jtools 1.1.1

A Python module that aids filtering, formatting, and transforming JSON-like objects

JTools

JTools is a robust library for interacting with JSON-like objects: providing the ability to quickly query, format, and filter JSON-like data.

There are three main components:

• Query: Extract and transform the value of nested fields.
  • Query("data.timestamp.$parse_timestamp.$attr('year')").many(items)
• Filter: Combine the querying capabilities of Query with the ability to define filtering conditions to find just the elements you want.
  • Filter(Key("data.timestamp.$parse_timestamp.$attr('year')") > 2015).many(data)
• Formatter: Take multiple queries and format them into a string
  • Formatter("Item @data.id was released in @data.timestamp.$parse_timestamp.$attr('year')").single(data[0])

A companion to the JavaScript version of this package: @blending_jake/jtools. The JavaScript version supports almost the exact same specials, filters, and formatting specification, with the goal of making it a seamless experience to go from querying/filtering/formatting in JavaScript to Python and back.

Recent Changes

• 1.1.1
  • Add antlr4-python3 requirement to setup.py so that installation will get the needed dependencies
  • Change JQL so that field and special names must only contain [-a-zA-Z0-9_]. $index can be used to get fields with prohibited characters. The change was to support more formatting use-cases, like Age: @age, DOB: @dob, which previously would have failed because the , would have been considered part of the field name.
  • Change Formatter so that fallback is just a string that is substituted for invalid queries, instead of being the entire return value. Previously, "Age: @missing" would result in None, not it results in "Age: <missing>". This change allows for better debugging as it becomes clear exactly which queries are failing.
  • Add function docstrings

Glossary

• Installation
• JQL
• Query
  • Specials
• Filter
  • Key
  • Condition
• Formatter
• Performance
• Changelog

Installation

pip install jtools

# import
from jtools import Query, Filter, Key, Condition, Formatter

JQL

JQL or the JSON Query Language is a custom built query language for JTools which supports powerful features like accessing nested fields, transforming values, and even using nested queries as arguments. The basic format of the language is:

(<field> | $<special>) (. (<field> | $<special>))*
EX: 'data', 'data.timestamp', 'data.$split', '$split.0'

field

A field is just a value that can be used as an index, like a string or integer key for a map/dict, or an integer for an array. By default, any field that can be treated as an integer will be. However, this assumes that any field containing only digits was intended to be an integer index, which isn't always the desired behavior. To stop digit-only strings from becoming integers, set convert_ints=False when creating the query/filter/formatter.

Fields can only contain the following characters: [-a-zA-Z0-9_]. However, fields with prohibited characters can still be indexed by using the $index special, so to index range[0] use $index("range[0]").

$special

A special is a function that is applied to the value that has been queried so far. There is a complete list of specials here. These specials can be passed arguments, which is one of the most powerful features of JQL. The syntax is similar to most programming languages: $<special>(<value>(, <value>)*). Just to note, $<special>() is valid, as is $<special>. Many of the specials don't require any arguments, or have default values, allowing the parenthesis to be left off.

value

A <value> can be:

[] or [<value>(, <value>)*] - List
{} or {<value>(, <value>)*} - Set
{:} or {<key>: <value>(, <key>: <value>)*} - Map/Dict/Object (see below for <key> spec)
Integer
Float
String w/ '' or ""
true
false
null
@<query> - Yep! Nested queries!

<key>:
    @<query>    
    Integer
    Float
    String
    true
    false
    null

As shown above, values and queries can be nested, so [[1, 2], ["bob"], {"Ann", 'Ralph'}, {'key': 4, 23: 5}] is valid. Additionally, the support for nesting queries is extremely powerful and allows for queries like: item.tag.$lookup(@table.colors), which, for {"item": {"tag": "product"}, "table": {"colors": {"product": "red"}}} results in "red"

Query

Query takes the power of JQL and puts it into practice querying and transforming values in JSON-like data.

Query(query, convert_ints=True, fallback=None)

• query: str | List[str] The field or fields to query
• convert_ints: Whether or not to convert any valid fields to integers
• fallback: The value that will result if a non-existent field is queried

.single(item)

Take a single item and query it using the query(ies) provided

Query(field).single(...) -> result

Query([field, field, ...]).single(...) -> [result, result, ...]

.many(items)

Take a list of items, and query each item using the query(ies) provided

Query(field).many(...) -> [result, result, ...]

Query([field, field, ...]).many(...) -> [[result, result, ...], [result, result, ...], ...]

