[perl5.git] / t / op / numconvert.t

#!./perl

#
# test the conversion operators
#
# Notations:
#
# "N p i N vs N N":  Apply op-N, then op-p, then op-i, then reporter-N
# Compare with application of op-N, then reporter-N
# Right below are descriptions of different ops and reporters.

# We do not use these subroutines any more, sub overhead makes a "switch"
# solution better:

# obviously, 0, 1 and 2, 3 are destructive.  (XXXX 64-bit? 4 destructive too)

# *0 = sub {--$_[0]};		# -
# *1 = sub {++$_[0]};		# +

# # Converters
# *2 = sub { $_[0] = $max_uv & $_[0]}; # U
# *3 = sub { use integer; $_[0] += $zero}; # I
# *4 = sub { $_[0] += $zero};	# N
# *5 = sub { $_[0] = "$_[0]" };	# P

# # Side effects
# *6 = sub { $max_uv & $_[0]};	# u
# *7 = sub { use integer; $_[0] + $zero};	# i
# *8 = sub { $_[0] + $zero};	# n
# *9 = sub { $_[0] . "" };	# p

# # Reporters
# sub a2 { sprintf "%u", $_[0] }	# U
# sub a3 { sprintf "%d", $_[0] }	# I
# sub a4 { sprintf "%g", $_[0] }	# N
# sub a5 { "$_[0]" }		# P

BEGIN {
    chdir 't' if -d 't';
    @INC = '../lib';
    require './test.pl';
}

use strict;

my $max_chain = $ENV{PERL_TEST_NUMCONVERTS} || 2;

# Bulk out if unsigned type is hopelessly wrong:
my $max_uv1 = ~0;
my $max_uv2 = sprintf "%u", $max_uv1 ** 6; # 6 is an arbitrary number here
my $big_iv = do {use integer; $max_uv1 * 16}; # 16 is an arbitrary number here
my $max_uv_less3 = $max_uv1 - 3;

print "# max_uv1 = $max_uv1, max_uv2 = $max_uv2, big_iv = $big_iv\n";
print "# max_uv_less3 = $max_uv_less3\n";
if ($max_uv1 ne $max_uv2 or $big_iv > $max_uv1 or $max_uv1 == $max_uv_less3) {
  eval { require Config; };
  my $message = 'unsigned perl arithmetic is not sane';
  $message .= " (common in 64-bit platforms)" if $Config::Config{d_quad};
  skip_all($message);
}
if ($max_uv_less3 =~ tr/0-9//c) {
  skip_all('this perl stringifies large unsigned integers using E notation');
}

my $st_t = 4*4;			# We try 4 initializers and 4 reporters

my $num = 0;
$num += 10**$_ - 4**$_ for 1.. $max_chain;
$num *= $st_t;
$num += $::additional_tests;
plan(tests => $num);		# In fact 15 times more subsubtests...

my $max_uv = ~0;
my $max_iv = int($max_uv/2);
my $zero = 0;

my $l_uv = length $max_uv;
my $l_iv = length $max_iv;

# Hope: the first digits are good
my $larger_than_uv = substr 97 x 100, 0, $l_uv;
my $smaller_than_iv = substr 12 x 100, 0, $l_iv;
my $yet_smaller_than_iv = substr 97 x 100, 0, ($l_iv - 1);

my @list = (1, $yet_smaller_than_iv, $smaller_than_iv, $max_iv, $max_iv + 1,
	    $max_uv, $max_uv + 1);
unshift @list, (reverse map -$_, @list), 0; # 15 elts
@list = map "$_", @list; # Normalize

note("@list");

# need to special case ++ for max_uv, as ++ "magic" on a string gives
# another string, whereas ++ magic on a string used as a number gives
# a number. Not a problem when NV preserves UV, but if it doesn't then
# stringification of the latter gives something in e notation.

my $max_uv_pp = "$max_uv"; $max_uv_pp++;
my $max_uv_p1 = "$max_uv"; $max_uv_p1+=0; $max_uv_p1++;

