Skip to content

chore(release): bump version to 6.3.0 - #600

Merged
dwolfson merged 23 commits into
odpi:mainfrom
dwolfson:prep/6.3.0
Nov 16, 2026
Merged

dwolfson merged 23 commits into
odpi:mainfrom
dwolfson:prep/6.3.0

Conversation

@dwolfson

Copy link
Copy Markdown
Member

pyegeria 6.3.0, to match Egeria 6.3 (tag V6.3, a81af02c).

What's in it (since 6.2.42)

  • The post-6.2.42 release/next work, including:
    • Limit Result by Status as element statuses (Rocky's fix);
    • defaultValueOverride sent only when the type declares it (PY-4);
    • Project Scope mapping (PY-5);
    • find-method/type validation (PY-15);
    • "at least N" counts (PY-17);
    • the drafts listing fix (PY-18);
    • spec-cache guards (PY-11).
  • #9454 release floor = 6.3.0: on a released 6.3 platform, get_relationships and get_related_elements are trusted to report their end (#9453 and #9454 are both in V6.3).
  • Status values (PY-21, part): DEPLOYMENT_STATUS uses REJECTED, matching Egeria.
  • Deprecation wording: the legacy ReportType:: loader is "removed in pyegeria 6.4 or the next release".
  • Release notes: docs/release-notes/next.md.
  • Version 6.2.42 -> 6.3.0. uv lock also takes click 8.5.1, which fixes main's stale lock.

Checks

  • Micro-tests on Python 3.13, run with EGERIA_USER unset and with EGERIA_USER=someoneelse: 4478 passed, 0 failed, 5 skipped either way. v6.2.42 has the same 5 skips; 3 old defaultValueOverride tests were replaced on purpose by PY-4.
  • uv lock --check passes. An import smoke run passes both ways. The wheel layout is unchanged (my_egeria top-level, serve_my_profile entry point).
  • Against released Egeria 6.3 (read-only):
    • /api/about reports "6.3", and fixes 9453/9454 are recognised.
    • Paged get_relationships and get_related_elements end trusted_null at the DB count.
    • get_elements(DataAsset) ends trusted_null at 1756 of 1798. All 42 missing elements are hidden from the caller (11 by zone, 31 anchored to a Promise).
    • By-exact-property-value reads equal the DB (38, 54).
    • Known issue (in the notes): type-specific searches such as find_data_assets can be incomplete on 6.3 and report "at least N".

🤖 Generated with Claude Code

…t is never a private draft

The Architect's ruling (a), 2026-11-14 late (D16):
- async_save, async_load and async_delete refuse an element zoned to the author and others, with "a private
  draft is zoned only to you; publish it to share it";
- async_list leaves it out, with a note in notes;
- prune reports it under not_private.
Before this, any element whose zones merely included the author's zone was treated as their private draft.

Tests: exact zone saved over; superset refused on save, load and delete; superset left out of the list with
the note; prune's sheet test updated.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…ationships and get_related_elements

G2 (the Architect's brief and ruling, 2026-11-15, option (B) plus (A)'s client fix):
- FIXES table in pyegeria/core/paging.py, with a snapshot cutoff, release floor and later-line rule per fix:
  - 9453: cutoff 2026-11-14T19:52:00Z, floor 6.3.0, later lines yes (unchanged);
  - 9454: cutoff 2026-11-15T07:26:00Z, floor None, later lines yes (ruling 4).
  METHOD_NEEDS: find / by-name / by-type need {9453}; get_relationships, _async_get_relationships,
  get_related_elements and _async_get_related_elements need {9454}.
  platform_fixes / async_platform_fixes: one /api/about read and cache per URL, fail-safe (doubt is no fixes),
  PYEGERIA_TRUSTED_END=off turns all fixes off, no force-on. platform_promoted and gate_from_about stay as
  9453 aliases.
- Without 9454: get_relationships stays NULL_NOT_END; _async_get_relationships, the same server call, now joins it
  (ruling 3). get_related_elements stays plain suspect (ruling 1). The old :114 assert is replaced by import-time
  invariants for both.
- saved_query._all_related and the dashboard_sheet related read pass their client, so they promote on a 9454
  platform (ruling 2).

G3: odpi/egeria has no 6.3.0 tag yet (git ls-remote --tags, 2026-11-15), so the 9454 release floor stays None and
no release is promoted for it. 970092b9 committed 2026-11-15T07:25:59Z.
Evidence: ~/.re-gate/CAPTURE-G1-9454-2026-11-15/ and READBACK-G1-9454-CAPTURE-AND-G2-BRIEF-2026-11-15.md.