Notes

• Lists can be indexed as long as Query(..., convert_ints=True), which is set to True by default. This allows paths like friends.0. However, convert_ints=False should be used if trying to access fields whose keys are strings containing digits, like {"index": {"0": ...}}. Use $index if you have mixed data where there are some cases where you need to use an integer and other times where the digits-only string needs to stay as a string, like in: Query("item.0.$index(0)").single({"item": {"0": ["tag"])}) -> 'tag'

• Fields can be indexed after specials, so $split.0 is completely valid

• You don't have to use () at the end of a special if there aren't any arguments, or the default arguments are acceptable.

• More specials can be added by using the class attribute .register_special() like so: Query.register_special(<name>, <func>). The function should take at least one argument, which is the current value in the query string: lambda value, *args: ...

Specials

Below is a list of all of the specials that can be used. Additionally, more can be added as discussed above.

General

• $length -> int
• $lookup(map: dict, fallback=None) -> any: Lookup the current value in the provided map/dict
• $inject(value: any) -> any: Inject a value into the query
• $print -> any: Print the current query value before continuing to pass that value along

Maps

• $keys -> list
• $values -> list
• $items -> List[tuple]
• $wildcard(next, just_value=true) -> List[any]: On a given map or list, go through all values and see if next is a defined field. If it is, then return just the value of next on that item, if just_value=true, or the entire item otherwise. This special allows a nested field to be extracted across multiple items where it it present. For example:

data = {
    "a": {"tag": "run"},
    "b": {"tag": "to-do", "other": "task"},
    "meta": None
}
Query('$wildcard("tag")').single(data)  # => ["run", "to-do"]
Query('$wildcard("tag", false)').single(data) # => [{"tag": "run"}, {"tag": "to-do", "other": "task"}]

Type Conversions

• $set -> set
• $float -> float
• $string -> str
• $int -> int
• $not -> bool: Returns !value
• $fallback(fallback) -> value or fallback: If the value is None, then it will be replaced with fallback.
• $ternary(if_true, if_false, strict=false) -> any: Return if_true if the value is truish, otherwise, return if_false. Pass true for strict if the value must be True and not just truish.

Datetime

• $parse_timestamp -> datetime: Take a Unix timestamp in seconds and return a corresponding datetime object
• $strptime(fmt=null) -> datetime: Parse a datetime string and return a corresponding datetime object. If fmt=null, then common formats will be tried. Refer to https://docs.python.org/3/library/datetime.html#strftime-strptime-behavior for formatting instructions
• $timestamp -> float: Dump a datetime object to a UTC timestamp as a float
• $strftime(fmt="%Y-%m-%dT%H:%M:%SZ") -> str: Format a datetime object as a string using fmt. Refer to https://docs.python.org/3/library/datetime.html#strftime-strptime-behavior for formatting instructions

Math / Numbers

• $add(num) -> Union[int, float]
• $subtract(num) -> Union[int, float]
• $multiply(num) -> Union[int, float]
• $divide(num) -> float
• $pow(num) -> Union[int, float]
• $abs(num) -> Union[int, float]
• $distance(other) -> float: Euler distance in N-dimensions
• $math(attr) -> any: Returns math.<attr>(value), which can be used for operations like floor, cos, sin, etc.
• $round(n=2) -> float

Strings

• $prefix(prefix) -> str: Prefix the value with the specified string
• $suffix(suffix) -> str: Concatenate a string to the end of the value
• $wrap(prefix, suffix) -> str: Wrap a string with a prefix and suffix. Combines features of above two specials.
• $strip -> str: Strip leading and trailing whitespace
• $replace(old, new) -> str: Replace all occurrences of a string
• $trim(length=50, suffix="...") -> str: Trim the length of a string
• $split(on=" ") -> List[str]: Split a string

Lists

• $sum -> Union[float, int]: Return the sum of the items in the value
• $join(sep=", ") -> str: Join a list using the specified separator
• $index(index, fallback=null) -> any: Index a list. Negative indices are allowed.
• $range(start, end=null) -> : Get a sublist. Defaults to value[start:], but an end value can be specified. Negative indices are allowed.
• $remove_nulls -> List[any]: Remove any values that are None
• $map(special, *args) -> list: Apply special to every element in the value. Arguments can be passed through to the special being used, like $map("index", "name")

Attributes

• $call(func, *args) -> any: Call a function that is on the current value, implemented as getattr(value, func)(*args)
• $attr(attr) -> any: Access an attribute of the given object, implemented as getattr(value, attr)

Filter

Filter takes the power of JQL and combines it with filtering conditions to allow lists of items to be filtered down to just those of interest. The filters can be manually built, or the Key and Condition classes can be used to simplify your code.

