0.1.0rc5

What's new since 0.1.0rc4

Fixes (unblock rc4 users)

• pandas 2.2.x excluded (pandas>=2.0,!=2.2.*) — pivot() on ArrowDtype dictionary columns crashed type_tableview with 'Series' object has no attribute '_pa_array'; fixed upstream in pandas 2.3 (#52)
• Export tolerates non-string VALUEs — numbers from edited tableviews no longer crash the compiled CIM XML export (Expected bytes, got a 'int' object); export functions also fail early with a clear message when triplet columns are missing (#50)
• Triplets string-or-null invariant enforced at producers — tableview_to_triplets returns nullable string dtype (empty cells stay null, not "nan"), set_value_at_key normalizes; None no longer becomes the literal "None" (#55)

Export

• Schema-driven profile resolution — the schema's own ProfileMetadata (keyword / versionIRI / conformsTo) is matched against the instance header, so CGMES 2.4.15, CGMES 3.0 and NetworkCode dcat headers all resolve correctly
• export_undefined now defaults to False — internal structures (Distribution, NamespaceMap) stay out of normal exports; with export_undefined=True they emit under the http://triplets# namespace (valid, strict-parser-safe RDF/XML)
• Optional rdf:datatype annotations in CIM XML export (datatypes=True, python_lxml engine)
• N-Quads literal datatypes from the export schema ("44.84"^^<xsd:float>)
• N-Quads polars engine — single lazy expression plan, ~2x faster than the previous polars path; engine="auto" routes to it

API & naming

• tools functions follow <action>_<format>_<qualifier> (e.g. filter_triplets_by_type, diff_triplets); 0.0 names keep DeprecationWarning aliases until 0.2
• first-class autocomplete aliases: get_types_count, tableview_by_type/key/id
• multivalue=True on all three tableview functions (shared pivot core)
• cgmes_tools accepts polars / arrow / duckdb input
• set_value_at_key (corrected from the rc4 set_triplets_value_by_key misnaming)

Housekeeping

• dropped aniso8601 core dependency (pandas parses CGMES timestamps) (#30)
• artifact actions bumped to Node 24 majors (#40)
• N-Quads vs CIM XML rdflib-load benchmarks (group rdflib-load)

Full changelog: 0.1.0rc4...0.1.0rc5

0.1.0rc4

What's new since 0.1.0rc3

Performance

• CIM XML export engine dispatch (#43, closes #32) — export_to_cimxml(engine="auto") uses the compiled Arrow→pugixml extension when available: 11.5x faster on RealGrid (1.14M rows: 9.3s → 0.8s). Both engines produce data-identical output (verified by equivalence tests).

N-Quads

• Schema datatype annotations (#46, closes #31) — literals carry their xsd type from the export schema ("44.84"^^<...XMLSchema#float>); xsd:string stays plain (RDF 1.1 default), xsd:anyURI keys keep IRI handling. Also fixes IdentifiedObject.mRID being mis-exported as a urn:uuid reference. Validated end-to-end with rdflib, including referential integrity against known dangling references.

API

• tools renames (#45) — 13 functions renamed to the <action>_<format>_<qualifier> convention (e.g. filter_by_type → filter_triplets_by_type, set_VALUE_at_KEY → set_triplets_value_by_key, diff_between_triplet → diff_triplets). Old names keep working with DeprecationWarning until 0.2 — see the migration guide.
• IDE-autocomplete aliases (#45) — first-class aliases get_types_count, tableview_by_type/key/id (no deprecation, both names permanent).
• cgmes_tools renames (#44) — draw_relations_graph (typo fix), count_GeneratingUnit_types, generate_instance_ids, get_model_triplets; old names deprecated.

Interop

• cgmes_tools accepts polars, arrow, and DuckDB input (#44) — converted automatically at the package boundary (~10 ms / 1.14M rows); DataFrame results come back in the input flavor.

Docs

• New Export Architecture guide mirroring the parser guide; parser guide updated with engine aliases and logging-level debug output.

Full changelog: 0.1.0rc3...0.1.0rc4

0.1.0rc3

Note: 0.1.0rc2 on PyPI is incomplete (wheels only, no sdist) — its first publish run uploaded wheels built from a dirty checkout before the sdist was rejected, and PyPI filenames cannot be reused. Use 0.1.0rc3 instead.

