Release 1.2.3a1#74
Open
github-actions[bot] wants to merge 128 commits into
Open
Conversation
Adds the remaining two intent primitives.
- resources.py — `LocaleResources`, the OVOS-INTENT-2 loader: the §3
common reader (UTF-8/BOM/CRLF, comments, blanks), recursive
`locale/<lang>/` search with case-insensitive tags, the user→skill→core
override precedence (§2.1), and the five resource roles (§4). The
user-data path is a parameter — no configuration import. Duplicate
resources and empty files raise `MalformedResource`.
- dialog.py — `render()`, the OVOS-INTENT-2 §4.2 dialog renderer: select
a phrase, expand its variety, fill `{name}` slots from caller values.
Expansion precedes fill so a slot value is never parsed as grammar.
An unfilled slot raises `UnfilledSlot`.
20 more conformance tests (58 total).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The package is the reference implementation of the OVOS formal specifications generally, not only the intent specs — the formal-specifications repo will grow beyond OVOS-INTENT-1/2/3. Renamed from ovos-intent-primitives: repo, distribution, and the import package (ovos_intent_primitives -> ovos_spec_tools). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A CLI and library function that validates every resource file under a locale directory and reports every problem found. - syntax (OVOS-INTENT-1): each template expanded; malformed forms, empty files, and named slots in slot-free roles are errors - naming/layout (OVOS-INTENT-2): base-name charset, .entity slot-name rule, duplicate (role, base name), language-tag form, files outside a language directory, legacy file types - `ovos-spec-lint <locale>` — accepts a locale/ or a <lang>/ directory; exit code non-zero on errors, or on warnings with --strict (CI-ready) 15 linter tests (73 total). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
examples/dirty-locale is a deliberately broken skill locale — every file trips at least one ovos-spec-lint check (two valid files produce no findings). examples/README.md shows the locale, the command, the verbatim linter output (10 errors, 4 warnings), and a table mapping each finding to its spec rule. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
An OO alternative to render(). Beyond holding the phrases, vocabularies and rng, it tracks the last phrase chosen and avoids repeating it on the next call — repetition avoidance, which a stateless function cannot do and which OVOS-INTENT-2 §4.2 explicitly allows. - DialogRenderer(phrases, vocabularies=, rng=) with .render(slots=) - DialogRenderer.from_resources(resources, name) builds one from a LocaleResources - render() refactored to share _render_phrase with the class 6 more tests (79 total). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…a Protocol - rng parameters are now typed `Chooser`, a Protocol (an object with a `choice` method) instead of the meaningless `object`. `random` and a `random.Random` instance both satisfy it. - DialogRenderer holds default slot values (set once, reused every render) and `.entity` value sets. A slot resolves in order: per-call value, default, a random `.entity` value, then UnfilledSlot. - DialogRenderer.from_resources now also loads `.entity` value sets and accepts default `slots`. - LocaleResources.entities() — every `.entity` for the language, as a name -> expanded value set map (mirrors vocabularies()). 6 more tests (85 total). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
One self-contained .py per feature worth demonstrating: - expand.py — the sentence template expander - load_resources.py — the locale resource loader - render_dialog.py — render() and DialogRenderer (repetition avoidance, default slots, .entity fallback) - lint.py — the linter used as a library Adds examples/skill-locale/, a small valid skill locale the loader and dialog scripts read. examples/README.md indexes the scripts. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When the requested language has no directory, LocaleResources resolves to the nearest available language (OVOS-INTENT-2 §2.2, non-normative) — e.g. en-AU finds en-US. - nearness is decided by a LanguageMatcher Protocol (an object with tag_distance(desired, supported)); the langcodes module satisfies it structurally and is the default - langcodes moved from a hard dependency to the [langcodes] extra; the package core now has zero dependencies. Without langcodes, resolution is exact-match only - max_language_distance caps the fallback (default 10, per §2.2); set 0 to disable 5 more tests (90 total). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Language is dynamic — a locale folder is a skill's multilingual unit — so it must not be pinned at construction. - LocaleResources: drops the `lang` constructor argument; every load method takes `lang`. One instance serves every language, and the smart fallback is resolved afresh per call. - DialogRenderer: now resource-backed and multilingual — built from a LocaleResources + a dialog name, with `render(lang, slots=)`. Repetition avoidance is tracked per language. The `render()` function stays as the stateless, language-agnostic primitive. - New ovos_spec_tools/language.py — `standardize_lang` and `closest_lang`, the single home for the language-tag logic OVOS reimplements across locale loading, TTS voices and STT models (ovos_utils.lang.get_language_dir, phoonnx.match_lang, ...). Mirrors their proven behaviour: tag standardization, distance-below-10 match, the Tagalog `tl`/`fil` quirk. langcodes stays optional. - LocaleResources uses closest_lang; a custom `lang_resolver` may be injected. The LanguageMatcher protocol is replaced by this. 9 more tests (99 total). