Perinci::Sub::Wrapper - A multi-purpose subroutine wrapping framework


This document describes version 0.82 of Perinci::Sub::Wrapper (from Perl distribution Perinci-Sub-Wrapper), released on 2016-02-04.


For dynamic usage:

 use Perinci::Sub::Wrapper qw(wrap_sub);
 my $res = wrap_sub(sub_name => "mysub", meta=>{...});
 my ($wrapped_sub, $meta) = ($res->[2]{sub}, $res->[2]{meta});
 $wrapped_sub->(); # call the wrapped function


Perinci::Sub::Wrapper (PSW for short) is an extensible subroutine wrapping framework. It generates code to do stuffs before calling your subroutine, like validate arguments, convert arguments from positional/array to named/hash or vice versa, etc; as well as generate code to do stuffs after calling your subroutine, like retry calling for a number of times if subroutine returns a non-success status, check subroutine result against a schema, etc). Some other things it can do: apply a timeout, currying, and so on.

PSW differs from other function composition or decoration system like Python decorators (or its Perl equivalent Python::Decorator) in a couple of ways:

  • Single wrapper

    Instead of multiple/nested wrapping for implementing different features, PSW is designed to generate a single large wrapper around your code, i.e.:

     sub _wrapper_for_your_sub {
         # do various stuffs before calling:
         # e.g. start timer
         # e.g. convert, prefill, validate arguments
         my @args = ...;
         # do various stuffs after calling
         # e.g. report times
         # e.g. perform retry
         # e.g. convert or envelope results
         # return result

    Multiple functionalities will be added and combined in this single wrapper subroutine in the appropriate location. This is done to reduce function call overhead or depth of nested call levels. And also to make it easier to embed the wrapping code to your source code (see Dist::Zilla::Plugin::Rinci::Wrap).

    Of course, you can still wrap multiple times if wanted.

  • Rinci

    The wrapper code is built according to the Rinci metadata you provide. Rinci allows you to specify various things for your function, e.g. list of arguments including the expected data type of each argument and whether an argument is required or optional. PSW can then be used to generate the necessary code to enforce this specification, e.g. generate validator for the function arguments.

    Since Rinci specification is extensible, you can describe additional stuffs for your function and write a PSW plugin to generate the necessary code to implement your specification. An example is timeout to specify execution time limit, implemented by Perinci::Sub::Property::timeout which generates code to call function inside an eval() block and use alarm() to limit the execution. Another example is retry property, implemented by Perinci::Sub::Property::retry which generates code to call function inside a simple retry loop.

Normally you do not use PSW directly in your applications. You might want to check out Perinci::Access::Perl and Perinci::Exporter on examples of wrapping function dynamically (during runtime), or Dist::Zilla::Plugin::Rinci::Wrap on an example of embedding the generated wrapping code to source code during build.


The framework is simple and extensible. Please delve directly into the source code for now. Some notes:

The internal uses OO.

The main wrapper building mechanism is in the wrap() method.

For each Rinci property, it will call handle_NAME() wrapper handler method. The handlemeta_NAME() methods are called first, to determine order of processing. You can supply these methods either by subclassing the class or, more simply, monkeypatching the method in the Perinci::Sub::Wrapper package.

The wrapper handler method will be called with a hash argument, containing these keys: value (property value), new (this key will exist if convert argument of wrap() exists, to convert a property to a new value).

For properties that have name in the form of NAME1.NAME2.NAME3 (i.e., dotted) only the first part of the name will be used (i.e., handle_NAME1()).


$Log_Wrapper_Code (BOOL)

Whether to log wrapper result. Default is from environment variable LOG_PERINCI_WRAPPER_CODE, or false. Logging is done with Log::Any at trace level.


x.perinci.sub.wrapper.disable_validate_args => bool

Can be set to 1 to set validate_args to 0 by default. This is used e.g. if you already embed/insert code to validate arguments by other means and do not want to repeat validating arguments. E.g. used if you use Dist::Zilla::Plugin::Rinci::Validate.

x.perinci.sub.wrapper.disable_validate_result => bool

Can be set to 1 to set validate_result to 0 by default. This is used e.g. if you already embed/insert code to validate result by other means and do not want to repeat validating result. E.g. used if you use Dist::Zilla::Plugin::Rinci::Validate.

x.perinci.sub.wrapper.logs => array

Generated/added by this module to the function metadata for every wrapping done. Used to avoid adding repeated code, e.g. to validate result or arguments.


The following numbers are produced on an Intel Core i5-2400 3.1GHz desktop using PSW v0.51 and Perl v5.18.2. Operating system is Debian sid (64bit).

For perspective, empty subroutine (sub {}) as well as sub { [200, "OK"] } can be called around 5.3 mil/sec.

Wrapping this subroutine sub { [200, "OK"] } and this simple metadata {v=>1.1} using default options yields call performance for $sub->() of about 0.9 mil/sec. With validate_args=>0 and validate_result=>0, it's 1.5 mil/sec.

As more (and more complex) arguments are introduced and validated, overhead will increase. The significant portion of the overhead is in argument validation. For example, this metadata {v=>1.1, args=>{a=>{schema=>"int"}}} yields 0.5 mil/sec.


wrap_sub(%args) -> [status, msg, result, meta]

Wrap subroutine to do various things, like enforcing Rinci properties.

This function is not exported by default, but exportable.

