Download the PHP package halaxa/json-machine without Composer

Very easy to use and memory efficient drop-in replacement for inefficient iteration of big JSON files or streams for PHP >=7.2. See TL;DR. No dependencies in production except optional ext-json. README in sync with the code

• TL;DR
• Introduction
• Parsing JSON documents
  • Parsing a document
  • Parsing a subtree
  • Parsing nested values in arrays
  • Parsing a single scalar value
  • Parsing multiple subtrees
  • What is JSON Pointer anyway?
• Options
• Parsing streaming responses from a JSON API
  • GuzzleHttp
  • Symfony HttpClient
• Tracking the progress
• Decoders
  • Available decoders
• Error handling
  • Catching malformed items
• Parser efficiency
  • Streams / files
  • In-memory JSON strings
• Troubleshooting
  • "I'm still getting Allowed memory size ... exhausted"
  • "That didn't help"
  • "I am still out of luck"
• Installation
• Development
  • Non containerized
  • Containerized
• Support
• License

TL;DR

Random access like $users[42] is not yet possible. Use above-mentioned foreach and find the item or use JSON Pointer.

Count the items via iterator_count($users). Remember it will still have to internally iterate the whole thing to get the count and thus will take about the same time.

Requires ext-json if used out of the box. See Decoders.

Follow CHANGELOG.

Introduction

JSON Machine is an efficient, easy-to-use and fast JSON stream/pull/incremental/lazy (whatever you name it) parser based on generators developed for unpredictably long JSON streams or documents. Main features are:

• Constant memory footprint for unpredictably large JSON documents.
• Ease of use. Just iterate JSON of any size with foreach. No events and callbacks.
• Efficient iteration on any subtree of the document, specified by JSON Pointer
• Speed. Performance critical code contains no unnecessary function calls, no regular expressions and uses native json_decode to decode JSON document items by default. See Decoders.
• Parses not only streams but any iterable that produces JSON chunks.
• Thoroughly tested. More than 200 tests and 1000 assertions.

Parsing JSON documents

Parsing a document

Let's say that fruits.json contains this huge JSON document:

It can be parsed this way:

Parsing a json array instead of a json object follows the same logic. The key in a foreach will be a numeric index of an item.

If you prefer JSON Machine to return arrays instead of objects, use new ExtJsonDecoder(true) as a decoder.

Parsing a subtree

If you want to iterate only results subtree in this fruits.json:

use JSON Pointer /results as pointer option:

Note:

Value of results is not loaded into memory at once, but only one item in results at a time. It is always one item in memory at a time at the level/subtree you are currently iterating. Thus, the memory consumption is constant.

Parsing nested values in arrays

The JSON Pointer spec also allows to use a hyphen (-) instead of a specific array index. JSON Machine interprets it as a wildcard which matches any array index (not any object key). This enables you to iterate nested values in arrays without loading the whole item.

Example:

To iterate over all colors of the fruits, use the JSON Pointer "/results/-/color".

Parsing a single scalar value

You can parse a single scalar value anywhere in the document the same way as a collection. Consider this example:

Get the scalar value of lastModified key like this:

When parser finds the value and yields it to you, it stops parsing. So when a single scalar value is in the beginning of a gigabytes-sized file or stream, it just gets the value from the beginning in no time and with almost no memory consumed.

The obvious shortcut is:

Single scalar value access supports array indices in JSON Pointer as well.

Parsing multiple subtrees

It is also possible to parse multiple subtrees using multiple JSON Pointers. Consider this example:

To iterate over all berries and citrus fruits, use the JSON pointers ["/berries", "/citrus"]. The order of pointers does not matter. The items will be iterated in the order of appearance in the document.

What is JSON Pointer anyway?

It's a way of addressing one item in JSON document. See the JSON Pointer RFC 6901. It's very handy, because sometimes the JSON structure goes deeper, and you want to iterate a subtree, not the main level. So you just specify the pointer to the JSON array or object (or even to a scalar value) you want to iterate and off you go. When the parser hits the collection you specified, iteration begins. You can pass it as pointer option in all Items::from* functions. If you specify a pointer to a non-existent position in the document, an exception is thrown. It can be used to access scalar values as well. JSON Pointer itself must be a valid JSON string. Literal comparison of reference tokens (the parts between slashes) is performed against the JSON document keys/member names.

Some examples:

Options

Options may change how a JSON is parsed. Array of options is the second parameter of all Items::from* functions. Available options are:

• pointer - A JSON Pointer string that tells which part of the document you want to iterate.
• decoder - An instance of ItemDecoder interface.
• debug - true or false to enable or disable the debug mode. When the debug mode is enabled, data such as line, column and position in the document are available during parsing or in exceptions. Keeping debug disabled adds slight performance advantage.

Parsing streaming responses from a JSON API

A stream API response or any other JSON stream is parsed exactly the same way as file is. The only difference is, you use Items::fromStream($streamResource) for it, where $streamResource is the stream resource with the JSON document. The rest is the same as with parsing files. Here are some examples of popular http clients which support streaming responses:

GuzzleHttp

Guzzle uses its own streams, but they can be converted back to PHP streams by calling \GuzzleHttp\Psr7\StreamWrapper::getResource(). Pass the result of this function to Items::fromStream function, and you're set up. See working GuzzleHttp example.