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…s one langcodes resolves a bare tag to its most-populous region, so "pt" matches "pt-BR" closer than "pt-PT". The unmarked form of a language should resolve to its reference variety: Portuguese is "from Portugal", and every Lusophone country except Brazil follows the pt-PT norm. closest_lang() now, for a bare language tag with a norm region (_NORM_REGION, seeded with pt -> PT), prefers a candidate in that region over the langcodes distance result. So a request for "pt" against a locale offering pt-PT and pt-BR resolves to pt-PT. An explicit pt-BR request is still respected; a bare tag still falls back by distance when the norm region is absent. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…bsent Without langcodes no tag distance can be computed, so closest_lang previously resolved exact matches only. It now adds a final fallback: a candidate sharing the primary subtag — so a request for en-AU still accepts en, en-GB, en-US, ... It prefers the bare language tag, then the norm region, then the first match. This fallback also covers the rare case where langcodes is installed but computes no in-threshold distance. 3 more tests (105 total). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous closest_lang was a cascade of special cases — exact, then norm region, then langcodes distance, then a primary-subtag fallback. That logic now lives in a single distance function. - lang_distance(a, b) — the one place policy lives: standardizes tags, measures a bare tag from its norm region (so pt -> pt-PT is 0, not a langcodes population guess), uses langcodes when present, and falls back to a coarse same-language measure (shared primary subtag is near, the generic form nearer than a sibling region) when it is not. - closest_lang(target, available, max_distance) — now branch-free: the candidate with the smallest lang_distance, accepted if below the cap. - lang_distance is exported as a public primitive. 108 tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New docs/ — an ordered guide that builds from a first install to the full API: - getting-started — install and a taste of every tool - templates — the OVOS-INTENT-1 grammar - locale-resources — the locale/ folder and the five file roles - dialog — render() and DialogRenderer - language-matching — standardize_lang, lang_distance, closest_lang - linting — ovos-spec-lint, on the CLI and in CI - api-reference — every public name The README drops its long per-tool sections in favour of a quick taste and links into docs/. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
34 new tests (142 total): - expansion: whitespace/tab normalization, optional vocabularies arg, bare <name> templates, multi-word vocabulary members, empty vocabulary, no cross-template slot-consistency check, optional around a slot, unicode literals, duplicate-branch de-duplication - resources: core fallback and skill-overrides-core precedence, empty .dialog/.voc, vocabularies()/entities() directly, indented and mid-line "#", nonexistent skill_locale, undefined <voc> reference - dialog: single-phrase renderer, no-slot phrases, a slot value containing braces stays literal, numeric values, missing dialog - language: empty available list, empty-string tag, regional vs different-language distance, regioned-request sibling fallback - lint: nonexistent path, empty locale, unknown extension ignored, a single-language directory argument Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New checks:
- §5.5 slot consistency: every template in one .intent/.dialog must
declare the same slot set — flagged when they differ (previously
unchecked)
- a non-UTF-8 file is now reported as an error instead of crashing the
linter (UnicodeDecodeError is not an OSError)
- a .blacklist with no matching .intent — a warning (orphan suppression)
- a language directory with no resource files — a warning
New --spec-version {0,1,2} flag — a forward-compatibility check that
flags features newer than a target runtime:
- spec-version 0: a .blacklist file warns (a v0 runtime ignores it)
- spec-version < 2: a <name> reference errors (an older runtime cannot
expand the template)
- default 2: nothing extra flagged
11 more linter tests (153 total). The dirty-locale example output is
unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…cifications) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Part of a documentation and interoperability effort for OpenVoiceOS, funded by NLnet under grant agreement 101135429. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
AI-Generated Change: - Model: claude-sonnet-4-6 - Intent: add missing CI/CD workflows - Impact: added build-tests.yml, coverage.yml, lint.yml, license_check.yml, pip_audit.yml, release_workflow.yml, publish_stable.yml, release-preview.yml, repo-health.yml, conventional-label.yml - Verified via: reviewed apply_workflows.py output
Implements the `.prompt` resource role: a localized, whole-file
plain-text prompt for a language model.