Filter(filters, convert_ints=True, empty_filters_response=True, missing_field_response=False)

• filters: Condition | List[dict] The filters to apply to any data. If List[dict], then the filters should be formatted as shown below.

[
    {"field": <field>, "operator": <op>, "value": <value>},

    OR

    {"or": <nested outer structure>},

    OR

    {"not": <nested outer structure>},

    ...
]
<field>: any valid `JQL` query
<op>: See list below
<value>: Anything that makes sense for the operator

• convert_ints: bool Corresponds with the argument with the same name in Query. Determines whether digit only fields are treated as integers or strings. Defaults to True.
• empty_filters_response: bool Determines what gets returned when no filters are supplied.
• missing_field_response: bool Determines the result of a filter where the field could not be found.

.single(item)

Take a single item and determine whether it satisfies the filters or not

Filter(filters).single(...) -> True/False

.many(items)

Take a list of items, and returns only those which satisfy the filters

Filter(filters).many(...) -> [result, result, ...]

Notes

{"or": [ 
    [ {filter1}, {filter2} ], 
    {filter3} 
]} === (filter1 AND filter2) OR filter3

Nesting in an or will cause those filters to be AND'd and then everything in the toplevel of that or will be OR'd.

Operators:

• >
• <
• >=
• <=
• ==
• !=
• ===: same as == in Python
• !==: same as != in Python
• in: <field> in <value>
• !in
• contains: <value> in <field>
• !contains
• interval: <field> in interval [value[0], value[1]] (closed/inclusive interval)
• !interval: <field> not in interval [value[0], value[1]]
• startswith
• endswith
• present
• !present

Key

Intended to simplify having to write {"field": <field>, "operator": <operator>, "value": value} a lot. The basic usage is: Key(<field>).<op>(<value>), or for the first six operators, the actual Python operators can be used, so Key(<field>) <op> <value>. For example: Key("meta.id").eq(12) is the same as Key("meta.id") == 12, which is the same as {"field": "meta.id", "operator": "==", "value": 12}.

The table below describes all of the functions which map to the underlying conditions, but, in addition, there is the .operator(op: str) function which can be use to build a filter. For example: Key(<field>).operator(<op>).value(<value>) is the same as {"field": <field>, "operator": <op>, "value": <value>}

Operators:

underlying operator	Key function	Python operator
>	gt	>
<	lt	<
<=	lte	<=
>=	gte	>=
==	eq	==
!=	ne	!=
===	seq	N/A
!==	sne	N/A
in	in_	N/A
!in	nin	N/A
contains	contains	N/A
!contains	not_contains	N/A
interval	interval	N/A
!interval	not_interval	N/A
startswith	startswith	N/A
endswith	endswith	N/A
present	present	N/A
!present	not_present	N/A

Condition

Intended to be used in combination with Key to make creating filters easier than manually creating the JSON. There are three conditions supported: and, or, and not. They can be manually accessed via and_(*args), or_(*args), and not_(), or through the overloaded operators &, |, and ~, respectively.

Caution: & and | bind tighter than the comparisons operators and ~ binds the tightest Key("first_name") == "John" | Key("first_name") == "Bill" is actually (Key("first_name") == ("John" | Key("first_name"))) == "Bill", not (Key("first_name") == "John") | (Key("first_name") == "Bill")

Examples

Key("state").eq("Texas") | Key("city").eq("New York")

(Key("gender") == "male") & (Key("age") >= 18) & (Key("selective_service") == False)

Key('creation_time.$parse_timestamp.$attr("year")').lt(2005).or_(
    Key('creation_time.$parse_timestamp.$attr("year")').gt(2015)
).and_(
    Key("product_id") == 15
)
# (year < 2005 OR year > 2015) AND product_id == 15

Formatter

Formatter allows fields to be queried from an object and then formatted into a string. Any JQL queries in a format string should be prefixed with @. For example, Formatter('Name: @name}').single({"name": "John Smith"}) results in Name: John Smith.

Formatter(spec, fallback="<missing>", convert_ints=True)

• spec: str The format string
• fallback: str The value that will be used in the formatted string if a query could not be performed. For example, if the field missing does exist, then the query "Age: @missing" will result in "Age: <missing>"
• convert_ints: bool Whether digit-only fields get treated as integers or strings

.single(item)

Return a formatted string or the fallback value if the query fails

.many(items)

Return a list of formatted strings or the fallback value.

