\x00\x00"); is(pack("U0a*", $high), "\xfeb"); is(pack("U0A*", $high), "\xfeb"); is(pack("U0Z*", $high), "\xfeb\x00"); } { # pack / my @array = 1..14; my @out = unpack("N/S", pack("N/S", @array) . "abcd"); is("@out", "@array", "pack N/S works"); @out = unpack("N/S*", pack("N/S*", @array) . "abcd"); is("@out", "@array", "pack N/S* works"); @out = unpack("N/S*", pack("N/S14", @array) . "abcd"); is("@out", "@array", "pack N/S14 works"); @out = unpack("N/S*", pack("N/S15", @array) . "abcd"); is("@out", "@array", "pack N/S15 works"); @out = unpack("N/S*", pack("N/S13", @array) . "abcd"); is("@out", "@array[0..12]", "pack N/S13 works"); @out = unpack("N/S*", pack("N/S0", @array) . "abcd"); is("@out", "", "pack N/S0 works"); is(pack("Z*/a0", "abc"), "0\0", "pack Z*/a0 makes a short string"); is(pack("Z*/Z0", "abc"), "0\0", "pack Z*/Z0 makes a short string"); is(pack("Z*/a3", "abc"), "3\0abc", "pack Z*/a3 makes a full string"); is(pack("Z*/Z3", "abc"), "3\0ab\0", "pack Z*/Z3 makes a short string"); is(pack("Z*/a5", "abc"), "5\0abc\0\0", "pack Z*/a5 makes a long string"); is(pack("Z*/Z5", "abc"), "5\0abc\0\0", "pack Z*/Z5 makes a long string"); is(pack("Z*/Z"), "1\0\0", "pack Z*/Z makes an extended string"); is(pack("Z*/Z", ""), "1\0\0", "pack Z*/Z makes an extended string"); is(pack("Z*/a", ""), "0\0", "pack Z*/a makes an extended string"); } { # unpack("A*", $unicode) strips general unicode spaces is(unpack("A*", "ab \n\xa0 \0"), "ab \n\xa0", 'normal A* strip leaves \xa0'); is(unpack("U0C0A*", "ab \n\xa0 \0"), "ab \n\xa0", 'normal A* strip leaves \xa0 even if it got upgraded for technical reasons'); is(unpack("A*", pack("a*(U0U)a*", "ab \n", 0xa0, " \0")), "ab", 'upgraded strings A* removes \xa0'); is(unpack("A*", pack("a*(U0UU)a*", "ab \n", 0xa0, 0x1680, " \0")), "ab", 'upgraded strings A* removes all unicode whitespace'); is(unpack("A5", pack("a*(U0U)a*", "ab \n", 0x1680, "def", "ab")), "ab", 'upgraded strings A5 removes all unicode whitespace'); is(unpack("A*", pack("U", 0x1680)), "", 'upgraded strings A* with nothing left'); } { # Testing unpack . and .! is(unpack(".", "ABCD"), 0, "offset at start of string is 0"); is(unpack(".", ""), 0, "offset at start of empty string is 0"); is(unpack("x3.", "ABCDEF"), 3, "simple offset works"); is(unpack("x3.", "ABC"), 3, "simple offset at end of string works"); is(unpack("x3.0", "ABC"), 0, "self offset is 0"); is(unpack("x3(x2.)", "ABCDEF"), 2, "offset is relative to inner group"); is(unpack("x3(X2.)", "ABCDEF"), -2, "negative offset relative to inner group"); is(unpack("x3(X2.2)", "ABCDEF"), 1, "offset is relative to inner group"); is(unpack("x3(x2.0)", "ABCDEF"), 0, "self offset in group is still 0"); is(unpack("x3(x2.2)", "ABCDEF"), 5, "offset counts groups"); is(unpack("x3(x2.*)", "ABCDEF"), 5, "star offset is relative to start"); my $high = chr(8188) x 6; is(unpack("x3(x2.)", $high), 2, "utf8 offset is relative to inner group"); is(unpack("x3(X2.)", $high), -2, "utf8 negative offset relative to inner group"); is(unpack("x3(X2.2)", $high), 1, "utf8 offset counts groups"); is(unpack("x3(x2.0)", $high), 0, "utf8 self offset in group is still 0"); is(unpack("x3(x2.2)", $high), 5, "utf8 offset counts groups"); is(unpack("x3(x2.*)", $high), 5, "utf8 star offset is relative to start"); is(unpack("U0x3(x2.)", $high), 2, "U0 mode utf8 offset is relative to inner group"); is(unpack("U0x3(X2.)", $high), -2, "U0 mode utf8 negative offset relative to inner group"); is(unpack("U0x3(X2.2)", $high), 1, "U0 mode utf8 offset counts groups"); is(unpack("U0x3(x2.0)", $high), 0, "U0 mode utf8 self offset in group is still 0"); is(unpack("U0x3(x2.2)", $high), 5, "U0 mode utf8 offset counts groups"); is(unpack("U0x3(x2.