- prompt.py — render_prompt() and PromptRenderer. {name} substitution is
conservative: only a well-formed name, only names the caller supplied,
never inside a ``` fenced code block. Unfilled slots and any other
braces are left literal.
- resources.py — read_prompt_file() (whole/verbatim read) and
LocaleResources.load_prompt(); PROMPT_ROLE constant.
- 22 tests (175 total); docs and an examples/render_prompt.py added.
The linter's .prompt awareness (and --spec-version 3) remains a
follow-up.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The linter now recognizes `.prompt` as a resource role instead of silently ignoring it as an unknown extension. - `.prompt` is a known role: collected, deduplicated, and naming-checked like the others; no longer mistaken for a legacy/unknown file. - A `.prompt` is checked as a whole-file document, not a template — only non-emptiness and a valid UTF-8 read; never template syntax, slot-free or §5.5 checks. - A non-UTF-8 `.prompt` is reported, not crashed. - --spec-version gains 3 (the .prompt role); default is now 3. Below it, a `.prompt` file is a warning — an older runtime ignores the role. 6 tests (181 total); docs/linting.md updated. dirty-locale output unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…g policy (#4) * feat: lang_matches and iter_locale_dirs — pin the cross-component lang policy Every place that crossed a component boundary (resource loaders, adapt/padatious/padacioso engine bucketing, TTS/STT plugin routing, `_get_closest_lang` boilerplate) reimplemented the same three steps — standardize, distance, threshold-check — with subtle drift between the copies. Bug surface: a skill stores resources under a full tag but an engine buckets under a macroed one (or vice versa), and the two never reconcile. This adds the two missing primitives so callers stop reinventing them: * `lang_matches(a, b, max_distance=10) -> bool` — the `if lang_distance(...) < threshold` check, named and shared. Pass `max_distance=0` for exact-match. * `iter_locale_dirs(root, native_langs=None) -> Iterator[(lang, path)]` — walks `<root>/locale/` and yields each subdir as a canonical `(standardize_lang(name), Path)` pair, optionally filtered against the skill's native langs via `closest_lang`. Resource loaders for `.rx`, `.dialog`, `.voc`, `.intent`, locale `.json` reinvent this walk by hand — switch them to this and the macro/full-tag disagreement disappears. The implicit policy: **full normalized lang tag everywhere**. Macro-stripping is no longer a knob; if an engine buckets under `en` and a resource is `en-US`, the engine reconciles at match time via `closest_lang`, not at registration time via macro flags. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix: remove unused MalformedTemplate import, rename ambiguous loop var, sync docstring - Drop MalformedTemplate from resources.py import (F401 — unused) - Rename loop variable l -> lang in iter_locale_dirs comprehension (E741) - Add lang_matches and iter_locale_dirs to the top-level module docstring Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * ci: pass install_extras as pip args and expose license classifier - install_extras was 'test' (bare) → reusable workflow ran 'pip install test', no such PyPI package. Pass '.[test]' instead. - pip-licenses categorised the package itself as 'Error' (unknown license) because pyproject had no License classifier. Add the SPDX Apache-2.0 classifier and matching Python-version classifiers. - Tighten requires-python to >=3.10 to match the tested matrix. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * ci: exclude ovos-spec-tools from auditing itself pilosus/action-pip-license-checker queries PyPI metadata for each installed package. The PyPI release predates the SPDX License classifier added in this PR, so the package is audited as 'Error' (unknown license) against itself. Exclude it from the audit until the next alpha republishes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * ci: build-tests install_extras takes the bare extras name The build-tests reusable workflow wraps it as `${WHEEL}[${EXTRAS}]`, so it expects `test` not `.