Skip to content

CHANGELOG.md

Latest commit

 

History

History
208 lines (155 loc) · 9.62 KB

CHANGELOG.md

File metadata and controls

208 lines (155 loc) · 9.62 KB

Changelog

All notable changes to this project will be documented in this file.

This project uses PEP 440 versioning with release candidates (rcN) for pre-release versions.

[2.0.0rc7] - 2026-03-24

Changed

  • BREAKING: Type resolution is now async and contextual; can_cast_to() and cast_to() receive a ValidationContext enabling context-dependent type compatibility (#94)
  • All node hooks (on_node_expand, on_node_finish, on_node_error) now consistently receive the post-cast input (#94)

[2.0.0rc6] - 2026-03-24

Added

  • Deep output field selection on edges: output fields can now be selected using dot-separated paths to extract nested values (#90)

[2.0.0rc5] - 2026-03-01

Added

  • remove_node_class and remove_value_class methods on registry builders; in lazy builders, removals are applied after all additions at build time so a removal always wins regardless of call order (#82)

Changed

  • BREAKING: Workflow execution methods now return WorkflowExecutionResult instead of tuple[WorkflowErrors, DataMapping]. The result object unifies output, errors, and node yields into a single value with a status field (SUCCESS, ERROR, or YIELDED) (#84)
  • BREAKING: WorkflowYield exception removed; yields are now surfaced via WorkflowExecutionResult and the on_workflow_yield context hook (previously on_workflow_yield received a WorkflowYield exception, now it receives partial_output and node_yields directly) (#84)
  • on_node_start hook now receives the fully type-cast input instead of the raw pre-cast input (#81)

Fixed

  • Data helper methods (to_dict, to_value_schema, field_annotations, only_field) moved to module-level functions (get_data_dict, get_data_schema, get_field_annotations, get_only_field) to eliminate the risk of user-defined field names shadowing them (#81)
  • BaseValueSchema.value_type alias (x-value-type) now handled via explicit serializer/validator instead of Pydantic field aliases, fixing pyright errors (#83)
  • Fixed pyright errors in constrained sequence and map value builders (#83)

[2.0.0rc4] - 2026-02-25

Added

  • on_workflow_yield context hook called when a workflow raises WorkflowYield; receives partial_output from nodes that completed before the yield (#79)
  • node_yields parameter on on_workflow_error hook to expose any ShouldYield exceptions collected before the error (#79)
  • AddNode is now variadic: accepts any number of inputs (≥2) via a num_arguments parameter, with input fields named using an infinite alphabetic sequence (a, b, …, z, aa, ab, …) (#80)
  • Titles and descriptions on all built-in node input, output, and parameter fields for non-technical display (#78)

Changed

  • Error state takes precedence over yield state in ParallelExecutionAlgorithm CONTINUE mode: on_workflow_error is called instead of on_workflow_yield when both occur (#79)

Fixed

  • ParallelExecutionAlgorithm CONTINUE mode no longer raises a spurious UserException for missing output when errors are present; partial output is computed directly and forwarded to on_workflow_error (#79)

[2.0.0rc3] - 2026-02-23

Added

  • ShouldYield / WorkflowYield control-flow signals for nodes that dispatch external work and cannot return a value in the current execution (#77)
  • on_node_yield(node, input, exception) context hook, called once per yielded node before WorkflowYield is raised (#77)
  • on_node_expand context hook called when a node's run() returns a Workflow instead of a DataMapping (#74)
  • Constrained Value subclasses: build_value_cls() on primitive schema classes now enforces JSON Schema keywords (minimum, maxLength, minItems, etc.) at runtime via Pydantic Field constraints (#76)
  • x-value-type JSON Schema extension field for stable Value type detection; title retained as a backwards-compatible fallback (#76)

Fixed

  • $defs key collisions when multiple constrained variants of the same base Value type appear in a single DataValue schema (#76)
  • Bug in ParallelExecutionAlgorithm where yielded nodes were missing from the ready-node exclusion filter, which would have caused an infinite re-dispatch loop (#77)

2.0.0rc2 - 2026-02-13

Added

  • Adaptive ForEachNode wiring based on inner workflow I/O (#72)
  • FieldSchemaMappingValue convenience data type for string-field schema mappings (#71)
  • First-class support for CSV, DocX, and Markdown file types (#69)
  • Iterable APIs and typed extend() method for registries (#64)
  • Comprehensive tests for round-trip type (de)serialization and failure cases (#73)