G4: tests/micro-tests/test_paging_9454_gate.py (new), plus updates to the tests that encoded the one-fix gate.
Full micro run, offline: 4380 tests, 0 failed, 5 skipped, both with EGERIA_USER unset and with EGERIA_USER=someoneelse.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…the #9454 gate

overview_metrics._all_relationships called paging_for("get_relationships") with no client, so its scans never
promoted, even on a #9454 platform (laz a5c3a195), and the Portal's unconfirmed markers there never cleared. It now
passes `ce`, as the find scan (:384) already passes `mgr`.

Tests (test_overview_unconfirmed.py):
- a 9454 client: trusted, confirmed, no unconfirmed marker;
- a 9453-only client: NULL_NOT_END (6 probes), marked unconfirmed.
- test_confirmed_figures_are_unchanged's paging_for seam now takes the client argument.

Full micro run, offline: 4382 tests, 0 failed, 5 skipped, both with EGERIA_USER unset and with EGERIA_USER=someoneelse.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
Egeria main c61bad65 added scope to ProjectProperties (odpi/egeria #9448). Dr.Egeria's
Project Scope was recorded as not applied; the owner's 2026-11-15 decision replaces
the 2026-11-12 one.

- Create and Update Project send scope (merge update drops None, so an omitted
  Project Scope leaves the stored scope alone); the not-applied note is gone.
- The Project Scope attribute's variable_name is now scope, so report specs read
  properties.scope (the old project_scope key read a property Egeria never returns).
- Regenerated Project templates, report specs and seed; generated_format_sets.json
  edited only for the Project Scope key.
- Tests: create, update, merge-omitted, read.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit 6c4793c)
…ble, 0 unrunnable)

The previous commit regenerated manifest.json in a venv without rdflib, which wrote
376 runnable / 2 unrunnable and failed three test_rs_egeria_tier_seed tests. Only the
manifest census differed; restoring upstream's file.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit 7ac8760)
…ared tier entries. By-name call sites pinned (PY-11, PY-8)

PY-11 (6.2.42 read-back advisories):
- fetched_refusal now has the same None guard as _fetched_hit: with no identity (a declared multi-user
  process outside a request, or spec_identity(None)) it makes no words and logs nothing. fetched_spec
  resolves the identity once for the hit and the words.
- _TierEntry and egeria_tier say that every reader with the same identity gets the same specs/meta
  objects: treat them as read-only and copy before editing.

PY-8: the unpaged by-name call sites were given dispositions in 6.2.42 (test_by_name_dispositions.py).
A source-scan test now pins which modules may call get_collections_by_name / get_communities_by_name /
_async_get_note_logs_by_name, so a new call site fails until it has a disposition. The two
GovernanceOfficer prototype files pass a GUID to a manager stub and are listed as prototypes, unchanged.

Tests: no identity gives no refusal words and no log key; another identity still gets the words; same
identity shares one entry, another identity does not. Full micro, full-deps venv (rdflib 7.6.0), both env
ways: 4312 tests, 0 failures, 5 skipped.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit 424b7bf)
Egeria main c61bad65 declares defaultValueOverride on SchemaAttribute
(OpenMetadataTypesArchive.java:1095); 757a5df0 did not, and sending it failed the whole create with
OMRS-REPOSITORY-400-028. The withholding is now a type-definition check: send it when the target type or a
super type declares it; otherwise report it as not applied; if the type can't be read, report that
(undetermined is not absent) and still create.

type_declares_property and type_super_chain (md_processing_utils/typed_properties.py) read the type's
definition and its super types through valid_metadata; a failed or missing read is None, never 'not declared'.

Tests: declared / not declared / unreadable for SchemaAttribute, DisplayDataContainer and DisplayDataField,
no lookup when the attribute isn't given, and the helper's super-type walk.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit efd8c1a)
…umn keys are checked strictly (PY-15)

BRIEF-AUTHORING-PHASE2 step 2.
1. FIND_ELEMENT_TYPES records what each find method returns (a test checks every value is a type in the
   Properties catalog). A spec whose target type isn't that type or a subtype (read from Egeria's type
   definitions, or already paired with that method in a shipped spec) is refused (prototype C06: Asset +
   find_projects passed). An unrecorded method, or a type whose super types can't be read, is a warning and
   listed in not_checked, never a pass.
2. A column key must be an attribute of the target type (super types included), a key the output formatter
   derives (computed from the formatter: its header properties, GUID, Mermaid and chart columns), a documented
   related-element or annotation field, or, for a generated -DrE- spec, a Dr.Egeria command attribute. An
   unknown key is refused naming it. The old loose check, which accepted any key some pyegeria spec uses,
   remains only as validate(..., legacy=True), with a warning per such key.