[test]`. The coverage workflow expects raw pip args, so that one keeps `.[test]`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * ci: warn_only license_check until alpha republishes The pilosus 'exclude' regex did not suppress self-audit; switch to warn_only so the job no longer blocks. Transitive dependencies still get audited and reported in the PR comment. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ng) (#6) Every callsite outside this module that needed to locate a resource file ended up reinventing the override-precedence + closest-lang walk the loader already encapsulates. Promote the previously private _locate method to a public find — same body, fuller docstring, covered by six new tests. The internal load_intent / load_dialog / load_prompt callers move to the new name. The duplicate-resource check is preserved, so a LocaleResources.find caller that violates the §2 uniqueness rule gets the same MalformedResource it would have got via load_*. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…_samples (#8) * feat: keyword_form / vocabulary_keywords / utterance_contains / strip_samples Every consumer that registered slot-free keywords or matched a vocab against an utterance reimplemented the same handful of operations. Promote them to ovos-spec-tools so the OVOS-INTENT-2 §4.3 keyword convention has exactly one implementation. New free functions in ovos_spec_tools.resources: - keyword_form(template_line, vocabularies=None) -> (entity, aliases) expands one slot-free template, lowercases, dedupes and sorts; the first item is the canonical entity, the rest are aliases that canonicalize to it. Malformed input yields ('', []) rather than poisoning a batch. - normalize_for_match(text, ensure_ascii=True) -> str lowercases, strips, and optionally folds accents and ASCII punctuation; the comparison normalization shared by the two below. - utterance_contains(utterance, samples, exact=False, ensure_ascii=True) true iff utterance matches any sample — whole-word substring by default, exact when requested, accent/punct-folded by default. - strip_samples(utterance, samples) -> str remove every whole-word occurrence of any sample, longest first so composite phrases consume their parts before fallback matches. New methods on LocaleResources: - vocabulary_keywords(lang) -> Iterator[(voc_name, entity, aliases)] - entity_keywords(lang) -> Iterator[(entity_name, entity, aliases)] yield one triple per template line, with the entity/alias split applied via keyword_form. Suit any consumer that registers keyword sets with a primary/alias distinction. Tests: 52 passed (16 new). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: granular normalization flags + permissive whole-word match - normalize_for_match now takes two keyword-only flags: strip_diacritics (NFD + drop combining marks) and strip_punct (drop ASCII punctuation, keep slot braces). The previous ensure_ascii knob conflated the two — French ou/où or technical names like c++ need separate control. Both default True; either flag can be flipped independently. - utterance_contains and strip_samples forward both flags and the whole-word match anchor moves from \b...\b to (?<!\w)...(?!\w), which preserves the \b semantics on word-char boundaries while also matching samples that begin or end in non-word characters (c++, yes!). Without this, a sample ending in punctuation could never match in non-stripped mode. Coverage on the new helpers expanded from 16 to 41 tests covering: each flag independently and in combination, exact-mode normalization, slot marker preservation, regex metachar escaping, unicode samples, malformed lines, blank-sample filtering, override precedence walking, voc-reference resolution at expansion time, and the missing-language branch in _keywords_for. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