# Also need to cope with %g notation for max_uv_p1 that actually gives an
# integer less than max_uv because of correct rounding for the limited
# precision. This bites for 12 byte long doubles and 8 byte UVs

my $temp = $max_uv_p1;
my $max_uv_p1_as_iv;
{use integer; $max_uv_p1_as_iv = 0 + sprintf "%s", $temp}
my $max_uv_p1_as_uv = 0 | sprintf "%s", $temp;

my @opnames = split //, "-+UINPuinp";

# @list = map { 2->($_), 3->($_), 4->($_), 5->($_),  } @list; # Prepare input

my $test = 1;
my $nok;
for my $num_chain (1..$max_chain) {
  my @ops = map [split //], grep /[4-9]/,
    map { sprintf "%0${num_chain}d", $_ }  0 .. 10**$num_chain - 1;

  #@ops = ([]) unless $num_chain;
  #@ops = ([6, 4]);

  for my $op (@ops) {
    for my $first (2..5) {
      for my $last (2..5) {
	$nok = 0;
	my @otherops = grep $_ <= 3, @$op;
	my @curops = ($op,\@otherops);

	for my $num (@list) {
	  my $inpt;
	  my @ans;

	  for my $short (0, 1) {
	    # undef $inpt;	# Forget all we had - some bugs were masked

	    $inpt = $num;	# Try to not contaminate $num...
	    $inpt = "$inpt";
	    if ($first == 2) {
	      $inpt = $max_uv & $inpt; # U 2
	    } elsif ($first == 3) {
	      use integer; $inpt += $zero; # I 3
	    } elsif ($first == 4) {
	      $inpt += $zero;	# N 4
	    } else {
	      $inpt = "$inpt";	# P 5
	    }

	    # Saves 20% of time - not with this logic:
	    #my $tmp = $inpt;
	    #my $tmp1 = $num;
	    #next if $num_chain > 1
	    #  and "$tmp" ne "$tmp1"; # Already the coercion gives problems...

	    for my $curop (@{$curops[$short]}) {
	      if ($curop < 5) {
		if ($curop < 3) {
		  if ($curop == 0) {
		    --$inpt;	# - 0
		  } elsif ($curop == 1) {
		    ++$inpt;	# + 1
		  } else {
		    $inpt = $max_uv & $inpt; # U 2
		  }
		} elsif ($curop == 3) {
		  use integer; $inpt += $zero;
		} else {
		  $inpt += $zero; # N 4
		}
	      } elsif ($curop < 8) {
		if ($curop == 5) {
		  $inpt = "$inpt"; # P 5
		} elsif ($curop == 6) {
		  my $dummy = $max_uv & $inpt; # u 6
		} else {
		  use integer; my $dummy = $inpt + $zero;
		}
	      } elsif ($curop == 8) {
		my $dummy = $inpt + $zero;	# n 8
	      } else {
		my $dummy = $inpt . "";	# p 9
	      }
	    }

