Skip to content
Simple functionality for testing PDL data in test scripts.
Branch: master
Clone or download
Pull request Compare This branch is 2 commits ahead of PDLPorters:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.



Test::PDL - Test Perl Data Language arrays (a.k.a. piddles) for equality


version 0.14


use PDL;
use Test::More tests => 3;
use Test::PDL qw( is_pdl :deep );

# an example of a test that succeeds
$got      = sequence 5;
$expected = pdl( 0,1,2,3,4 );
is_pdl( $got, $expected, 'sequence() works as expected' );
# ok 1 - sequence() works as expected

# if a test fails, detailed diagnostics are printed; the output is
# similar to that of is() from L<Test::More>
$got      = pdl( 0,-1,-2,3,4 );
$expected = sequence 5;
is_pdl( $got, $expected, 'demonstrate the output of a failing test' );
# not ok 2 - demonstrate the output of a failing test
#   Failed test 'demonstrate the output of a failing test'
#   at aux/pod.t line 16.
#     values do not match
#          got: Double   D [5]        (P    ) [0 -1 -2 3 4]
#     expected: Double   D [5]        (P    ) [0 1 2 3 4]

# piddles within other data structures can be tested with Test::Deep
use Test::Deep qw( cmp_deeply );
$got      = { name => 'Histogram', data => long( 17,0,1 ) };
$expected = { name => 'Histogram', data => test_long( 17,0,0,1 ) };
cmp_deeply( $got, $expected, 'demonstrate the output of a failing deep comparison' );
# not ok 3 - demonstrate the output of a failing deep comparison
#   Failed test 'demonstrate the output of a failing deep comparison'
#   at aux/pod.t line 30.
# Comparing $data->{"data"} as a piddle:
# dimensions do not match in extent
#    got : Long     D [3]        (P    ) [17 0 1]
# expect : Long     D [4]        (P    ) [17 0 0 1]


With Test::PDL, you can compare two piddles for equality. The comparison is performed as thoroughly as possible, comparing types, dimensions, bad value patterns, and finally the values themselves. The exact behaviour can be configured by setting certain options (see set_options() and %OPTIONS below). Test::PDL is mostly useful in test scripts.

Test::PDL is to be used with the Perl Data Language (PDL).

By default, Test::PDL exports only one function: is_pdl(). The other functions are exported on demand only. The export tag :deep exports test_pdl() and one function for each PDL type constructor (like short(), double(), etc.), prefixed with test_: test_short(), test_double(), ...



The comparison criteria used by Test::PDL can be configured by setting the values in the %OPTIONS hash. This can be done directly, by addressing %Test::PDL::OPTIONS directly. However, it is preferred that set_options() is used instead.


The tolerance used to compare floating-point values. Initially set to 1e-6. This is currently an absolute tolerance, meaning that two values compare equal if the absolute value of their difference is below the tolerance.


If true, only piddles with equal type can be considered equal. If false, the types of the piddles being compared is not taken into consideration. Defaults to true: types must match for the comparison to succeed. If you want to write tests like

is_pdl( $got, pdl([ 1, 3, 5, 6 ]) );

without having to worry about the type of the piddle being exactly double (which is the default type of the pdl() constructor), set EQUAL_TYPES equal to 0.



Custom importer that recognizes configuration options specified at use time, as in

use Test::PDL -equal_types => 0;

This invocation is equivalent to

use Test::PDL;
Test::PDL::set_options( EQUAL_TYPES => 0 );

but is arguably somewhat nicer.


Internal function reimplementing the functionality of PDL::approx(), but with a tolerance that is not remembered across invocations. Rather, the tolerance can be set by the user (see set_options() and $OPTIONS{TOLERANCE}), and defaults to 1e-6.


Internal function which does the real work of comparing two piddles. If the comparison fails, _comparison_fails() returns a string containing the reason for failure. If the comparison succeeds, _comparison_fails() returns zero.

The criteria for equality are the following:

  • Both arguments must be piddles for the comparison to succeed. Currently, there is no implicit conversion from scalar to piddle.

  • The type of both piddles must be equal if (and only if) EQUAL_TYPES is true.

  • The number of dimensions must be equal. That is, a two-dimensional piddle only compares equal with another two-dimensional piddle.

  • The extent of the dimensions are compared one by one and must match. That is, a piddle with dimensions (5,4) cannot compare equal with a piddle of dimensions (5,3). Note that degenerate dimensions are not treated specially, and thus a piddle with dimensions (5,4,1) is considered different from a piddle with dimensions (5,4).

  • For piddles that conform in type and shape, the bad value pattern is examined. If the two piddles have bad values in different positions, the piddles are considered different. Note that two piddles may compare equal even though their bad flag is different, if there are no bad values.