) The INTENT-3 §4.2 well-formedness MUSTs were enforced only in lint_locale, not in the IntentBuilder/Intent data model, so an invalid intent could be built and registered on the bus. Add Intent.validate() enforcing both §4.2 MUSTs and call it from the construction path (IntentBuilder.build) and the emission path (Intent.to_keyword_payload): - (a) a keyword intent MUST declare at least one required or one-of constraint — an optional/excluded-only intent has nothing that must be present and is malformed; - (b) a vocabulary MUST appear under at most one role — the same vocab under two roles (e.g. required and excluded) is contradictory and malformed. Raises the new MalformedIntent. Raw Intent(...) construction stays permissive (the scaffold / wire round-trip path via open_intent_envelope), so validation fires only at build/emit and a subclass overriding build() is unaffected. INTENT-1 §5.5 (DialogRenderer rejecting non-identical slot sets) is already enforced on dev via verify_slot_consistency in both render paths. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…3 §4.2) (#59) IntentBuilder.build() and Intent.to_keyword_payload() validated the intent and RAISED MalformedIntent. Because ovos-workshop re-exports this IntentBuilder, build()-raises broke skills that build a constraint-less intent — a backward-compat regression on consumers' dev (e.g. workshop's test_build_preserves_name). Build/emit now LOG a warning instead; the raising enforcement stays available via the explicit Intent.validate() method and the locale linter, so §4.2 is still enforced where rejection is wanted, without breaking construction. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Both classes defined __eq__ but no __hash__, so Python set __hash__ to
None and they were unhashable -- forcing downstream lru_cache callers
(padacioso/linha-fina/nebulento) to pass frozenset workarounds instead
of the objects themselves.
Add __hash__ to both, derived from the SAME fields each __eq__ compares
so equal objects always hash equal (the hash/eq contract):
- Session.__hash__ from to_dict()
- Message.__hash__ from (msg_type, data, context)
A shared _freeze() helper (in message.py, imported by session.py)
recursively converts nested dict/list/set payloads into a deterministic
hashable form. It preserves value-equality: {"x": 1} and {"x": 1.0}
freeze equal (unlike a json.dumps digest, which would diverge "1" vs
"1.0" and violate the contract). The hash is a point-in-time snapshot of
these mutable objects -- safe for lru_cache keys and short-lived dict/set
membership, documented in the method docstrings.
Not a spec change: no message/session semantics or spec docs touched.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
PIPELINE-1 §8 makes the handler-lifecycle trio orchestrator-owned: the
orchestrator (ovos-core) emits ovos.intent.handler.{start,complete,error}
authoritatively, while the skill framework keeps emitting the legacy
mycroft.skill.handler.* as a private done-signal. Bridging the two would
(a) double-emit the spec trio (core's emission + the mirrored legacy) and
(b) reshape a shape-changing event with a lossy best-effort transform.
Remove the 3 trio entries from MIGRATION_MAP, their MIGRATION_PAYLOAD_TRANSFORMS
pair, and the now-unused _handler_legacy_to_spec/_handler_spec_to_legacy
helpers. The INTENT_HANDLER_* SpecMessage members stay (still the spec topics
core emits); only the legacy<->spec bridge for them is removed. Tests assert
the trio no longer migrates.
Make SpecMessage the spec-complete and spec-only vocabulary of static
ovos.* bus topics. Adds the topics defined by specs not previously
catalogued and traces every member to its owning spec clause.
Added (34 static members):
- SESSION-2 §2.7: SESSION_SYNC (ovos.session.sync)
- CONVERSE-1 §6.1: CONVERSE_ACTIVE_LIST(+_RESPONSE)
- PERSONA-1 §11: PERSONA_{QUERY,ANSWER,LIST,LIST_RESPONSE,REGISTER,
DEREGISTER,ACTIVATED,DISMISSED}
- FALLBACK-1 §3: FALLBACK_{REGISTER,DEREGISTER}
- COMMON-QUERY-1 §6: COMMON_QUERY_{PING,PONG}
- TRANSFORM-1 §6: TRANSFORMER_{AUDIO,UTTERANCE,METADATA,INTENT,DIALOG,
TTS}_LIST(+_RESPONSE) — six static query/response pairs
- OCP-1 §4: COMMON_PLAY_{PLAY,SEARCH,PAUSE,RESUME,STOP,NEXT,PREVIOUS,
SEEK,PLAYER_STATE,MEDIA_STATE,TRACK_STATE}
Fixed: UTTERANCE_CANCELLED is owned by TRANSFORM-1 §8.2 ("defined
here"), not PIPELINE-1 §6.4 (which only references it).