Every shipped find spec validates with no column or fit refusal (a test runs the registry). The allow-lists
name only keys a shipped spec uses (a test). The Projects spec's start_date reads nothing today and is
allow-listed with that reason, to be fixed against live output.

Relationship-path keys are step 1 and are not accepted here.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit e329690)
…d key from the target's own command

- Regenerated md_processing_utils/properties_class_fields.py from Egeria c61bad65 (scripts/gen_properties_class_fields.py,
  read-only on the Egeria checkout). ProjectProperties gains scope (#9448), so a hand-written Project spec with
  a scope column passes strict validation. The same regeneration drops 'queries' from the SchemaType
  properties classes, as Egeria did; nothing in pyegeria reads it from this catalog.
- test_a_generated_dre_spec_may_use_command_attribute_keys_but_a_hand_spec_may_not no longer hard-codes
  project_scope (renamed to scope by the Project Scope mapping): the odd key is taken at test time from
  Create Project's own command attributes. New test: a hand-written Project spec may show scope.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit a527624)
…ad65

Adds a Changed entry to docs/release-notes/next.md: ProjectProperties gains scope; the SchemaType-family classes
drop queries (#9448); Dr.Egeria and authoring validation follow Egeria 6.3's type system.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit 27cc1e3)
….*" property search

EgeriaDraftStore.async_list searched classification-explorer by-property-value for ".*" (starts_with=False).
Egeria matches a property value literally, so on a live server it listed no drafts, and prune_drafts, built on it,
removed none. Found by the D16 run on laz, 2026-11-15: W1 was saved zoned exactly [erinoverview], and R1 listed 0
drafts (~/.re-gate/READBACK-D16-SECOND-PERSONA-PROOF-2026-11-15T1948Z.md; the Architect's ruling
RULING-D16-VOID-PY18-ASYNC-LIST-2026-11-15.md).

- async_list now reads each kind by type through classification_manager._async_get_elements, with
  governance_zone_filter=[zone], paged with async_paging_for("_async_get_elements"). That's the by-type
  findMetadataElements read: NULL_NOT_END before #9453, gate-trusted after it. LIST_PAGE_SIZE = 50.
- The exact-zone rule (a superset-zoned element is noted, never listed) and the "at least" notes are unchanged.

Tests: test_drafts_list_py18.py (new):
- a literal-matching fake gives 0 drafts for the old ".*" call;
- the by-type read carries the zone filter;
- exact vs superset zone;
- confirmed on a #9453 platform, "at least" before it;
- prune sees the drafts;
- an AST guard: no regex literal reaches a *by_property_value call in pyegeria, md_processing, commands or
  my_egeria.
The fakes in test_authoring_api, test_paging_step4_loops and test_authoring_d14_published_versions now serve the
by-type read. test_authoring_api's property search now matches literally, like Egeria, and its by-type read ends
with null, not [].

Full micro run, offline, with rn-step5's full-deps python and PYTHONPATH set to this worktree: 4417 tests, 0
failed, 5 skipped, both with EGERIA_USER unset and with EGERIA_USER=someoneelse.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…ort runs page to their max rows

Also covers PY-16's MCP part. The rest of PY-16 was resolved by the pager steps: preview already pages with
page_to=max_rows, and EA answer() was fixed in EA (2c6e4667).

- result_count(result) (pyegeria.view.format_set_executor) returns {"count", "count_text", "complete", "end",
  "reason"} for a find run's result (kind table / json list / empty, with end). count_text is "N" only on a
  "confirmed" end; any other end is "at least N", and reason is the result's own notes or the words for the end
  (END_REASONS_WORDS). It returns None for a result that isn't a counted find (an analytic, a chart, text, or no
  end). One helper the Portal and EA can reuse.
- The MCP report runs pass page_to: the run_report tool (_async_run_report_tool), mcp_adapter.run_report and
  _execute_egeria_call_blocking. page_to is page_size, else MCP_MAX_ROWS (100), so a list format pages like preview
  does, instead of reading one page and ending "first_page". Each result gains "count" and is otherwise unchanged.