Arguments ('*' denotes required arguments):

  • compile => bool (default: 1)

    Whether to compile the generated wrapper.

    Can be set to 0 to not actually wrap but just return the generated wrapper source code.

  • convert => hash

    Properties to convert to new value.

    Not all properties can be converted, but these are a partial list of those that can: v (usually do not need to be specified when converting from 1.0 to 1.1, will be done automatically), args_as, result_naked, default_lang.

  • core => bool

    If set to true, will avoid the use of non-core modules.

  • core_or_pp => bool

    If set to true, will avoid the use of non-core XS modules.

    In other words, will stick to core or pure-perl modules only.

  • debug => bool (default: 0)

    Generate code with debugging.

    If turned on, will produce various debugging in the generated code. Currently what this does:

    • add more comments (e.g. for each property handler)

  • meta* => hash

    The function metadata.

  • meta_name => str

    Where to find the metadata, e.g. "$SPEC{foo}".

    Some wrapper code (e.g. handler for dep property) needs to refer to the function metadata. If not provided, the wrapper will store the function metadata in a unique variable (e.g. $Perinci::Sub::Wrapped::meta34127816).

  • pp => bool

    If set to true, will avoid the use of XS modules.

  • sub => str

    The code to be wrapped.

    At least one of sub or sub_name must be specified.

  • sub_name => str

    The name of the subroutine, e.g. func or Foo::func (qualified).

    At least one of sub or sub_name must be specified.

  • validate_args => bool

    Whether wrapper should validate arguments.

    If set to true, will validate arguments. Validation error will cause status 400 to be returned. The default is to enable this unless previous wrapper(s) have already done this.

  • validate_result => bool

    Whether wrapper should validate arguments.

    If set to true, will validate sub's result. Validation error will cause wrapper to return status 500 instead of sub's result. The default is to enable this unless previous wrapper(s) have already done this.

Returns an enveloped result (an array).

First element (status) is an integer containing HTTP status code (200 means OK, 4xx caller error, 5xx function error). Second element (msg) is a string containing error message, or 'OK' if status is 200. Third element (result) is optional, the actual result. Fourth element (meta) is called result metadata and is optional, a hash that contains extra information.

Return value: The wrapped subroutine along with its new metadata (hash)

Aside from wrapping the subroutine, the wrapper will also create a new metadata for the subroutine. The new metadata is a clone of the original, with some properties changed, e.g. schema in args and result normalized, some values changed according to the convert argument, some defaults set, etc.

The new metadata will also contain (or append) the wrapping log located in the x.perinci.sub.wrapper.logs attribute. The wrapping log marks that the wrapper has added some functionality (like validating arguments or result) so that future nested wrapper can choose to avoid duplicating the same functionality.


The OO interface is only used internally or when you want to extend the wrapper.



  • What is a function wrapper?

    A wrapper function calls the target function but with additional behaviors. The goal is similar to function composition or decorator system like in Python (or its Perl equivalent Python::Decorator) where you use a higher-order function which accepts another function and modifies it.

    It is used to add various functionalities, e.g.: cache/memoization, singleton, adding benchmarking/timing around function call, logging, argument validation (parameter checking), checking pre/post-condition, authentication/authorization checking, etc. The Python folks use decorators quite a bit; see discussions on the Internet on those.

  • How is PSW different from Python::Decorator?

    PSW uses dynamic code generation (it generates Perl code on the fly). It also creates a single large wrapper instead of nested wrappers. It builds wrapper code according to Rinci specification.

  • Why use code generation?

    Mainly because Data::Sah, which is the module used to do argument validation, also uses code generation. Data::Sah allows us to do data validation at full Perl speed, which can be one or two orders of magnitude faster than "interpreter" modules like Data::FormValidator.

  • Why use a single large wrapper?

    This is just a design approach. It can impose some restriction for wrapper code authors, since everything needs to be put in a single subroutine, but has nice properties like less stack trace depth and less function call overhead.


  • How to display the wrapper code being generated?

    If environment variable LOG_PERINCI_WRAPPER_CODE or package variable $Log_Perinci_Wrapper_Code is set to true, generated wrapper source code is logged at trace level using Log::Any. It can be displayed, for example, using Log::Any::App:

       perl -MLog::Any::App -MPerinci::Sub::Wrapper=wrap_sub \
       -e 'wrap_sub(sub=>sub{}, meta=>{v=>1.1, args=>{a=>{schema=>"int"}}});'

    Note that Data::Sah (the module used to generate validator code) observes LOG_SAH_VALIDATOR_CODE, but during wrapping this environment flag is currently disabled by this module, so you need to set LOG_PERINCI_WRAPPER_CODE instead.

caller() doesn't work from inside my wrapped code!

Wrapping adds at least one or two levels of calls: one for the wrapper subroutine itself, the other is for the eval trap when necessary.

This poses a problem if you need to call caller() from within your wrapped code; it will also be off by at least one or two.

The solution is for your function to use the caller() replacement, provided by Perinci::Sub::Util. Or use embedded mode, where the wrapper code won't add extra subroutine calls.



If set to 1, will log the generated wrapper code. This value is used to set $Log_Wrapper_Code if it is not already set.


Set default for wrap argument core.


Set default for wrap argument core_or_pp.


Set default for wrap argument pp.


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.


Perinci, Rinci





perlancar <>


This software is copyright (c) 2016 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.