	    if ($last == 2) {
	      $inpt = sprintf "%u", $inpt; # U 2
	    } elsif ($last == 3) {
	      $inpt = sprintf "%d", $inpt; # I 3
	    } elsif ($last == 4) {
	      $inpt = sprintf "%g", $inpt; # N 4
	    } else {
	      $inpt = "$inpt";	# P 5
	    }
	    push @ans, $inpt;
	  }
	  if ($ans[0] ne $ans[1]) {
	    my $diag = "'$ans[0]' ne '$ans[1]',\t$num\t=> @opnames[$first,@{$curops[0]},$last] vs @opnames[$first,@{$curops[1]},$last]";
	    my $excuse;
	    # XXX ought to check that "+" was in the list of opnames
	    if ((($ans[0] eq $max_uv_pp) and ($ans[1] eq $max_uv_p1))
		or (($ans[1] eq $max_uv_pp) and ($ans[0] eq $max_uv_p1))) {
	      # string ++ versus numeric ++. Tolerate this little
	      # bit of insanity
	      $excuse = "ok, as string ++ of max_uv is \"$max_uv_pp\", numeric is $max_uv_p1";
	    } elsif ($opnames[$last] eq 'I' and $ans[1] eq "-1"
		     and $ans[0] eq $max_uv_p1_as_iv) {
              # Max UV plus 1 is NV. This NV may stringify in E notation.
              # And the number of decimal digits shown in E notation will depend
              # on the binary digits in the mantissa. And it may be that
              # (say)  18446744073709551616 in E notation is truncated to
              # (say) 1.8446744073709551e+19 (say) which gets converted back
              # as    1.8446744073709551000e+19
              # ie    18446744073709551000
              # which isn't the integer we first had.
              # But each step of conversion is correct. So it's not an error.
              # (Only shows up for 64 bit UVs and NVs with 64 bit mantissas,
              #  and on Crays (64 bit integers, 48 bit mantissas) IIRC)
	      $excuse = "ok, \"$max_uv_p1\" correctly converts to IV \"$max_uv_p1_as_iv\"";
	    } elsif ($opnames[$last] eq 'U' and $ans[1] eq ~0
		     and $ans[0] eq $max_uv_p1_as_uv) {
              # as aboce
	      $excuse = "ok, \"$max_uv_p1\" correctly converts to UV \"$max_uv_p1_as_uv\"";
	    } elsif (grep {defined $_ && /^N$/} @opnames[@{$curops[0]}]
		     and $ans[0] == $ans[1] and $ans[0] <= ~0
                     # First must be in E notation (ie not just digits) and
                     # second must still be an integer.
		     # eg 1.84467440737095516e+19
		     # 1.84467440737095516e+19 for 64 bit mantissa is in the
		     # integer range, so 1.84467440737095516e+19 + 0 is treated
		     # as integer addition. [should it be?]
		     # and 18446744073709551600 + 0 is 18446744073709551600
		     # Which isn't the string you first thought of.
                     # I can't remember why there isn't symmetry in this
                     # exception, ie why only the first ops are tested for 'N'
                     and $ans[0] != /^-?\d+$/ and $ans[1] !~ /^-?\d+$/) {
	      $excuse = "ok, numerically equal - notation changed due to adding zero";
	    } else {
	      $nok++,
	      diag($diag);
	    }
	    if ($excuse) {
	      note($diag);
	      note($excuse);
	    }
	  }
	}
	ok($nok == 0);
      }
    }
  }
}