Tests: tests/micro-tests/test_py17_count_words.py (new):
- confirmed vs every other end;
- notes as the reason;
- empty;
- no count for non-finds;
- page_to and count on all three MCP entry points.
Full micro run, offline, with rn-step5's full-deps python and PYTHONPATH set to this worktree: 4429 tests, 0
failed, 5 skipped, both with EGERIA_USER unset and with EGERIA_USER=someoneelse.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
(cherry picked from commit 1f5fcf0)
… add lifecycle filters

Builds on bf45d29 (Egeria's ElementStatus check in
pyegeria.models.element_status). Rebased from a2bd7aa onto release/next.

- Limit Result by Status (Create Report, View Report) is a Simple List of
  the element statuses (UNKNOWN, ACTIVE, INVALID, DELETED, OTHER), default
  ACTIVE, validated locally. A lifecycle value such as DRAFT is dropped
  with a warning that points at the new filters, rather than reaching
  Egeria. Old forms (['ACTIVE'], JSON lists) still parse.
- New attributes Limit Result by Content Status, Limit Result by
  Deployment Status and Limit Result by Activity Status, with no default,
  valued from Egeria's ContentStatus / DeploymentStatus / ActivityStatus.
  They are sent only by find methods that accept content_status_list /
  deployment_status_list / activity_status_list
  (format_set_executor.LIFECYCLE_FILTERS); elsewhere they are dropped with
  an UnknownFilterWarning.
- The parsed-attributes table shows "Warning" rather than "Valid" for an
  attribute whose value was partly dropped.
- refresh_specs for Views & Reports: templates, seed and report specs.
  Only Report-DrE-Advanced changes in generated_format_sets.json and
  base_report_formats.py; the regeneration's null render/max_items/
  max_lines/label fields on every column were left out.

Checked against the local Egeria (2026-11-16): limitResultsByStatus
accepts all five element statuses and refuses DRAFT (400);
get_valid_metadata_values gives deploymentStatus REJECTED (not
REJECTED_FOR_DEPLOYMENT) and activityStatus FOR_INFO, as used here.

Tests: tests/micro-tests/test_limit_result_by_status_semantics.py (34);
micro-tests 4471 passed, 5 skipped on Python 3.13, with EGERIA_USER
unset and with EGERIA_USER=someoneelse.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
Claude-Session: https://claude.ai/code/session_015pdMuSi43tjoHBv3ssZfu5
…ent search

dr_egeria_manual.md
- New "Techniques and Best Practices" section, each rule checked against
  md_processing/v2/extraction.py: how a document is split into blocks and
  commands, the rules that follow (close commands with ___, prose that
  starts with a verb, commands in code fences still run, attributes as ###
  headings), re-running documents, and the status filters.
- The ___ separator is described as what ends a command, not an optional
  delimiter. TOC renumbered.

search-and-get-filters.md
- "Element status and lifecycle status": limitResultsByStatus takes
  Egeria's ElementStatus (UNKNOWN, ACTIVE, INVALID, DELETED, OTHER;
  elements hold ACTIVE or DELETED), lifecycle values are refused before
  sending and by Egeria (HTTP 400), and which methods take
  content/deployment/activity status filters.
- "Searching efficiently": list by type once in pages of 100
  (STAR_PAGE_MAX) instead of one search per item, with timings.
- The example that implied ACTIVE meant lifecycle is corrected.

The status-filter text describes the preceding commit (Limit Result by
Status is a list of element statuses); merge with or after it.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
Claude-Session: https://claude.ai/code/session_015pdMuSi43tjoHBv3ssZfu5
The #9454 fix (970092b9: get_relationships / get_related_elements send []
for a filtered page and null only at the end) had release_floor None, to be
set only once the 6.3.0 tag was shown to carry it. Egeria 6.3 was released
2026-11-16: the V6.3 tag is a81af02c, and GitHub's compare 970092b9...V6.3
is "ahead", so the tag carries the fix. 2a27dad5 (#9453) is in it too, which
confirms the existing #9453 floor of 6.3.0.

A released 6.3 (or later) platform now trusts the relationship reads as well
as the find paths; releases before 6.3 trust neither.

Tests: a release from 6.3 carries both fixes; one before 6.3 carries
neither; with the 9454 floor removed, a release carries only 9453.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…ity lists offer FOR_INFO (PY-21)

Checked against Egeria's valid metadata values on laz (2026-11-16):
- deploymentStatus has REJECTED; there is no REJECTED_FOR_DEPLOYMENT.
  _globals.DEPLOYMENT_STATUS, the fallback used when the values can't be
  fetched, offered the wrong name.
- activityStatus includes FOR_INFO. The seed report specs' activity_status
  columns (Actor Manager 4, Feedback 8, Projects 4) left it out, so a
  FOR_INFO activity could not be chosen there. The compact specs and
  generated formats already had it.