What's new since 0.1.0rc1

Visualization

• Relations graphs rewritten without pyvis (#41) — self-contained HTML with vendored vis-network 9.1.13, works offline. Full-viewport graph; selecting a node opens a copyable object data table panel (deduplicated rows, source filename instead of instance UUID). draw_relations* open the browser automatically (open_browser=True).

Features & refactors

• Schema-aware N-Quads export — enums get namespace, UUIDs get urn:uuid (#38)
• DuckDB engine split into proper modules matching pandas/polars pattern
• Registry-driven accessors — _accessor.py rewritten around method-name registries (#39)
• Debug logging across all modules; parser auto-enables debug output when the logger is at DEBUG level

Release pipeline

• Untracked generated .pyc/.cpp files that made CI builds dirty and broke the versioneer version (#39)
• skip-existing: true on PyPI publish so partial publishes can be re-run (#42)

Full changelog: 0.1.0rc1...0.1.0rc3

0.1.0rc2

What's new since 0.1.0rc1

• Debug logging across all modules — replaces print() in cython parser with proper logging.getLogger(), adds debug logging for engine selection, accessor registration, and export dispatch
• Schema-aware N-Quads export — enums get namespace, UUIDs get urn:uuid (#38)
• DuckDB engine split into proper modules matching pandas/polars pattern
• Registry-driven accessors — _accessor.py rewritten around method-name registries instead of hand-written delegation methods (#39)
• Release fix — untrack generated .pyc/.cpp files that made CI builds dirty and broke the versioneer version (#39)
• Docs — suppress duplicate warnings, fix docstrings, cleanup orphan files
• CI — use absolute paths in test_parser.py for cibuildwheel compatibility

Full changelog: 0.1.0rc1...0.1.0rc2

0.1.0rc1

Release Candidate: Three-tier CIMXML parser with engine= API

Install

# Core (python_lxml_pandas engine, no extra deps)
pip install --pre triplets

# With pyarrow (enables python_lxml_arrow + cython_pugixml_arrow engines, ~12x faster)
pip install --pre triplets[arrow]

The cython_pugixml_arrow engine is a compiled C++ extension included in the wheels.
It requires pyarrow at runtime — install with triplets[arrow] to enable it.

What's new

• Three parser engines with automatic fallback (python_lxml_pandas, python_lxml_arrow, cython_pugixml_arrow)
• engine= parameter on parse() and pandas.read_RDF()
• polars.read_rdf() registered alongside pandas.read_RDF()
• pyarrow moved to optional dep (pip install triplets[arrow])
• Dictionary-encoded KEY + INSTANCE_ID columns — categories are free (no overhead)
• Cross-platform wheels with compiled cython extension (Linux x86+ARM, macOS ARM, Windows)

Performance (RealGrid ~82MB, 1.14M rows)

Engine	Install	Speed	Memory
python_lxml_pandas	pip install triplets	1,599 ms	314 MB
python_lxml_arrow	pip install triplets[arrow]	1,477 ms	145 MB
cython_pugixml_arrow	pip install triplets[arrow]	129 ms	50 MB

Backwards compatible

• pandas.read_RDF() works unchanged
• df.type_tableview() works unchanged
• Old rdf_parser.py functions emit deprecation warnings

0.0.17

What's Changed

• Updated export schema and simplified cim/xml export by @Haigutus in #17
• Missing package imports by @Haigutus in #18
• Fixed imports by @Haigutus in #19
• Imports and documentation fix by @Haigutus in #20
• Still import issues, making all init imports relative by @Haigutus in #21
• Still import issues, changed import order, added addtional resources … by @Haigutus in #22

Full Changelog: 0.0.11...0.0.17

0.0.11

What's Changed

• escapes print_duration not founding the start time if not debug mode by @mr0321 in #10

New Contributors

• @mr0321 made their first contribution in #10

Full Changelog: 0.0.10...0.0.11

Minor fix

Full Changelog: 0.0.9...0.0.10

0.0.9

What's Changed

• RDF import optimizations by @Haigutus in #1

Full Changelog: 0.0.8...0.0.9

0.0.8

1. Import and export speed optimizations and parallel processing initial support
2. Arrow backed support for reduced memory footprint

Full Changelog: 0.0.7...0.0.8