Symfony HttpClient

A stream response of Symfony HttpClient works as iterator. And because JSON Machine is based on iterators, the integration with Symfony HttpClient is very simple. See HttpClient example.

Tracking the progress (with debug enabled)

Big documents may take a while to parse. Call Items::getPosition() in your foreach to get current count of the processed bytes from the beginning. Percentage is then easy to calculate as position / total * 100. To find out the total size of your document in bytes you may want to check:

• strlen($document) if you parse a string
• filesize($file) if you parse a file
• Content-Length http header if you parse a http stream response
• ... you get the point

If debug is disabled, getPosition() always returns 0.

Decoders

Items::from* functions also accept decoder option. It must be an instance of JsonMachine\JsonDecoder\ItemDecoder. If none is specified, ExtJsonDecoder is used by default. It requires ext-json PHP extension to be present, because it uses json_decode. When json_decode doesn't do what you want, implement JsonMachine\JsonDecoder\ItemDecoder and make your own.

Available decoders

• ExtJsonDecoder - Default. Uses json_decode to decode keys and values. Constructor has the same parameters as json_decode.

• PassThruDecoder - Does no decoding. Both keys and values are produced as pure JSON strings. Useful when you want to parse a JSON item with something else directly in the foreach and don't want to implement JsonMachine\JsonDecoder\ItemDecoder. Since 1.0.0 does not use json_decode.

Example:

• ErrorWrappingDecoder - A decorator which wraps decoding errors inside DecodingError object thus enabling you to skip malformed items instead of dying on SyntaxError exception. Example:

Error handling

Since 0.4.0 every exception extends JsonMachineException, so you can catch that to filter any error from JSON Machine library.

Skipping malformed items

If there's an error anywhere in a json stream, SyntaxError exception is thrown. That's very inconvenient, because if there is an error inside one json item you are unable to parse the rest of the document because of one malformed item. ErrorWrappingDecoder is a decoder decorator which can help you with that. Wrap a decoder with it, and all malformed items you are iterating will be given to you in the foreach via DecodingError. This way you can skip them and continue further with the document. See example in Available decoders. Syntax errors in the structure of a json stream between the iterated items will still throw SyntaxError exception though.

Parser efficiency

The time complexity is always O(n)

Streams / files

TL;DR: The memory complexity is O(2)

JSON Machine reads a stream (or a file) 1 JSON item at a time and generates corresponding 1 PHP item at a time. This is the most efficient way, because if you had say 10,000 users in JSON file and wanted to parse it using json_decode(file_get_contents('big.json')), you'd have the whole string in memory as well as all the 10,000 PHP structures. Following table shows the difference:

This means, that JSON Machine is constantly efficient for any size of processed JSON. 100 GB no problem.

In-memory JSON strings

TL;DR: The memory complexity is O(n+1)

There is also a method Items::fromString(). If you are forced to parse a big string, and the stream is not available, JSON Machine may be better than json_decode. The reason is that unlike json_decode, JSON Machine still traverses the JSON string one item at a time and doesn't load all resulting PHP structures into memory at once.

Let's continue with the example with 10,000 users. This time they are all in string in memory. When decoding that string with json_decode, 10,000 arrays (objects) is created in memory and then the result is returned. JSON Machine on the other hand creates single structure for each found item in the string and yields it back to you. When you process this item and iterate to the next one, another single structure is created. This is the same behaviour as with streams/files. Following table puts the concept into perspective:

The reality is even better. Items::fromString consumes about 5x less memory than json_decode. The reason is that a PHP structure takes much more memory than its corresponding JSON representation.

Troubleshooting

"I'm still getting Allowed memory size ... exhausted"

One of the reasons may be that the items you want to iterate over are in some sub-key such as "results" but you forgot to specify a JSON Pointer. See Parsing a subtree.

"That didn't help"

The other reason may be, that one of the items you iterate is itself so huge it cannot be decoded at once. For example, you iterate over users and one of them has thousands of "friend" objects in it. Use PassThruDecoder which does not decode an item, get the json string of the user and parse it iteratively yourself using Items::fromString().

"I am still out of luck"

It probably means that the JSON string $user itself or one of the friends are too big and do not fit in memory. However, you can try this approach recursively. Parse "/friends" with PassThruDecoder getting one $friend json string at a time and then parse that using Items::fromString()... If even that does not help, there's probably no solution yet via JSON Machine. A feature is planned which will enable you to iterate any structure fully recursively and strings will be served as streams.

Installation

Using Composer

Without Composer

Clone or download this repository and add the following to your bootstrap file:

Development

Clone this repository. This library supports two development approaches:

1. non containerized (PHP and composer already installed on your machine)
2. containerized (Docker on your machine)

Non containerized

Run composer run -l in the project dir to see available dev scripts. This way you can run some steps of the build process such as tests.

Containerized

Install Docker and run make in the project dir on your host machine to see available dev tools/commands. You can run all the steps of the build process separately as well as the whole build process at once. Make basically runs composer dev scripts inside containers in the background.

make build: Runs complete build. The same command is run via GitHub Actions CI.

Support

Do you like this library? Star it, share it, show it :) Issues and pull requests are very welcome.

License

Apache 2.0

Cogwheel element: Icons made by TutsPlus from www.flaticon.com is licensed by CC 3.0 BY

Table of contents generated with markdown-toc