Found by Rocky while rebasing the Limit Result by Status fix. A new test
pins both fallback lists and every seed activity_status column to
Egeria's values.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
next.md wasn't reset when 6.2.42 was cut, so it still listed as unreleased
items that shipped in 6.2.42 (NO_PASSWORD, AuthoringRefused.answer, can_edit,
used_by, the per-identity spec cache, the unique-name read fix, get_elements
paging, the load_egeria_report_specs deprecation). Each was checked against
the 6.2.42 tree (9b3eb8e) and removed, with a pointer to the 6.2.42 notes.

Added what the post-6.2.42 commits bring and the notes lacked: Limit Result by
Status as a list of element statuses with three lifecycle filters,
defaultValueOverride (PY-4), Project Scope (PY-5), strict authoring validation
(PY-15) and the status-value fix (PY-21). The #9454 entry now says releases
from 6.3.0 are trusted (the V6.3 tag carries 970092b9).

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
ca93542 edited three seed report-spec files by hand to add FOR_INFO to
their activity_status columns. The seed is generated from the
constant-defined report formats (auto-generated from the Dr.Egeria
command specs), so test_rs_egeria_tier_seed failed: the seed no longer
equalled its source. This reverts the seed files to their generated form
and drops the seed assertion from test_status_value_lists.py.

Kept: the deployment-status fallback fix (REJECTED, not
REJECTED_FOR_DEPLOYMENT) and its tests. FOR_INFO for the generated
activity_status columns is a follow-up through the Dr.Egeria command
specs and a regeneration. next.md says only what this release fixes.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
… January 2027)"

Dan, 2026-11-16: the next release will probably be 6.4, and then 7, so the
legacy-loader deprecation and the planned removal of the default password
say "the next release" instead of naming pyegeria 7 and a date.

- LEGACY_LOADER_DEPRECATION (the DeprecationWarning and logged WARNING
  text), load_egeria_report_specs' docstring and the load-report-specs
  CLI's DEPRECATION_WORDS now say "will be removed in the next release".
- next.md's "Planned for pyegeria 7" section is "Planned for the next
  release".
- Comments in _no_password.py, config.py, mcp_server.py, used_by.py,
  load_report_specs.py and the report-spec migration design doc follow.
- The two tests that pinned the old words now pin the new ones.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…ext release"

Dan's wording (2026-11-16): a shipped package's deprecation text should name the
release, so every place f6e3851 changed now says "pyegeria 6.4 or the next
release": the legacy loader's DeprecationWarning and CLI words, the legacy
ReportType:: prefix note, the default-password lines, the design doc and
next.md's "Planned for" heading.

- The two tests that pin the words now pin the new ones.
- next.md line 3 (about the release/next branch in general) is unchanged.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…t least N")

The Architect's release ruling for 6.3.0 asks for this line. Checks on
released Egeria 6.3 (2026-11-16) confirmed that #9453/#9454 make
get_relationships and get_related_elements trusted. The find methods are not:
an unfiltered find_data_assets("*") read 1357 of 1797 assets, with its end not
confirmed. pyegeria already reports such a count as "at least N".

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
…e out

The 0521acd line said find methods are not trusted on 6.3. On a 6.3 release
#9453 does make the find_metadata_elements family trusted (by type, by name,
by property value). Checked on released 6.3 (2026-11-16, C2b): get_elements
for DataAsset ended trusted_null at 1756 of 1798. All 42 missing were shown,
one by one, to be invisible to the caller: 11 in the "security" zone and 31
anchored to a Promise. The exact property-value reads matched the DB (38, 54).
Only type-specific searches such as find_data_assets stay "at least N".

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
pyegeria 6.3.0, to match Egeria 6.3 (tag V6.3, a81af02c).

- pyproject version 6.2.42 -> 6.3.0. uv lock (uv 0.13.8, the same lockfile
  revision as main) also picks up click 8.5.1, which fixes main's stale lock.
- Release notes: elements anchored to a Promise are left out unless the
  request asks for lineage (pyegeria.lineage_visible()), not "on every path".

Checked on released Egeria 6.3 (Rocky, 2026-11-16): /api/about says "6.3";
#9453 and #9454 are trusted; paged relationship and by-type/by-property reads
end trusted_null at the DB count, after removing what the caller may not see.

Signed-off-by: Dan Wolfson <dan.wolfson@pdr-associates.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant