Data::StreamDeserializer(3) non-blocking deserializer.

SYNOPSIS


my $sr = new Data::StreamDeserializer
data => $very_big_dump;
... somewhere
unless($sr->next) {
# deserialization hasn't been done yet
}
...
if ($sr->next) {
# deserialization has been done
...
if ($sr->is_error) {
printf "%s\n", $sr->error;
printf "Unparsed string tail: %s\n", $sr->tail;
}
my $result = $sr->result; # first deserialized object
my $result = $sr->result(first); # the same
my $results = $sr->result('all'); # all deserialized objects
# (ARRAYREF)
}
# stream deserializer
$sr = new Data::StreamDeserializer;
while(defined (my $block = read_next_data_block)) {
$sr->next($block);
...
}
$sr->next(undef); # eof signal
until ($sr->next) {
... do something
}
# all data were parsed

DESCRIPTION

Sometimes You need to deserialize a lot of data. If You use 'eval' (or Safe->reval, etc) it can take You too much time. If Your code is executed in event machine it can be inadmissible. So using the module You can deserialize Your stream progressively and do something else between deserialization itearions.

Recognized statements

HASHES

 { something }

ARRAYS

 [ something ]

REFS

 \ something
 \[ ARRAY ]
 \{ HASH }

Regexps

 qr{something}

SCALARS

 "something"
 'something'
 q{something}
 qq{something}

METHODS

new

Creates new deserializer. It can receive a few named arguments:

block_size

The size of block which will be serialized in each 'next' cycle. Default value is 512 bytes.

data

If You know (have) all data to deserialize before constructing the object, You can use this argument.

NOTE: You must not use the function part or next with arguments if You used this argument.

block_size

Set/get the same field.

part

Append a part of input data to serialize. If there is no argument (or undef), deserializer will know that there will be no data in the future.

next

Processes to parse next block_size bytes. Returns TRUE if an error was detected or all input datas were parsed.

next_object

The same as next but returns true after new object is found. Drop previous results.

For example You have the string:

    $str = "1, 2, [ 0, 1 ], { 'a' => 'b' }";

You can extract objects:

    my $dsr = new Data::StreamDeserializer data => $str;
    1 until $dsr->next_object;
    my $first = $dsr->result;       # scalar: 1
    1 until $dsr->next_object;
    my $second = $dsr->result;      # scalar: 2
    1 until $dsr->next_object;
    my $third = $dsr->result;       # arrayref: [ 0, 1 ]
    1 until $dsr->next_object;
    my $third = $dsr->result;       # hashref: { 'a' => 'b' }

skip_divider

If You have a string:

    Object Object Object

(there are no dividers between objects), You can call skip_divider after fetching the next object.

Example:

    $str = "1 2 [ 0, 1 ]{ 'a' => 'b' }";
    my $dsr = new Data::StreamDeserializer data => $str;
    1 until $dsr->next_object;
    my $first = $dsr->result;       # scalar: 1
    $dsr->skip_divider;
    1 until $dsr->next_object;
    my $second = $dsr->result;      # scalar: 2
    $dsr->skip_divider;
    1 until $dsr->next_object;
    my $third = $dsr->result;       # arrayref: [ 0, 1 ]

Important: You can't skip dividers inside nested object. The function will croak if You call it in the point that isn't between objects.

is_error

Returns TRUE if an error was detected.

error

Returns error string.

tail

Returns unparsed data.

result

Returns result of parsing. By default the function returns only the first parsed object.

You can call the function with argument 'all' to get all parsed objects. In this case the function will receive ARRAYREF.

is_done

Returns TRUE if all input data were processed or an error was found. If You didn't call part without arguments, and didn't call next or next_object with undef the function could return TRUE only if an error occured.

PRIVATE METHODS

_push_error

Pushes error into deserializer's error stack.

BENCHMARKS

This module is almost fully written using XS/C language. So it works a bit faster or slowly than CORE::eval.

You can try a few scripts in benchmark/ directory. There are a few test arrays in this directory.

Here are a few test results of my system.

Array which contains 100 hashes:

It works faster than eval:

    $ perl benchmark/ds_vs_eval.pl -n 1000 -b 512 benchmark/tests/01_100x10
    38296 bytes were read
    First deserializing by eval... done
    First deserializing by Data::DeSerializer... done
    Check if deserialized objects are same... done
    Starting 1000 iterations for eval... done (3.755 seconds)
    Starting 1000 iterations for Data::StreamDeserializer... done (3.059 seconds)
    Eval statistic:
            1000 iterations were done
            maximum deserialization time: 0.0041 seconds
            minimum deserialization time: 0.0035 seconds
            average deserialization time: 0.0036 seconds
    StreamDeserializer statistic:
            1000 iterations were done
            75000 SUBiterations were done
            512 bytes in one block in one iteration
            maximum deserialization time: 0.0045 seconds
            minimum deserialization time: 0.0028 seconds
            average deserialization time: 0.0029 seconds
            average subiteration time:    0.00004 seconds

Array which contains 1000 hashes:

It works slowly than eval:

    $ perl benchmark/ds_vs_eval.pl -n 1000 -b 512 benchmark/tests/02_1000x10
    355623 bytes were read
    First deserializing by eval... done
    First deserializing by Data::DeSerializer... done
    Check if deserialized objects are same... done
    Starting 1000 iterations for eval... done (43.920 seconds)
    Starting 1000 iterations for Data::StreamDeserializer... done (71.668 seconds)
    Eval statistic:
            1000 iterations were done
            maximum deserialization time: 0.0490 seconds
            minimum deserialization time: 0.0416 seconds
            average deserialization time: 0.0426 seconds
    StreamDeserializer statistic:
            1000 iterations were done
            689000 SUBiterations were done
            512 bytes in one block in one iteration
            maximum deserialization time: 0.0773 seconds
            minimum deserialization time: 0.0656 seconds
            average deserialization time: 0.0690 seconds
            average subiteration time:    0.00010 seconds

You can see, that one block is parsed in a very short time period. So You can increase block_size value to reduce total parsing time.

If block_size is equal string size the module works two times faster than eval:

    $ perl benchmark/ds_vs_eval.pl -n 1000 -b 355623 benchmark/tests/02_1000x10
    355623 bytes were read
    First deserializing by eval... done
    First deserializing by Data::DeSerializer... done
    Check if deserialized objects are same... done
    Starting 1000 iterations for eval... done (44.456 seconds)
    Starting 1000 iterations for Data::StreamDeserializer... done (19.702 seconds)
    Eval statistic:
            1000 iterations were done
            maximum deserialization time: 0.0474 seconds
            minimum deserialization time: 0.0423 seconds
            average deserialization time: 0.0431 seconds
    StreamDeserializer statistic:
            1000 iterations were done
            1000 SUBiterations were done
            355623 bytes in one block in one iteration
            maximum deserialization time: 0.0179 seconds
            minimum deserialization time: 0.0168 seconds
            average deserialization time: 0.0171 seconds
            average subiteration time:    0.01705 seconds

AUTHOR

Dmitry E. Oboukhov, <[email protected]>

COPYRIGHT AND LICENSE

Copyright (C) 2011 by Dmitry E. Oboukhov

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.10.1 or, at your option, any later version of Perl 5 you may have available.

VCS

The project is placed in my git repo. See here: <http://git.uvw.ru/?p=data-stream-deserializer;a=summary>