PyJolt v2.0.0

[2.0.0] — 2026-10-02

pyjolt now behaves exactly like the reference Java JOLT implementation. It
passes all 178 cases of the reference project's own test suite (shift,
default, remove, cardinality, sort, all three modify flavours and chained
transforms), which runs in CI as tests/conformance. A spec built on the
JOLT demo site now gives the same output in
pyjolt. Before this release, 35 of those cases passed.

Every transform was rebuilt on a port of the reference path-matching engine.
The public API (Chainr, Shift(spec).apply(data), …) is unchanged, but
some 1.x behaviour that differed from JOLT has changed — see Upgrading below.

Added

• ModifyDefine / modify-define-beta: write only where a key does not
  exist.
• Shift: @ and @(n,path) as spec keys (use input values as keys, e.g.
  turn [{"name": k, "value": v}] into {k: v}), & in spec keys, \
  escaping of special characters, [#n] and [@(n,path)] array indexes,
  "" to write to the output root.
• Default: a|b keys, and key[] for arrays (including a top-level
  array input).
• Modify: all reference functions (isPresent, notNull, isNull,
  squashDuplicates, sort, divide, divideAndRound, intSubtract, …),
  + / ~ / _ per-key mode prefixes, ? "only if present" suffix, [n]
  array indexes, ^path context lookups, and custom functions via
  ModifyOverwrite(spec, functions={...}). pyjolt.MISSING is exported for
  custom functions that produce no value.
• Chainr: apply(data, context=...), and JOLT's Java class names
  (com.bazaarvoice.jolt.Shiftr, …) as operation names.
• Remove: a|b keys, multiple * in a key, removing array indexes.
• Sort: keys starting with ~ sort first, as in JOLT.
• A reference-conformance test suite (tests/conformance).

Changed — upgrading from 1.x

• [] always appends a new element. In 1.x, "out[].a" and "out[].b"
  written under the same * match were grouped into one object; that was a
  pyjolt-only behaviour. Write to a shared index instead:
  "out[&1].a", "out[&1].b".
• Shift returns None when nothing matches (1.x returned {}).
• Function arguments. "=fn" still applies fn to the current value, but
  "=fn(args)" now uses only the given arguments, as in JOLT. 1.x passed the
  current value as a hidden first argument. Reference the current value with
  @(1,key): "=concat(@(1,name),'-x')", "=min(@(1,n),10)",
  "=split(',',@(1,s))". The split separator is a regular expression.
• Function results follow JOLT. For example =toBoolean only converts
  "true"/"false", =toInteger on 3.7 gives 3, and a function that
  can't produce a value leaves the field unchanged.
• Unknown modify functions emit a UserWarning and produce no value, so a
  list of alternatives falls through to the next one (1.x raised SpecError).
• Default on arrays requires JOLT's key[] syntax:
  {"items[]": {"*": {...}}}. A plain {"items": {"*": {...}}} no longer
  reaches into a list, and a top-level list input needs a "*" or index spec.
• Cardinality: MANY turns null into [] (1.x gave [null]).
  ONE/MANY are case-sensitive.
• Invalid specs are rejected as in JOLT: an empty {} spec for shift or
  cardinality, # in a shift output path, or array indexes mixed with object
  keys in a modify spec raise SpecError.
• Modify uses JOLT's function names =toUpper / =toLower; the 1.x names
  =toUpperCase / =toLowerCase still work. The other 1.x-only functions
  (toFloat, floatSum, sum, sqrt, not, coalesce, startsWith,
  endsWith, contains, indexOf) remain available, with JOLT-style
  arguments.

Notes

• Transforms still never modify the input you pass in. Chainr now copies the
  input once instead of once per step.
• The engine does more work per key than 1.x did; expect roughly 1.4× the
  runtime of 1.2.2 on large inputs.

PyJolt v1.2.2

[1.2.2] — 2026-10-02

Fixed

• Shift Transform: Output paths now support explicit array indices such as
  items.[&2].field, items[&1].field, list[0] and grid[&1][&0].
  Previously the bracketed index was written as a literal object key
  (e.g. "[0]") instead of building an array
  (#8).

PyJolt v1.2.1

[1.2.1] — 2026-04-10

Fixed

• License: Updated to official Apache 2.0 text and standardised metadata
  for better PyPI compatibility.

PyJolt v1.2.0

[1.2.0] — 2026-04-10

Added

• CLI: Introduced a command-line interface (pyjolt) for running
  transformations from the shell. Supports file/stdin input and output.

PyJolt v1.1.0

[1.1.0] — 2026-04-10

Fixed

• Shift Transform: Improved JOLT spec compliance and stability in complex
  scenarios (multiple output paths, shared slot coordination).
• Code Quality: Fixed 60+ linting violations (Ruff) and resolved all Mypy
  type errors in core transforms.
• Tests: Cleaned up unused imports and standardized formatting across the
  entire test suite.

Added

• Complex Scenarios: Added dedicated test suite for complex JOLT-like
  transformations (list-to-object mapping, deep nested array append).

PyJolt v1.0.0

1.0.0 — 2024-04-10

Added

Core transforms (full Java JOLT parity)

• Shift — re-map fields with full spec language support:

  • Literal key matching
  • Wildcard * matching (prefix/suffix combinable, e.g. prefix_*_suffix)
  • OR patterns (a|b)
  • Self-reference @ — uses the current input node as its own value
  • $ / $N — emits the matched key name N levels up as a value
  • #literal — emits a constant string as a value
  • Output path tokens: &/&N back-references, &(N,M) capture groups,
    @(N,path) value lookups, [] array-append
  • Array-of-objects slot coordination — multiple fields from the same wildcard
    iteration land in the same output list element
  • Multiple output paths (list of strings)

• Default — fill absent or null fields; wildcard * applies to every key;
  list-aware ({"key": {"*": sub_spec}} applies sub_spec to each list element)

• Remove — delete specified fields; "*" removes all keys at a level

• Sort — recursively sort all dict keys alphabetically

• Cardinality — enforce ONE (scalar) or MANY (list) cardinality

• ModifyOverwrite / ModifyDefault — apply functions or literal values to
  fields; 39 built-in functions covering type conversions, string manipulation,
  numeric operations, and collection utilities:
  toInteger, toLong, toDouble, toFloat, toString, toBoolean,
  trim, toUpperCase, toLowerCase, abs, min, max, intSum,
  doubleSum, longSum, floatSum, sum, avg, sqrt, not,
  size, concat, join, split, leftPad, rightPad, substring,
  startsWith, endsWith, contains, squashNulls, recursivelySquashNulls,
  toList, firstElement, lastElement, elementAt, indexOf, coalesce,
  noop

• Chainr — chain multiple transforms sequentially; supports both direct
  instantiation and JOLT-spec lists (from_spec)

Packaging

• PEP 561 py.typed marker — full inline type annotations
• pyproject.toml with hatchling build backend, classifiers, URLs
• Python 3.10–3.13 support

Tests

• 193 tests across all transforms and real-world integration scenarios