# Tests that use test.pl start here.
BEGIN { $::additional_tests = 4 }

ok(-0.0 eq "0", 'negative zero stringifies as 0');
ok(!-0.0, "neg zero is boolean false");
my $nz = -0.0;
{ my $dummy = "$nz"; }
ok(!$nz, 'previously stringified -0.0 is boolean false');
$nz = -0.0;
is sprintf("%+.f", - -$nz), sprintf("%+.f", - -$nz),
  "negation does not coerce negative zeroes";
Commit	Line	Data
25da4f38 IZ	1	#!./perl
	2
	3	#
	4	# test the conversion operators
	5	#
	6	# Notations:
	7	#
	8	# "N p i N vs N N": Apply op-N, then op-p, then op-i, then reporter-N
	9	# Compare with application of op-N, then reporter-N
	10	# Right below are descriptions of different ops and reporters.
	11
	12	# We do not use these subroutines any more, sub overhead makes a "switch"
	13	# solution better:
	14
	15	# obviously, 0, 1 and 2, 3 are destructive. (XXXX 64-bit? 4 destructive too)
	16
	17	# *0 = sub {--$_[0]}; # -
	18	# *1 = sub {++$_[0]}; # +
	19
	20	# # Converters
	21	# *2 = sub { $_[0] = $max_uv & $_[0]}; # U
	22	# *3 = sub { use integer; $_[0] += $zero}; # I
	23	# *4 = sub { $_[0] += $zero}; # N
	24	# *5 = sub { $_[0] = "$_[0]" }; # P
	25
	26	# # Side effects
	27	# *6 = sub { $max_uv & $_[0]}; # u
	28	# *7 = sub { use integer; $_[0] + $zero}; # i
	29	# *8 = sub { $_[0] + $zero}; # n
	30	# *9 = sub { $_[0] . "" }; # p
	31
	32	# # Reporters
	33	# sub a2 { sprintf "%u", $_[0] } # U
	34	# sub a3 { sprintf "%d", $_[0] } # I
	35	# sub a4 { sprintf "%g", $_[0] } # N
	36	# sub a5 { "$_[0]" } # P
	37
	38	BEGIN {
	39	chdir 't' if -d 't';
	40	@INC = '../lib';
afa74577	41	require './test.pl';
25da4f38 IZ	42	}
25da4f38 IZ	43
e8fbc27b	44	use strict;
25da4f38	45
45c0de28	46	my $max_chain = $ENV{PERL_TEST_NUMCONVERTS} \|\| 2;
25da4f38 IZ	47
	48	# Bulk out if unsigned type is hopelessly wrong:
	49	my $max_uv1 = ~0;
	50	my $max_uv2 = sprintf "%u", $max_uv1 ** 6; # 6 is an arbitrary number here
	51	my $big_iv = do {use integer; $max_uv1 * 16}; # 16 is an arbitrary number here
59d8ce62	52	my $max_uv_less3 = $max_uv1 - 3;
25da4f38	53
972b05a9	54	print "# max_uv1 = $max_uv1, max_uv2 = $max_uv2, big_iv = $big_iv\n";
59d8ce62 NC	55	print "# max_uv_less3 = $max_uv_less3\n";
59d8ce62 NC	56	if ($max_uv1 ne $max_uv2 or $big_iv > $max_uv1 or $max_uv1 == $max_uv_less3) {
028bf728 NC	57	eval { require Config; };
	58	my $message = 'unsigned perl arithmetic is not sane';
	59	$message .= " (common in 64-bit platforms)" if $Config::Config{d_quad};
	60	skip_all($message);
25da4f38	61	}
59d8ce62	62	if ($max_uv_less3 =~ tr/0-9//c) {
028bf728	63	skip_all('this perl stringifies large unsigned integers using E notation');
59d8ce62	64	}
25da4f38 IZ	65
	66	my $st_t = 4*4; # We try 4 initializers and 4 reporters
	67
	68	my $num = 0;
	69	$num += 10$_ - 4$_ for 1.. $max_chain;
	70	$num *= $st_t;
afa74577	71	$num += $::additional_tests;
e8fbc27b	72	plan(tests => $num); # In fact 15 times more subsubtests...
25da4f38 IZ	73
	74	my $max_uv = ~0;
	75	my $max_iv = int($max_uv/2);
	76	my $zero = 0;
	77
	78	my $l_uv = length $max_uv;
	79	my $l_iv = length $max_iv;
	80
	81	# Hope: the first digits are good
	82	my $larger_than_uv = substr 97 x 100, 0, $l_uv;
	83	my $smaller_than_iv = substr 12 x 100, 0, $l_iv;
	84	my $yet_smaller_than_iv = substr 97 x 100, 0, ($l_iv - 1);
	85
	86	my @list = (1, $yet_smaller_than_iv, $smaller_than_iv, $max_iv, $max_iv + 1,
