[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';
}

use strict 'vars';

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 == $max_uv_less3) {
  print "1..0 # skipped: unsigned perl arithmetic is not sane";
  eval { require Config; import Config };
  use vars qw(%Config);
  if ($Config{d_quad} eq 'define') {
      print " (common in 64-bit platforms)";
  }
  print "\n";
  exit 0;
}
if ($max_uv_less3 =~ tr/0-9//c) {
  print "1..0 # skipped: this perl stringifies large unsigned integers using E notation\n";
  exit 0;
}

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

my $num = 0;
$num += 10**$_ - 4**$_ for 1.. $max_chain;
$num *= $st_t;
print "1..$num\n";		# 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

print "# @list\n";

# 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
# precisision. 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

#print "@list\n";
#print "'@ops'\n";

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]);

  # print "'@ops'\n";
  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) {
		  $max_uv & $inpt; # u 6
		} else {
		  use integer; $inpt + $zero;
		}
	      } elsif ($curop == 8) {
		$inpt + $zero;	# n 8
	      } else {
		$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 $ans[1]) {
	    print "# '$ans[0]' ne '$ans[1]',\t$num\t=> @opnames[$first,@{$curops[0]},$last] vs @opnames[$first,@{$curops[1]},$last]\n";
	    # 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 $max_uv_p1))) {
	      # string ++ versus numeric ++. Tolerate this little
	      # bit of insanity
	      print "# ok, as string ++ of max_uv is \"$max_uv_pp\", numeric is $max_uv_p1\n"
	    } elsif ($opnames[$last] eq 'I' and $ans[1] eq "-1"
		     and $ans[0] eq $max_uv_p1_as_iv) {
              # Max UV plus 1 is NV. This NV may stringify in E notation.
              # And the number of decimal digits shown in E notation will depend
              # on the binary digits in the mantissa. And it may be that
              # (say)  18446744073709551616 in E notation is truncated to
              # (say) 1.8446744073709551e+19 (say) which gets converted back
              # as    1.8446744073709551000e+19
              # ie    18446744073709551000
              # which isn't the integer we first had.
              # But each step of conversion is correct. So it's not an error.
              # (Only shows up for 64 bit UVs and NVs with 64 bit mantissas,
              #  and on Crays (64 bit integers, 48 bit mantissas) IIRC)
	      print "# ok, \"$max_uv_p1\" correctly converts to IV \"$max_uv_p1_as_iv\"\n";
	    } elsif ($opnames[$last] eq 'U' and $ans[1] eq ~0
		     and $ans[0] eq $max_uv_p1_as_uv) {
              # as aboce
	      print "# ok, \"$max_uv_p1\" correctly converts to UV \"$max_uv_p1_as_uv\"\n";
	    } elsif (grep {/^N$/} @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 $ans[1] !~ /^-?\d+$/) {
	      print "# ok, numerically equal - notation changed due to adding zero\n";
	    } else {
	      $nok++,
	    }
	  }
	}
        if ($nok) {
          print "not ok $test\n";
        } else {
          print "ok $test\n";
        }
	#print $txt if $nok;
	$test++;
      }
    }
  }
}
Commit	Line	Data
25da4f38 IZ	1	#!./perl
	2
	3	#
	4	# test the conversion operators
	5	#
	6	# Notations:
	7	#
	8	# "N p i N vs N N": Apply op-N, then op-p, then op-i, then reporter-N
	9	# Compare with application of op-N, then reporter-N
	10	# Right below are descriptions of different ops and reporters.
	11
	12	# We do not use these subroutines any more, sub overhead makes a "switch"
	13	# solution better:
	14
	15	# obviously, 0, 1 and 2, 3 are destructive. (XXXX 64-bit? 4 destructive too)
	16
	17	# *0 = sub {--$_[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';
	41	}
	42
	43	use strict 'vars';
	44
