Skip to content

CHANGELOG.md

Latest commit

 

History

History
83 lines (74 loc) · 4.67 KB

CHANGELOG.md

File metadata and controls

83 lines (74 loc) · 4.67 KB
Raw

Changelog

0.6.1 — 2026-09-02

Fixed

  • The build doc carries AFFECTS_PRIMARY / AFFECTS_TOTAL. 0.6.0 made those flags decide whether a purchase contributes to a temporal characteristic, but the (legacy, curated) build doc did not emit them — so a consumer that stores the doc relationally and rebuilds from it took the class default of True and counted purchases HD keeps out of the character's totals. Measured downstream: Cobra.hdc carries a +2 DCV power marked AFFECTS_TOTAL="No", and the database rebuilt him at DCV 12 where the canonical load says 10. Emitted only when False, since True is the default on both, so a character with nothing situational produces the same doc as before. Read from the raw attributes rather than the affect_* properties — affect_total's getter writes to affects_total, and serialising a character is not allowed to change it.

    This is the flag-carrying half of 0.6.0's behaviour change. Consumers persisting the doc need columns for both, or they inherit the old answer.

0.6.0 — 2026-09-02

The theme is one source of truth for what a build says. Three places had copies of a rule the engine already owned, and each copy was quietly wrong.

Added

  • A front door and a back door. load_build(source, format=…) and hero.export(format=…), with formats registered (@import_format / @export_format) so adding one is additive rather than an edit to a dispatcher. Formats: hdc (bytes) and json (dict).
  • kirby_cost.model.modifiers — has_modifier / modifier_levels / find_modifier, the one walk answering "does this modifier bind this purchase", with recursion through containers and inheritance from an enclosing purchase (minus PRIVATE modifiers).
  • LoadedHero.characteristic_states(xmlids) — the whole stat block from one walk of the purchases. characteristic_state(xmlid) is that call with a single xmlid.
  • tests/test_build_doc_fidelity.py — the JSON chain held to the same attribute-level property as the .hdc chain, sharing one survey (tests/export_survey.py) and carrying its own shrink-only ledger.

Changed — this moves numbers

  • A purchase HD keeps out of the character's totals no longer contributes to a temporal characteristic. AFFECTS_PRIMARY / AFFECTS_TOTAL is HD's own record of whether a purchase raises the characteristic or merely sits on the sheet as a situational ability, and it was not being read. Gorgon's "Tail" +20 STR is AFFECTS_TOTAL="No" with its limitation aliased "Only With Tail" — a restrainable limb — and it was making him STR 80 instead of 60 in every calculation, thrown-object damage included. Ravel's pooled "Reinforced String" +30 STR is the same shape.
  • JSON is now a transport encoding of the HDC element tree, not a curated subset. The build doc was hand-written in both directions while the XML side wrote from declared descriptors, so five fields had been lost one at a time (TEXT, NOTES, a power's NAME, a modifier's ALIAS, AFFECTS_PRIMARY / AFFECTS_TOTAL) — all cost-neutral, and the doc's only gate compared summed cost. Measured over 794 corpus characters, .hdc → hero → json → hero → .hdc went from 0/794 clean to 791/794, the same score the .hdc chain gets, with the same two matches_hd entries.
  • HD's element IDs survive the JSON trip. They were replaced by synthetic O<n>, so a character came back as a document HD would not recognise, every PARENTID target renamed. extract_costs keys per_object by the same rule, or those keys join to nothing.

Fixed

  • BuildNode had no .attrib, so a campaign's whole RULES block took the character down on the JSON path while the .hdc path loaded it fine.
  • hero_to_element re-appends preserved elements with deepcopy and lxml rejects a BuildNode: 15 corpus characters carrying an embedded template failed to write at all.
  • Statedness (which attributes the source stated, in its order) is carried through the JSON encoding. Without it every explicitly-stated empty value (NAME="") was dropped and a dozen defaults per element invented.
  • source_encoding rides in the JSON envelope; two characters came back XML-identical and byte-different without it.
  • ChangeEnvironment.can_add removed. It called a super().can_add that exists nowhere, so any call raised AttributeError, while its docstring advertised logic its body (return True) did not contain. Nothing called it.

Unchanged

  • Oracle parity: 656/656, residual ledger empty. None of this touches the cost path — characteristic_value is untouched and remains what costs derive from. Suite 1798 passing.