	87	$max_uv, $max_uv + 1);
	88	unshift @list, (reverse map -$_, @list), 0; # 15 elts
	89	@list = map "$_", @list; # Normalize
	90
e8fbc27b	91	note("@list");
25da4f38	92
28e5dec8 JH	93	# need to special case ++ for max_uv, as ++ "magic" on a string gives
	94	# another string, whereas ++ magic on a string used as a number gives
	95	# a number. Not a problem when NV preserves UV, but if it doesn't then
	96	# stringification of the latter gives something in e notation.
	97
	98	my $max_uv_pp = "$max_uv"; $max_uv_pp++;
	99	my $max_uv_p1 = "$max_uv"; $max_uv_p1+=0; $max_uv_p1++;
25da4f38	100
bd658be6 NC	101	# Also need to cope with %g notation for max_uv_p1 that actually gives an
bd658be6 NC	102	# integer less than max_uv because of correct rounding for the limited
93f09d7b	103	# precision. This bites for 12 byte long doubles and 8 byte UVs
bd658be6 NC	104
	105	my $temp = $max_uv_p1;
	106	my $max_uv_p1_as_iv;
	107	{use integer; $max_uv_p1_as_iv = 0 + sprintf "%s", $temp}
	108	my $max_uv_p1_as_uv = 0 \| sprintf "%s", $temp;
	109
25da4f38 IZ	110	my @opnames = split //, "-+UINPuinp";
	111
	112	# @list = map { 2->($_), 3->($_), 4->($_), 5->($_), } @list; # Prepare input
	113
25da4f38 IZ	114	my $test = 1;
	115	my $nok;
	116	for my $num_chain (1..$max_chain) {
	117	my @ops = map [split //], grep /[4-9]/,
	118	map { sprintf "%0${num_chain}d", $_ } 0 .. 10**$num_chain - 1;
	119
	120	#@ops = ([]) unless $num_chain;
	121	#@ops = ([6, 4]);
	122
25da4f38 IZ	123	for my $op (@ops) {
	124	for my $first (2..5) {
	125	for my $last (2..5) {
	126	$nok = 0;
	127	my @otherops = grep $_ <= 3, @$op;
	128	my @curops = ($op,\@otherops);
	129
	130	for my $num (@list) {
	131	my $inpt;
	132	my @ans;
	133
	134	for my $short (0, 1) {
	135	# undef $inpt; # Forget all we had - some bugs were masked
	136
	137	$inpt = $num; # Try to not contaminate $num...
	138	$inpt = "$inpt";
	139	if ($first == 2) {
	140	$inpt = $max_uv & $inpt; # U 2
	141	} elsif ($first == 3) {
	142	use integer; $inpt += $zero; # I 3
	143	} elsif ($first == 4) {
	144	$inpt += $zero; # N 4
	145	} else {
	146	$inpt = "$inpt"; # P 5
	147	}
	148
	149	# Saves 20% of time - not with this logic:
	150	#my $tmp = $inpt;
	151	#my $tmp1 = $num;
	152	#next if $num_chain > 1
	153	# and "$tmp" ne "$tmp1"; # Already the coercion gives problems...
	154
	155	for my $curop (@{$curops[$short]}) {
	156	if ($curop < 5) {
	157	if ($curop < 3) {
	158	if ($curop == 0) {
	159	--$inpt; # - 0
	160	} elsif ($curop == 1) {
	161	++$inpt; # + 1
	162	} else {
	163	$inpt = $max_uv & $inpt; # U 2
	164	}
	165	} elsif ($curop == 3) {
	166	use integer; $inpt += $zero;
	167	} else {
	168	$inpt += $zero; # N 4
	169	}
	170	} elsif ($curop < 8) {
	171	if ($curop == 5) {
	172	$inpt = "$inpt"; # P 5
	173	} elsif ($curop == 6) {
e8fbc27b	174	my $dummy = $max_uv & $inpt; # u 6
25da4f38	175	} else {
e8fbc27b	176	use integer; my $dummy = $inpt + $zero;
25da4f38 IZ	177	}
25da4f38 IZ	178	} elsif ($curop == 8) {
e8fbc27b	179	my $dummy = $inpt + $zero; # n 8
25da4f38	180	} else {
e8fbc27b	181	my $dummy = $inpt . ""; # p 9
25da4f38 IZ	182	}
	183	}
	184
	185	if ($last == 2) {
	186	$inpt = sprintf "%u", $inpt; # U 2
	187	} elsif ($last == 3) {
	188	$inpt = sprintf "%d", $inpt; # I 3
	189	} elsif ($last == 4) {
	190	$inpt = sprintf "%g", $inpt; # N 4
	191	} else {
	192	$inpt = "$inpt"; # P 5
	193	}
	194	push @ans, $inpt;