45c0de28	45	my $max_chain = $ENV{PERL_TEST_NUMCONVERTS} \|\| 2;
25da4f38 IZ	46
	47	# Bulk out if unsigned type is hopelessly wrong:
	48	my $max_uv1 = ~0;
	49	my $max_uv2 = sprintf "%u", $max_uv1 ** 6; # 6 is an arbitrary number here
	50	my $big_iv = do {use integer; $max_uv1 * 16}; # 16 is an arbitrary number here
59d8ce62	51	my $max_uv_less3 = $max_uv1 - 3;
25da4f38	52
972b05a9	53	print "# max_uv1 = $max_uv1, max_uv2 = $max_uv2, big_iv = $big_iv\n";
59d8ce62 NC	54	print "# max_uv_less3 = $max_uv_less3\n";
59d8ce62 NC	55	if ($max_uv1 ne $max_uv2 or $big_iv > $max_uv1 or $max_uv1 == $max_uv_less3) {
0be093c9 LC	56	print "1..0 # skipped: unsigned perl arithmetic is not sane";
	57	eval { require Config; import Config };
	58	use vars qw(%Config);
	59	if ($Config{d_quad} eq 'define') {
	60	print " (common in 64-bit platforms)";
	61	}
	62	print "\n";
25da4f38 IZ	63	exit 0;
25da4f38 IZ	64	}
59d8ce62 NC	65	if ($max_uv_less3 =~ tr/0-9//c) {
	66	print "1..0 # skipped: this perl stringifies large unsigned integers using E notation\n";
	67	exit 0;
	68	}
25da4f38 IZ	69
	70	my $st_t = 4*4; # We try 4 initializers and 4 reporters
	71
	72	my $num = 0;
	73	$num += 10$_ - 4$_ for 1.. $max_chain;
	74	$num *= $st_t;
	75	print "1..$num\n"; # In fact 15 times more subsubtests...
	76
	77	my $max_uv = ~0;
	78	my $max_iv = int($max_uv/2);
	79	my $zero = 0;
	80
	81	my $l_uv = length $max_uv;
	82	my $l_iv = length $max_iv;
	83
	84	# Hope: the first digits are good
	85	my $larger_than_uv = substr 97 x 100, 0, $l_uv;
	86	my $smaller_than_iv = substr 12 x 100, 0, $l_iv;
	87	my $yet_smaller_than_iv = substr 97 x 100, 0, ($l_iv - 1);
	88
	89	my @list = (1, $yet_smaller_than_iv, $smaller_than_iv, $max_iv, $max_iv + 1,
	90	$max_uv, $max_uv + 1);
	91	unshift @list, (reverse map -$_, @list), 0; # 15 elts
	92	@list = map "$_", @list; # Normalize
	93
28e5dec8	94	print "# @list\n";
25da4f38	95
28e5dec8 JH	96	# need to special case ++ for max_uv, as ++ "magic" on a string gives
	97	# another string, whereas ++ magic on a string used as a number gives
	98	# a number. Not a problem when NV preserves UV, but if it doesn't then
	99	# stringification of the latter gives something in e notation.
	100
	101	my $max_uv_pp = "$max_uv"; $max_uv_pp++;
	102	my $max_uv_p1 = "$max_uv"; $max_uv_p1+=0; $max_uv_p1++;
25da4f38	103
bd658be6 NC	104	# Also need to cope with %g notation for max_uv_p1 that actually gives an
	105	# integer less than max_uv because of correct rounding for the limited
	106	# precisision. This bites for 12 byte long doubles and 8 byte UVs
	107
	108	my $temp = $max_uv_p1;
	109	my $max_uv_p1_as_iv;
	110	{use integer; $max_uv_p1_as_iv = 0 + sprintf "%s", $temp}
	111	my $max_uv_p1_as_uv = 0 \| sprintf "%s", $temp;
	112
25da4f38 IZ	113	my @opnames = split //, "-+UINPuinp";
	114
	115	# @list = map { 2->($_), 3->($_), 4->($_), 5->($_), } @list; # Prepare input
	116
	117	#print "@list\n";
	118	#print "'@ops'\n";
	119
	120	my $test = 1;
	121	my $nok;
	122	for my $num_chain (1..$max_chain) {
	123	my @ops = map [split //], grep /[4-9]/,
	124	map { sprintf "%0${num_chain}d", $_ } 0 .. 10**$num_chain - 1;
	125
	126	#@ops = ([]) unless $num_chain;
	127	#@ops = ([6, 4]);
	128
	129	# print "'@ops'\n";
	130	for my $op (@ops) {
	131	for my $first (2..5) {
	132	for my $last (2..5) {
	133	$nok = 0;
	134	my @otherops = grep $_ <= 3, @$op;
	135	my @curops = ($op,\@otherops);
	136
	137	for my $num (@list) {
	138	my $inpt;
	139	my @ans;
	140
	141	for my $short (0, 1) {
	142	# undef $inpt; # Forget all we had - some bugs were masked
	143
	144	$inpt = $num; # Try to not contaminate $num...
	145	$inpt = "$inpt";
	146	if ($first == 2) {
	147	$inpt = $max_uv & $inpt; # U 2
	148	} elsif ($first == 3) {
	149	use integer; $inpt += $zero; # I 3
	150	} elsif ($first == 4) {
	151	$inpt += $zero; # N 4
	152	} else {
	153	$inpt = "$inpt"; # P 5