Deliberately OUT of the enum:
- ovos.session.update_default / ovos.session.start: used by
ovos-bus-client but no spec defines them (SESSION-2 §1 defers
lifecycle topics); legitimately bare strings in the bus client.
- ovos.context.set/.unset/.clear: only a stray TRANSFORM-1 reference
mis-cites CONTEXT-1 §5; CONTEXT-1 §5.3 routes context mutations
through ovos.session.sync, so no ovos.context.* topic is spec-defined.
- runtime-templated topics (MSG-1 §2.1.1): <skill_id>.converse.{ping,
pong}, <skill_id>:{converse,response}, <skill_id>.fallback.{ping,
pong}, <skill_id>:fallback, <skill_id>:common_query, <pipeline_id>...
No members removed (every pre-existing member is spec-defined).
MIGRATION_MAP unchanged: the new topics' legacy counterparts are not
clean payload-compatible 1:1 renames, so they are flagged rather than
mapped (consistent with the #63 handler-trio decision).
Tests: TestSpecCompleteness asserts the enum equals the golden static
spec-topic set exactly (complete + spec-only), that bus-client
internals stay out, and that no templated topic is a member.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…ing (#67) * feat: SessionManager singleton registry + forward/reply session stamping Add a SessionManager (implementation detail, not a spec) that keeps one live Session per id and folds incoming snapshots onto it (SESSION-1 §2/§4 value-passing). Message.forward/reply now re-stamp the derived message with the live session for its id, so a snapshot deep-copied before a handler mutated the session cannot desync the wire. Documented with the why-it-works argument (a stamp is always either a meaningful refresh or a no-op, never a discard) and a canonical get->mutate->forward example. - Session.update_from: spec-compliant fold via serialize/deserialize round-trip (present-empty overrides, null->default), resolved through type(self) so a subclass rebuilds as itself. - SessionManager.session_cls is pluggable so a downstream layer (ovos-bus-client) builds its richer Session subclass; construction via deserialize stays agnostic of subclass __init__ signatures. - Stamping is transparent to pure spec usage: a session-less message or an id the registry never folded is carried verbatim (§5), so MSG-1 forward/reply semantics are preserved (all 513 tests green). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix: idless/absent session normalizes to default on stamp (§4.3) A Message with no session, or a session naming no session_id, IS the default session per OVOS-SESSION-1 §4.1/§4.3 — stamp it. Only a NAMED id the registry never folded is carried verbatim (a relay's remote session we must not overwrite). Update the MSG-1 deep-copy test to a named (verbatim) session. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…ror) (#69) get_default_session now keys off the shared sessions[DEFAULT_SESSION_ID] rather than a per-class default_session mirror. When a subclass (ovos-bus-client) and this base both reach the one registry, a class-attribute mirror would shadow per class and diverge — the shared dict is a single object, so both see the same default. default_session is kept as a re-synced convenience mirror. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…ion) (#71) The default session is a normal session per SESSION-1 §4 — the wire is value-passing, so a message carrying the default id folds onto the live default like any other id. Drop the owner-only guard in SessionManager.get that refused to fold default-id snapshots (legacy baggage that violated the spec). A session naming no id is normalized to the default id and folded. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…mp) (#73) When the caller supplies a session via the context= kwarg of reply/response, that is a deliberate author choice — honour it and skip the live-session refresh, so a hand-constructed session is never overwritten by the registry. A context= that overrides only routing keys still gets the session refreshed. forward has no context kwarg, so it always stamps (unchanged). Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Human review requested!