Changed

  • Schemas now allow extra fields (#70)
  • Updated cryptography dependency from 46.0.3 to 46.0.5

2.0.0rc1 - 2025-02-05

Added

  • WorkflowEngine class with node and workflow deserialization capabilities
  • Execution capabilities to WorkflowEngine

Changed

  • BREAKING: Migrated workflow format from InputEdge/OutputEdge to InputNode/OutputNode
  • Moved node/edge validation from Workflow to WorkflowEngine
  • Updated examples and documentation for new InputNode/OutputNode format

1.2.0rc1 - 2025-02-01

Added

  • Generic ModelValue[M] type for arbitrary Pydantic models
  • ExtractionResultValue type
  • Comprehensive tests for NodeRegistry implementations
  • Coding best practices to CLAUDE.md about explicit methods vs dunder overrides

Changed

  • Refactored ValueRegistry with builder pattern and comprehensive tests
  • Updated codebase to use new ValueRegistry API

Fixed

  • Pytest warning by renaming TestValue to ExampleValue

1.1.0rc1 - 2025-01-30

Added

  • Workflow migration utilities for safely loading and migrating workflows (load_workflow_with_migration, clean_edges_after_migration)
  • Typecasts from JSONValue to primitive types (NullValue, BooleanValue, IntegerValue, FloatValue)
  • Optional input_type and output_type fields on Workflow for explicit type declarations
  • Logging for previously unlogged exception handlers
  • Project documentation: getting started guide, architecture overview, node reference, value types, execution, contexts

Fixed

  • NodeExpansionException property setter bug
  • Unlogged exceptions in execution error handlers now produce proper log output

Changed

  • Version bump from 1.0.0rc2 to 1.1.0rc1
  • Documentation URL updated to point to GitHub docs

1.0.0rc2 - 2025-01-15

Changed

  • Converted optional dependencies (pyright, pytest, ruff, etc.) into dev dependencies
  • Updated to Python 3.12+ minimum
  • Made type checking non-blocking in CI
  • Improved release script

1.0.0rc1 - 2025-01-10

Added

  • PyPI publishing via GitHub Actions
  • Node retry system with exponential backoff (ShouldRetry, RetryTracker)
  • Rate limiting per node type (RateLimitConfig, RateLimiter, RateLimitRegistry)
  • Node versioning and migration system (Migration, MigrationRegistry, MigrationRunner)
  • Parallel node execution algorithm (ParallelExecutionAlgorithm) with eager dispatch
  • Value typecasting graph visualization in README

Changed

  • Registration logs changed from INFO to DEBUG level
  • Migrated from Poetry to uv for dependency management

0.3.3 - 2024-12-20

Added

  • Release script for streamlined publishing

0.3.2 - 2024-12-15

Added

  • JSON type support
  • CLAUDE.md for Claude Code guidance

Fixed

  • Double-typecast issue with Files to JSONFile

0.3.1 - 2024-12-10

Added

  • MIME type support for file values
  • Iteration and property accessors on collection types to reduce .root usage
  • __getitem__ access on SequenceValue and StringMapValue
  • Direct equality comparison with Values

Fixed

  • MIME type inconsistencies
  • Type inconsistencies in file handling
  • Missing exports

0.3.0 - 2024-12-01

Added

  • ForEachNode for workflow iteration over sequences
  • IfNode and IfElseNode for conditional branching
  • Composite node system (ExpandSequenceNode, GatherSequenceNode, ExpandMappingNode, GatherMappingNode, ExpandDataNode, GatherDataNode)
  • Serializable metadata for node types
  • Input/output schemas on nodes
  • Support for type unions in JSON schemas

Changed

  • Require semantic versioning in node type definitions

0.2.0 - 2024-11-15

Added

  • Asynchronous workflow execution
  • Value type casting system (async)
  • Error handling with partial results (WorkflowErrors)

Changed

  • All node I/O now uses Value types
  • Workflows are fully asynchronous

[0.1.x] - 2024-10-01

Added

  • Initial workflow engine implementation
  • Core types: Node, Workflow, Edge, Context, Value, Data
  • Storage backends: SupabaseContext, LocalContext, InMemoryContext
  • Basic arithmetic, constant, text, and error nodes
  • Topological execution algorithm