	195	}
28e5dec8	196	if ($ans[0] ne $ans[1]) {
e8fbc27b NC	197	my $diag = "'$ans[0]' ne '$ans[1]',\t$num\t=> @opnames[$first,@{$curops[0]},$last] vs @opnames[$first,@{$curops[1]},$last]";
e8fbc27b NC	198	my $excuse;
28e5dec8 JH	199	# XXX ought to check that "+" was in the list of opnames
	200	if ((($ans[0] eq $max_uv_pp) and ($ans[1] eq $max_uv_p1))
	201	or (($ans[1] eq $max_uv_pp) and ($ans[0] eq $max_uv_p1))) {
	202	# string ++ versus numeric ++. Tolerate this little
	203	# bit of insanity
e8fbc27b	204	$excuse = "ok, as string ++ of max_uv is \"$max_uv_pp\", numeric is $max_uv_p1";
bd658be6 NC	205	} elsif ($opnames[$last] eq 'I' and $ans[1] eq "-1"
bd658be6 NC	206	and $ans[0] eq $max_uv_p1_as_iv) {
3e93e9fb NC	207	# Max UV plus 1 is NV. This NV may stringify in E notation.
	208	# And the number of decimal digits shown in E notation will depend
	209	# on the binary digits in the mantissa. And it may be that
	210	# (say) 18446744073709551616 in E notation is truncated to
	211	# (say) 1.8446744073709551e+19 (say) which gets converted back
	212	# as 1.8446744073709551000e+19
	213	# ie 18446744073709551000
	214	# which isn't the integer we first had.
	215	# But each step of conversion is correct. So it's not an error.
	216	# (Only shows up for 64 bit UVs and NVs with 64 bit mantissas,
	217	# and on Crays (64 bit integers, 48 bit mantissas) IIRC)
e8fbc27b	218	$excuse = "ok, \"$max_uv_p1\" correctly converts to IV \"$max_uv_p1_as_iv\"";
bd658be6 NC	219	} elsif ($opnames[$last] eq 'U' and $ans[1] eq ~0
bd658be6 NC	220	and $ans[0] eq $max_uv_p1_as_uv) {
3e93e9fb	221	# as aboce
e8fbc27b NC	222	$excuse = "ok, \"$max_uv_p1\" correctly converts to UV \"$max_uv_p1_as_uv\"";
e8fbc27b NC	223	} elsif (grep {defined $_ && /^N$/} @opnames[@{$curops[0]}]
3e93e9fb NC	224	and $ans[0] == $ans[1] and $ans[0] <= ~0
	225	# First must be in E notation (ie not just digits) and
	226	# second must still be an integer.
6ff52716 NC	227	# eg 1.84467440737095516e+19
	228	# 1.84467440737095516e+19 for 64 bit mantissa is in the
	229	# integer range, so 1.84467440737095516e+19 + 0 is treated
	230	# as integer addition. [should it be?]
	231	# and 18446744073709551600 + 0 is 18446744073709551600
	232	# Which isn't the string you first thought of.
3e93e9fb NC	233	# I can't remember why there isn't symmetry in this
3e93e9fb NC	234	# exception, ie why only the first ops are tested for 'N'
6ff52716	235	and $ans[0] != /^-?\d+$/ and $ans[1] !~ /^-?\d+$/) {
e8fbc27b	236	$excuse = "ok, numerically equal - notation changed due to adding zero";
28e5dec8 JH	237	} else {
28e5dec8 JH	238	$nok++,
e8fbc27b NC	239	diag($diag);
	240	}
	241	if ($excuse) {
	242	note($diag);
	243	note($excuse);
28e5dec8 JH	244	}
28e5dec8 JH	245	}
25da4f38	246	}
e8fbc27b	247	ok($nok == 0);
25da4f38 IZ	248	}
	249	}
	250	}
	251	}
afa74577 FC	252
afa74577 FC	253	# Tests that use test.pl start here.
886a4465	254	BEGIN { $::additional_tests = 4 }
afa74577	255
afa74577 FC	256	ok(-0.0 eq "0", 'negative zero stringifies as 0');
afa74577 FC	257	ok(!-0.0, "neg zero is boolean false");
e8fbc27b NC	258	my $nz = -0.0;
e8fbc27b NC	259	{ my $dummy = "$nz"; }
afa74577	260	ok(!$nz, 'previously stringified -0.0 is boolean false');
886a4465 FC	261	$nz = -0.0;
	262	is sprintf("%+.f", - -$nz), sprintf("%+.f", - -$nz),
	263	"negation does not coerce negative zeroes";