  • And last but not least, the values themselves are examined one by one. For integer types, the comparison is performed exactly, whereas an approximate equality is used for floating-point types. The approximate comparison is implemented using a private reimplementation of PDL::approx(). See _approx() for more information.


Internal function which compares the extent of each of the dimensions of two piddles, one by one. The dimensions must be passed in as two array references. Returns 1 if all dimensions match pairwise. Returns 0 otherwise.

This function will not operate correctly if the number of dimensions does not match between the piddles, so be sure to check that before calling this function.


Run a test comparing a piddle to an expected piddle, and fail with detailed diagnostics if they don't compare equal.

is_pdl( $got, $expected, $test_name );

Yields ok if the first two arguments are piddles that compare equal, not ok if the piddles are different, or if at least one is not a piddle. Prints a diagnostic when the comparison fails, with the reason and a brief printout of both arguments. See the documentation of _comparison_fails() for the comparison criteria. $test_name is optional.

Named after is() from Test::More.


Return true if two piddles compare equal, false otherwise.

my $equal = eq_pdl( $got, $expected );

eq_pdl() contains just the comparison part of is_pdl(), without the infrastructure required to write tests with Test::More. It could be used as part of a larger test in which the equality of two piddles must be verified. By itself, eq_pdl() does not generate any output, so it should be safe to use outside test suites.


Return true if two piddles compare equal, false otherwise, and the reason why the comparison failed (if it did).

my( $ok ) = eq_pdl_diag( $got, $expected );
my( $ok, $diag ) = eq_pdl_diag( $got, $expected );

eq_pdl_diag() is like eq_pdl(), except that it also returns the reason why the comparison failed (if it failed). $diag will be false if the comparison succeeds. Does not need Test::Builder, so you can use it as part of something else, without side effects (like generating output). It was written to support deep comparisons with Test::Deep.


Special comparison to be used in conjunction with Test::Deep to test piddles inside data structures.

my $expected = { ..., some_field => test_pdl( 1,2,-7 ), ... };
my $expected = [ ..., test_short( 1,2,-7 ), ... ];

Suppose you want to compare data structures that happen to contain piddles. You use is_deeply() (from Test::More) or cmp_deeply() (from Test::Deep) to compare the structures element by element. Unfortunately, you cannot just write

my $got = my_sub( ... );
my $expected = {
        some_field => pdl( ... ),
is_deeply $got, $expected;

Neither does cmp_deeply() work in the same situation. is_deeply() tries to compare the piddles using the (overloaded) == comparison operator, which doesn't work. It simply dies with an error message saying that multidimensional piddles cannot be compared, whereas cmp_deeply() performs only a shallow comparison of the references.

What you need is a special comparison, which is provided by this function, to be used with cmp_deeply(). You need to rewrite $expected as follows

my $expected = {
        some_field => test_pdl( ... ),
cmp_deeply $got, $expected;

Note that you need to write test_pdl() instead of pdl(). You could achieve the same thing with

my $expected = {
        some_field => code( sub { eq_pdl_diag( shift, pdl( ... ) ) } ),

but the diagnostics provided by test_pdl() are better, and it's easier to use. test_pdl() accepts the same arguments as the PDL constructor pdl() does. If you need to compare a piddle with a type different from the default type, use one of the provided test_byte(), test_short(), test_long(), etc.:

my $expected = { data => test_short( -4,-9,13 ) };

If you need to manipulate the expected value, you should keep in mind that the return value of test_pdl() and the like are not piddles. Therefore, in-place modification of the expected value won't work:

my $expected = { data => test_short( -99,-9,13 )->inplace->setvaltobad( -99 ) }; # won't work!

You should rather do

my $expected = { data => test_pdl( short(-99,-9,13)->inplace->setvaltobad(-99) ) };

test_pdl() will correctly set the type of the expected value to short in the above example.


Configure the comparison carried out by Test::PDL's testing functions.

# e.g., if a tolerance of 1e-6 is too tight
Test::PDL::set_options( TOLERANCE => 1e-4 );

The preferred way to set the options to this module. See %OPTIONS for all allowed options. set_options() dies with an error if an unknown option is passed. Note that sensible default values are provided for all options, so you needn't use this routine if you are fine with the defaults.

This function is not exported. Rather, it must be called as

Test::PDL::set_options( KEY => VALUE, ... );


None reported so far.


PDL, Test::More, Test::Deep, Test::PDL::Deep


Thanks to PDL Porters Joel Berger, Chris Marshall, and David Mertens for feedback and improvements.


Edward Baudrez <>


This software is copyright (c) 2019 by Edward Baudrez.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.

You can’t perform that action at this time.