NAME
JSON::Streaming::Reader - Read JSON strings in a streaming manner
DESCRIPTION
This module is effectively a tokenizer for JSON strings. With it you can process JSON strings in customizable ways without first creating a Perl data structure from the data. For some applications, such as those where the expected data structure is known ahead of time, this may be a more efficient way to process incoming data.
SYNOPSIS
my $jsonr = JSON::Streaming::Reader->for_stream($fh);
$jsonr->process_tokens(
start_object => sub {
...
},
end_object => sub {
},
start_property => sub {
my ($name) = @_;
},
# ...
);CREATING A NEW INSTANCE
This module can operate on either an IO::Handle instance or a string.
JSON::Streaming::Reader->for_stream($fh)
Create a new instance that will read from the provided IO::Handle instance. If you want to operate on a raw Perl filehandle, you currently must wrap it up in an IO::Handle instance yourself.
JSON::Streaming::Reader->for_string(\$string)
Create a new instance that will read from the provided string. Uses IO::Scalar to make a stream-like wrapper around the string, and passes it into for_stream.
CALLBACK API
The recommended way to use this library is via the callback-based API. In this API you make a single method call on the reader object and pass it a CODE ref for each token type. The reader object will then consume the entire stream and call the callback responding to the type of each token it encounters.
An error token will be raised if an error is encountered during parsing.
For tokens that themselves have data, the data items will be passed in as arguments to the callback.
The handlers for the start_property, start_array and start_object tokens may use the skip method from the pull API, as described below, to avoid processing the remainder of the corresponding container.
$jsonr->process_tokens(%callbacks)
Read the whole stream and call a callback corresponding to each token encountered.
PULL API
A lower-level API is provided that allows the caller to pull single tokens from the stream as necessary. The callback API is implemented in terms of the pull API.
$jsonr->get_token()
Get the next token from the stream and advance. If the end of the stream is reached, this will return undef. Otherwise it returns an ARRAY ref whose first member is the token type and its subsequent members are the token type's data items, if any.
$jsonr->skip()
Quickly skip to the end of the current container. This can be used after a start_property, start_array or start_object token is retrieved to signal that the remainder of the container is not actually required. The next call to get_token will return the token that comes after the corresponding end_ token for the current container. The corresponding end_ token is never returned.
This is most useful for skipping over unrecognised properties when populating a known data structure.
It is better to use this method than to implement skipping in the caller because skipping is done using a lightweight mechanism that does not need to allocate additional memory for tokens encountered during skipping. However, since this method uses a simpler state model it may cause less-intuitive error messages to be raised if there is a JSON syntax error within the content that is skipped.
Note that errors encountered during skip are actually raised via die rather than via the return value as with get_token.
$jsonr->slurp()
Skip to the end of the current container, capturing its value. This allows you to handle a start_property, start_array or start_object token as if it were an add_-type token, dealing with its entire contents in one go.
The next call to get_token will return the token that comes after the corresponding end_ token for the current container. The corresponding end_ token is never returned.
The return value of this method call will be a Perl data structure representing the data that was skipped. This uses the same mappings as other popular Perl JSON libraries: objects become hashrefs, arrays become arrayrefs, strings and integers become scalars, boolean values become references to either 1 or 0, and null becomes undef.
This is useful if there is a part of the tree that you would rather handle via an in-memory data structure like you'd get from a non-streaming JSON parser. It allows you to mix-and-match streaming parsing and one-shot parsing within a single data stream.
Note that errors encountered during skip are actually raised via die rather than via the return value as with get_token.
If you call this when in property state it will return the value of the property and parsing will continue after the corresponding end_property. In object or array state it will return the object or array and continue after the corresponding end_object or end_array.
EVENT-BASED API
This module has an experimental event-based API which can be used to do streaming JSON processing in event-driven applications or those which do non-blocking I/O.
In event-based mode it is the caller's responsibility to obtain data and when data is available provide it to the reader for processing. When enough data is available to unambigously represent a complete, atomic token a callback function is called in a similar fashion to the callback-based API described above.
The event-based API implementation is currently somewhat hacky and inefficient. Caution is advised when making use of it in production applications, since it is currently merely a shim over the existing blocking API which may introduce strange packet-boundary bugs and other misbehavior.
JSON::Streaming::Reader->event_based(%callbacks)
Creates and returns an event-based reader. Callbacks are provided in the same way as to the process_tokens method in the callback-based API, though here there is an additional pseudo-token type called 'eof' which signals that the end of the stream has been reached.
Note that at present it is not possible to use the skip or slurp methods on an event-based reader, since their implementations expect to be able to block. This ought to be fixed in a future version.
$jsonr->feed_buffer(\$data)
The caller must call this method whenever new data becomes available for processing. A call to this method causes the reader to append the supplied data to any existing buffer and then try to consume as many tokens as possible from the buffer before returning. A callback will be run for each complete token encountered in the buffer.
If the additional data does not allow a complete token to be recognised, the reader will retain the leftover buffer and attempt parsing again at the next call to feed_buffer.
In most cases this method will be called in response to some event, such as a notification that more data is available to read on a socket.
$jsonr->signal_eof()
The caller must call this method to signal the end of the data stream. This will cause the parser to process any remaining bytes in the buffer, possibly running token callbacks in the process, and then call the special eof callback.
In most cases this method will be called in response to some event, such as a notification that a socket stream has been closed.
TOKEN TYPES
There are two major classes of token types. Bracketing tokens enclose other tokens and come in pairs, named with start_ and end_ prefixes. Leaf tokens stand alone and have add_ prefixes.
For convenience the token type names match the method names used in the "raw" API of JSON::Streaming::Writer, so it is straightforward to implement a streaming JSON normalizer by feeding the output from this module into the corresponding methods on that module. However, this module does have an additional special token type 'error' which is used to indicate tokenizing errors and does not have a corresponding method on the writer.
start_object, end_object
These token types delimit a JSON object. In a valid JSON stream an object will contain only properties as direct children, which will result in start_property and end_property tokens.
start_array, end_array
These token types delimit a JSON array. In a valid JSON stream an object will contain only values as direct children, which will result in one of the value token types described below.
start_property($name), end_property
These token types delimit a JSON property. The name of the property is given as an argument. In a valid JSON stream a start_property token will always be followed by one of the value token types which will itself be immediately followed by an end_property token.
add_string($value)
Represents a JSON string. The value of the string is passed as an argument.
add_number($value)
Represents a JSON number. The value of the number is passed as an argument.
add_boolean($value)
Represents a JSON boolean. If it's true then 1 is passed as an argument, or if false 0 is passed.
add_null
Represents a JSON null.
error($string)
Indicates a tokenization error. A human-readable description of the error is included in $string.
STREAM BUFFERING
Except in event-based mode, this module doesn't do any buffering. It expects the underlying stream to do appropriate read buffering if necessary.
In event-based mode an internal buffer is used which retains bytes that are not yet enough to unambiguously represent a complete token so that it can retry when more data is available. In this situation it is up to the caller to read from its data source in an appropriate manner, but it is best to provide as much data as possible in a single data notification.