	154	}
	155
	156	# Saves 20% of time - not with this logic:
	157	#my $tmp = $inpt;
	158	#my $tmp1 = $num;
	159	#next if $num_chain > 1
	160	# and "$tmp" ne "$tmp1"; # Already the coercion gives problems...
	161
	162	for my $curop (@{$curops[$short]}) {
	163	if ($curop < 5) {
	164	if ($curop < 3) {
	165	if ($curop == 0) {
	166	--$inpt; # - 0
	167	} elsif ($curop == 1) {
	168	++$inpt; # + 1
	169	} else {
	170	$inpt = $max_uv & $inpt; # U 2
	171	}
	172	} elsif ($curop == 3) {
	173	use integer; $inpt += $zero;
	174	} else {
	175	$inpt += $zero; # N 4
	176	}
177	} elsif ($curop < 8) {
178	if ($curop == 5) {
179	$inpt = "$inpt"; # P 5
180	} elsif ($curop == 6) {
181	$max_uv & $inpt; # u 6
182	} else {
183	use integer; $inpt + $zero;
184	}
185	} elsif ($curop == 8) {
186	$inpt + $zero; # n 8
187	} else {
188	$inpt . ""; # p 9
189	}
190	}
191
192	if ($last == 2) {
193	$inpt = sprintf "%u", $inpt; # U 2
194	} elsif ($last == 3) {
195	$inpt = sprintf "%d", $inpt; # I 3
196	} elsif ($last == 4) {
197	$inpt = sprintf "%g", $inpt; # N 4
198	} else {
199	$inpt = "$inpt"; # P 5
200	}
201	push @ans, $inpt;
202	}
28e5dec8 JH	203	if ($ans[0] ne $ans[1]) {
	204	print "# '$ans[0]' ne '$ans[1]',\t$num\t=> @opnames[$first,@{$curops[0]},$last] vs @opnames[$first,@{$curops[1]},$last]\n";
	205	# XXX ought to check that "+" was in the list of opnames
	206	if ((($ans[0] eq $max_uv_pp) and ($ans[1] eq $max_uv_p1))
	207	or (($ans[1] eq $max_uv_pp) and ($ans[0] eq $max_uv_p1))) {
	208	# string ++ versus numeric ++. Tolerate this little
	209	# bit of insanity
	210	print "# ok, as string ++ of max_uv is \"$max_uv_pp\", numeric is $max_uv_p1\n"
bd658be6 NC	211	} elsif ($opnames[$last] eq 'I' and $ans[1] eq "-1"
bd658be6 NC	212	and $ans[0] eq $max_uv_p1_as_iv) {
3e93e9fb NC	213	# Max UV plus 1 is NV. This NV may stringify in E notation.
	214	# And the number of decimal digits shown in E notation will depend
	215	# on the binary digits in the mantissa. And it may be that
	216	# (say) 18446744073709551616 in E notation is truncated to
	217	# (say) 1.8446744073709551e+19 (say) which gets converted back
	218	# as 1.8446744073709551000e+19
	219	# ie 18446744073709551000
	220	# which isn't the integer we first had.
	221	# But each step of conversion is correct. So it's not an error.
	222	# (Only shows up for 64 bit UVs and NVs with 64 bit mantissas,
	223	# and on Crays (64 bit integers, 48 bit mantissas) IIRC)
bd658be6 NC	224	print "# ok, \"$max_uv_p1\" correctly converts to IV \"$max_uv_p1_as_iv\"\n";
	225	} elsif ($opnames[$last] eq 'U' and $ans[1] eq ~0
	226	and $ans[0] eq $max_uv_p1_as_uv) {
3e93e9fb	227	# as aboce
bd658be6 NC	228	print "# ok, \"$max_uv_p1\" correctly converts to UV \"$max_uv_p1_as_uv\"\n";
bd658be6 NC	229	} elsif (grep {/^N$/} @opnames[@{$curops[0]}]
3e93e9fb NC	230	and $ans[0] == $ans[1] and $ans[0] <= ~0
	231	# First must be in E notation (ie not just digits) and
	232	# second must still be an integer.
6ff52716 NC	233	# eg 1.84467440737095516e+19
	234	# 1.84467440737095516e+19 for 64 bit mantissa is in the
	235	# integer range, so 1.84467440737095516e+19 + 0 is treated
	236	# as integer addition. [should it be?]
	237	# and 18446744073709551600 + 0 is 18446744073709551600
	238	# Which isn't the string you first thought of.
3e93e9fb NC	239	# I can't remember why there isn't symmetry in this
3e93e9fb NC	240	# exception, ie why only the first ops are tested for 'N'
6ff52716	241	and $ans[0] != /^-?\d+$/ and $ans[1] !~ /^-?\d+$/) {
bd658be6	242	print "# ok, numerically equal - notation changed due to adding zero\n";
28e5dec8 JH	243	} else {
	244	$nok++,
	245	}
	246	}
25da4f38	247	}
3e93e9fb NC	248	if ($nok) {
	249	print "not ok $test\n";
	250	} else {
	251	print "ok $test\n";
	252	}
25da4f38 IZ	253	#print $txt if $nok;
	254	$test++;
	255	}
	256	}
	257	}
	258	}