*)", $high), 5, "U0 mode utf8 star offset is relative to start"); is(unpack("x3(x2.!)", $high), 2*3, "utf8 offset is relative to inner group"); is(unpack("x3(X2.!)", $high), -2*3, "utf8 negative offset relative to inner group"); is(unpack("x3(X2.!2)", $high), 1*3, "utf8 offset counts groups"); is(unpack("x3(x2.!0)", $high), 0, "utf8 self offset in group is still 0"); is(unpack("x3(x2.!2)", $high), 5*3, "utf8 offset counts groups"); is(unpack("x3(x2.!*)", $high), 5*3, "utf8 star offset is relative to start"); is(unpack("U0x3(x2.!)", $high), 2, "U0 mode utf8 offset is relative to inner group"); is(unpack("U0x3(X2.!)", $high), -2, "U0 mode utf8 negative offset relative to inner group"); is(unpack("U0x3(X2.!2)", $high), 1, "U0 mode utf8 offset counts groups"); is(unpack("U0x3(x2.!0)", $high), 0, "U0 mode utf8 self offset in group is still 0"); is(unpack("U0x3(x2.!2)", $high), 5, "U0 mode utf8 offset counts groups"); is(unpack("U0x3(x2.!*)", $high), 5, "U0 mode utf8 star offset is relative to start"); } { # Testing pack . and .! is(pack("(a)5 .", 1..5, 3), "123", ". relative to string start, shorten"); eval { () = pack("(a)5 .", 1..5, -3) }; like($@, qr{'\.' outside of string in pack}, "Proper error message"); is(pack("(a)5 .", 1..5, 8), "12345\x00\x00\x00", ". relative to string start, extend"); is(pack("(a)5 .", 1..5, 5), "12345", ". relative to string start, keep"); is(pack("(a)5 .0", 1..5, -3), "12", ". relative to string current, shorten"); is(pack("(a)5 .0", 1..5, 2), "12345\x00\x00", ". relative to string current, extend"); is(pack("(a)5 .0", 1..5, 0), "12345", ". relative to string current, keep"); is(pack("(a)5 (.)", 1..5, -3), "12", ". relative to group, shorten"); is(pack("(a)5 (.)", 1..5, 2), "12345\x00\x00", ". relative to group, extend"); is(pack("(a)5 (.)", 1..5, 0), "12345", ". relative to group, keep"); is(pack("(a)3 ((a)2 .)", 1..5, -2), "1", ". relative to group, shorten"); is(pack("(a)3 ((a)2 .)", 1..5, 2), "12345", ". relative to group, keep"); is(pack("(a)3 ((a)2 .)", 1..5, 4), "12345\x00\x00", ". relative to group, extend"); is(pack("(a)3 ((a)2 .2)", 1..5, 2), "12", ". relative to counted group, shorten"); is(pack("(a)3 ((a)2 .2)", 1..5, 7), "12345\x00\x00", ". relative to counted group, extend"); is(pack("(a)3 ((a)2 .2)", 1..5, 5), "12345", ". relative to counted group, keep"); is(pack("(a)3 ((a)2 .*)", 1..5, 2), "12", ". relative to start, shorten"); is(pack("(a)3 ((a)2 .*)", 1..5, 7), "12345\x00\x00", ". relative to start, extend"); is(pack("(a)3 ((a)2 .*)", 1..5, 5), "12345", ". relative to start, keep"); is(pack('(a)5 (. @2 a)', 1..5, -3, "a"), "12\x00\x00a", ". based shrink properly updates group starts"); is(pack("(W)3 ((W)2 .)", 0x301..0x305, -2), "\x{301}", "utf8 . relative to group, shorten"); is(pack("(W)3 ((W)2 .)", 0x301..0x305, 2), "\x{301}\x{302}\x{303}\x{304}\x{305}", "utf8 . relative to group, keep"); is(pack("(W)3 ((W)2 .)", 0x301..0x305, 4), "\x{301}\x{302}\x{303}\x{304}\x{305}\x00\x00", "utf8 . relative to group, extend"); is(pack("(W)3 ((W)2 .!)", 0x301..0x305, -2), "\x{301}\x{302}", "utf8 . relative to group, shorten"); is(pack("(W)3 ((W)2 .!)", 0x301..0x305, 4), "\x{301}\x{302}\x{303}\x{304}\x{305}", "utf8 . relative to group, keep"); is(pack("(W)3 ((W)2 .!)", 0x301..0x305, 6), "\x{301}\x{302}\x{303}\x{304}\x{305}\x00\x00", "utf8 . relative to group, extend"); is(pack('(W)5 (. @2 a)', 0x301..0x305, -3, "a"), "\x{301}\x{302}\x00\x00a", "utf8 . based shrink properly updates group starts"); } { # Testing @! is(pack('a* @3', "abcde"), "abc", 'Test basic @'); is(pack('a* @!3', "abcde"), "abc", 'Test basic @!'); is(pack('a* @2', "\x{301}\x{302}\x{303}\x{304}\x{305}"), "\x{301}\x{302}", 'Test basic utf8 @'); is(pack('a* @!2', "\x{301}\x{302}\x{303}\x{304}\x{305}"), "\x{301}", 'Test basic utf8 @!'); is(unpack('@4 a*', "abcde"), "e", 'Test basic @'); is(unpack('@!4 a*', "abcde"), "e", 'Test basic @!'); is(unpack('@4 a*', "\x{301}\x{302}\x{303}\x{304}\x{305}"), "\x{305}", 'Test basic utf8 @'); is(unpack('@!4 a*', "\x{301}\x{302}\x{303}\x{304}\x{305}"), "\x{303}\x{304}\x{305}", 'Test basic utf8 @!'); }