Notes

The differences between Query and Formatter are:

• Query can return a value of any type, Formatter just returns strings
• Formatter supports multiple queries, end-to-end, Query does not
• All queries must be prefixed with @ with Formatter, not just when used as an argument like with Query
• Both support all the features of JQL
• Query actually can theoretically do everything Formatter does by using $prefix, $suffix, and $string. For example, '@name @age' -> 'name.$suffix(" ").$suffix(@age)'. However, the latter is much longer than the former.

Example (flattening operations):

errors = {
    "errors": {
        "Process Error": "Could not communicate with the subprocess",
        "Connection Error": "Could not connect with the database instance"
    }
}

Formatter('Errors: \n@errors.$items.$map("join", ": \\n\\t").$join("\\n")').single(errors)
# Errors:
# Process Error: 
#   Could not communicate with the subprocess
# Connection Error: 
#   Could not connect with the database instance

The above example shows a powerful usage of flattening errors into its items, then joining each item; splitting the error name and message between lines, then joining all the errors together.

Example (nested replacement):

item = {
    "x1": 1,
    "y1": 1,
    "x2": 12,
    "y2": 54
}

Formatter(
    "Midpoint: [@x2.$subtract(@x1).$divide(2), @y2.$subtract(@y1).$divide(2)]"
)
# Midpoint: [5.5, 26.5]

Additionally, the speed of formatting is very quick. The above statement can be preformed 10,000 times in around 0.36 seconds.

Performance

There are several ways to increase the performance of querying, filtering, and formatting. The performance gains can be had by limiting the amount of times a query string has to be parsed. This means that using a Query, Filter, or Formatter object multiple times will be faster then creating a new object every time.

For example:

# slower
for item in items:
    f = Query("timestamp.$parse_timestamp").single(item)
    # do other stuff

# faster
query = Query("timestamp.$parse_timestamp")
for item in items:
    f = query.single(item)
    # do other stuff

Across 10,000 runs:

• reusing Query can improve performance by 302x
• reusing Filter can improve performance by 132x
• reusing Formatter can improve performance by 377x.

Changelog

• 1.1.0

  • Rename Getter to Query to more accurately describe what the class does
  • Migrate queries to use JQL
    • The migration opens the door to nested queries in Query, allowing queries, prefixed with @ to be used as arguments to specials, or even as values in the supported argument data structures
    • Special arguments are no longer parsed as JSON, allowing features like sets, query nesting, and support for single and double quoted strings.
    • Formatter no longer uses {{}} to surround queries. Instead, all queries must be prefixed with @, so "{{name}} {{age}}" -> "@name @age". @@ must be used to get a literal @ in a formatted string: "bob@@gmail.com" -> "[e-mail]"
    • Formatter got about a 2x performance boost
  • Added $wrap(prefix, suffix) to combine $prefix and $suffix
  • Added $remove_nulls
  • Added $lookup(map, fallback=None)
  • Added $wildcard(next, just_value=True), which allows level of nesting to be "skipped", such that a list of sub-values where next is present
  • Added a fallback argument to $index
  • Added $print to display the current value in the query
  • Added $inject to allow any valid argument value to be injected into the query to be accessed and transformed by subsequent fields and specials

• 1.0.6

  • Add === and !== to match the strict equality checking needed in the JS version. The methods seq and sne have been added to Key to correspond with the new filters. === is the same as == and !== in the same as != in the Python version.
  • Rename null -> !present and !null -> present. Corresponding methods have been renamed to not_present and present. This filter will catch values that are null or undefined.
  • Make membership filters (in, contains, !in and !contains) work properly with strings, lists, dicts, and sets.
  • Remove $datetime. See below for replacement.
  • Add $call and $attr for calling a function and accessing an attribute. Can be used to replace $datetime functionality.
  • Remove Formatter.format and add Formatter.single and Formatter.many to be consistent across other classes and support formatting arrays of items.
  • Add more tests to increase coverage and do basic performance testing

• 1.0.5

  • Query strings can now start with specials to allow operations on the entire object being passed.
  • Bug fixes and more unit tests

• 1.0.4

  • Added new specials, mostly relating to time
    • $parse_timestamp
    • $datetime
    • $strptime
    • $strftime
  • Added not filtering and the interval and !interval operators
  • Made Filter consistent with Query by removing .filter() and adding .single() and .many()
  • Added fallback to Getter
  • added numerous unit tests

• 1.0.3

  • Rename Getter.get to Getter.single
  • Add Getter.many
  • Support getting multiple fields at once by changing Getter to allow Getter(<field>) and Getter([<field>, <field>, ...])
  • Change Filter's before for when there are no filters. Now, by default, all items will be returned unless Filter(..., empty_filters_response=False)