The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.


Data::Clean::ForJSON - Clean data so it is safe to output to JSON


This document describes version 0.395 of Data::Clean::ForJSON (from Perl distribution Data-Clean-ForJSON), released on 2019-11-26.


 use Data::Clean::ForJSON;
 my $cleanser = Data::Clean::ForJSON->get_cleanser;
 my $data     = { code=>sub {}, re=>qr/abc/i };

 my $cleaned;

 # modifies data in-place
 $cleaned = $cleanser->clean_in_place($data);

 # ditto, but deep clone first, return
 $cleaned = $cleanser->clone_and_clean($data);

 # now output it
 use JSON;
 print encode_json($cleaned); # prints '{"code":"CODE","re":"(?^i:abc)"}'

Functional shortcuts:

 use Data::Clean::ForJSON qw(clean_json_in_place clone_and_clean_json);

 # equivalent to Data::Clean::ForJSON->get_cleanser->clean_in_place($data)

 # equivalent to Data::Clean::ForJSON->get_cleanser->clone_and_clean($data)
 $cleaned = clone_and_clean_json($data);


This class cleans data from anything that might be problematic when encoding to JSON. This includes coderefs, globs, and so on. Here's what it will do by default:

  • Change DateTime and Time::Moment object to its epoch value

  • Change Regexp and version object to its string value

  • Change scalar references (e.g. \1) to its scalar value (e.g. 1)

  • Change other references (non-hash, non-array) to its ref() value (e.g. "GLOB", "CODE")

  • Clone circular references

    With a default limit of 1, meaning that if a reference is first seen again for the first time, it will be cloned. But if it is seen again for the second time, it will be replaced with "CIRCULAR".

    To change the default limit, customize your cleanser object:

     $cleanser = Data::Clean::ForJSON->new(
         -circular => ["clone", 4],

    or you can perform other action for circular references, see Data::Clean for more details.

  • Unbless other types of objects

Cleaning recurses into objects.

Data that has been cleaned will probably not be convertible back to the original, due to information loss (for example, coderefs converted to string "CODE").

The design goals are good performance, good defaults, and just enough flexibility. The original use-case is for returning JSON response in HTTP API service.

This module is significantly faster than modules like Data::Rmap or Data::Visitor::Callback because with something like Data::Rmap you repeatedly invoke callback for each data item. This module, on the other hand, generates a cleanser code using eval(), using native Perl for() loops.

If LOG_CLEANSER_CODE environment is set to true, the generated cleanser code will be logged using Log::ger at trace level. You can see it, e.g. using Log::ger::Output::Screen:

 % LOG_CLEANSER_CODE=1 perl -MLog::ger::Output=Screen -MLog::ger::Level::trace -MData::Clean::ForJSON \
   -e'$c=Data::Clean::ForJSON->new; ...'


None of the functions are exported by default.


A shortcut for:


clone_and_clean_json($data) => $cleaned

A shortcut for:

 $cleaned = Data::Clean::ForJSON->get_cleanser->clone_and_clean($data)


CLASS->get_cleanser => $obj

Return a singleton instance, with default options. Use new() if you want to customize options.

CLASS->new() => $obj

Create a new instance.

$obj->clean_in_place($data) => $cleaned

Clean $data. Modify data in-place.

$obj->clone_and_clean($data) => $cleaned

Clean $data. Clone $data first.


Why clone/modify? Why not directly output JSON?

So that the data can be used for other stuffs, like outputting to YAML, etc.

Why is it slow?

If you use new() instead of get_cleanser(), make sure that you do not construct the Data::Clean::ForJSON object repeatedly, as the constructor generates the cleanser code first using eval(). A short benchmark (run on my slow Atom netbook):

 % bench -MData::Clean::ForJSON -b'$c=Data::Clean::ForJSON->new' \
     'Data::Clean::ForJSON->new->clone_and_clean([1..100])' \
 Benchmarking sub { Data::Clean::ForJSON->new->clean_in_place([1..100]) }, sub { $c->clean_in_place([1..100]) } ...
 a: 302 calls (291.3/s), 1.037s (3.433ms/call)
 b: 7043 calls (4996/s), 1.410s (0.200ms/call)
 Fastest is b (17.15x a)

Second, you can turn off some checks if you are sure you will not be getting bad data. For example, if you know that your input will not contain circular references, you can turn off circular detection:

 $cleanser = Data::Clean::ForJSON->new(-circular => 0);


 $ perl -MData::Clean::ForJSON -MBench -E '
   $data = [[1],[2],[3],[4],[5]];
   bench {
       circ   => sub { state $c = Data::Clean::ForJSON->new;               $c->clone_and_clean($data) },
       nocirc => sub { state $c = Data::Clean::ForJSON->new(-circular=>0); $c->clone_and_clean($data) }
   }, -1'
 circ: 9456 calls (9425/s), 1.003s (0.106ms/call)
 nocirc: 13161 calls (12885/s), 1.021s (0.0776ms/call)
 Fastest is nocirc (1.367x circ)

The less number of checks you do, the faster the cleansing process will be.

Why am I getting 'Not a CODE reference at lib/Data/ line xxx'?

[2013-08-07 ] This error message is from Data::Clone::clone() when it is cloning an object. If you are cleaning objects, instead of using clone_and_clean(), try using clean_in_place(). Or, clone your data first using something else like Sereal.



Bool. Can be set to true to log cleanser code using Log::ger at trace level.


Please visit the project's homepage at


Source repository is at


Please report any bugs or feature requests on the bugtracker website

When submitting a bug or request, please include a test-file or a patch to an existing test-file that illustrates the bug or desired feature.




Data::Abridge is similar in goal, which is to let Perl data structures (which might contain stuffs unsupported in JSON) be encodeable to JSON. But unlike Data::Clean::ForJSON, it has some (currently) non-configurable rules, like changing a coderef with a hash {CODE=>'\&main::__ANON__'} or a scalar ref with {SCALAR=>'value'} and so on. Note that the abridging process is similarly unidirectional (you cannot convert back the original Perl data structure).

Some benchmarks in Bencher::Scenarios::DataCleansing. You can see that Data::Clean::ForJSON can be several times faster than, say, Data::Rmap.


perlancar <>


This software is copyright (c) 2019, 2018, 2017, 2016, 2015, 2014, 2013, 2012 by

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