pubid 2.0.18 → 2.0.20

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 185124a260b71738ef7c333b0f9ceb23f867cd3886cece12810ed6420aff7a46
-  data.tar.gz: '081f7312e5c76b76bdd7fc8c61d4a7e5c70c1de4a70f2491c69227f1fa7b8f70'
+  metadata.gz: 5bf271d5120b52861eb84e3c8cf375c0bdcebf61ecf3c5666af4a888590ccbc7
+  data.tar.gz: 2516fdc850d72aab4de4a589a9187d840d685cc2bc101f1a501e088a43d6287e
 SHA512:
-  metadata.gz: 07d685cf60518b38c66d21e65186aa942df8a279536ddf1907edd4bb5fb0eb2b52eb9de627c44994aee7a520d51aeedf05de9f917992c194cfc829883eef587d
-  data.tar.gz: 9c153f3a31750860224084b31802eb65654af9948b9688b5493ca51cb946a18da5601710d3c74b18c7250d1edba33d855020bead796558c087a772ba16a99174
+  metadata.gz: 7127733e72ab23c53b55b62bc17258bd936055fef0a3840d61ef81d66dc98d193a9e30cedab3d1178e50b51a26021a8b13c91141b1a9c49e4c5b67448b346987
+  data.tar.gz: b19d01682a51d7b7864643b093e3379fe273fa34370c1f151809e0f84d82f965501dd14a6e95d0956ee9f2206d9e9757eb24c89eec166c8fa7395ed2c7bd6bd6

data/lib/pubid/components/typed_stage.rb

@@ -8,12 +8,7 @@ module Pubid
     # URN render: stage code (RFC 5141-bis: stage-XX.XX format handled by urn_generator).
     class TypedStage < Lutaml::Model::Serializable
       include ::Pubid::SubsetMatch
-
-      # `original_abbr` records the input spelling ("Amd" or "AMD"), not the
-      # stage, so a subset match skips it.
-      def self.subset_ignored_attributes
-        %i[original_abbr]
-      end
+      include ::Pubid::EqualityIgnored
 
       attribute :name, :string
       attribute :code, :string
@@ -25,6 +20,12 @@ module Pubid
       attribute :short_abbr, :string
       attribute :long_abbr, :string
 
+      # `original_abbr` records the input spelling ("Amd.", "AMD", "DAmd"),
+      # not the stage, so `==`, `hash` and `===` skip it (issue #656). It is
+      # not serialized: the parse path keeps the input spelling and
+      # `from_hash` rebuilds the canonical one, and the two ids must be ==.
+      equality_ignored :original_abbr
+
       def to_stage
         Stage.new(
           name: name,

data/lib/pubid/identifier.rb

@@ -957,9 +957,10 @@ module Pubid
     # The shared table names the components a printed identifier usually
     # carries — publisher, type, number, part, year. A flavor that keeps its
     # identity somewhere else answers with its own names appended: IEC's
-    # working documents are `technical_committee`/`wd_number`/`wd_stage`, and
-    # CSA's electrical code prints `cec_part`/`no_number` rather than the
-    # `number` it stores. Without this they carried no annotatable token at all.
+    # working documents are `technical_committee`/`wd_stage` (their document
+    # number IS `number`, issue #687), and CSA's electrical code prints
+    # `cec_part`/`no_number` rather than the `number` it stores. Without this
+    # they carried no annotatable token at all.
     #
     # Appending rather than replacing keeps whatever the shared table does
     # find; a token whose value is not in the string is skipped anyway.

data/lib/pubid/iec/CLAUDE.md

@@ -4,7 +4,12 @@ IEC typed stages, `from_hash` equality, and the `number` retype that deleted `Ie
 
 These notes were part of the root `CLAUDE.md`. Read them before you change `lib/pubid/iec/` or `spec/pubid/iec/`. The root file keeps the cross-flavor contract that every flavor obeys.
 
-- **IEC `from_hash(to_hash) == parse` — two `from_hash` defaults that diverged from the parse path**: `==` was false for **every** IEC identifier, while `to_s`, `to_urn` and `to_hash` were all correct — so the relaton index gate (`from_hash(to_hash) == to_hash`) passed and a lookup (parsed reference vs `from_hash`-ed row) silently returned nothing. Two independent causes, one family. **(1) `original_abbr`** — the JCGM defect verbatim, with a **second** injector: `Identifier.published_typed_stage` (the attribute default) **and** `#stage_from_kv` both `dup`ped the registry entry and set `original_abbr = canonical_abbreviation`, where the parse path (`Builder#locate_typed_stage`, the `:type_with_stage` cast) leaves it nil. Both writes are deleted; the `dup` stays (`TYPED_STAGES` is frozen, its elements are not). **The hand-off warned this fix was rendering-sensitive because IEC reads the field at `iec/builder.rb:161-163` (long-vs-short `Amd 1` vs `AMD1`). It is not: that read is UNREACHABLE.** The block is gated on `identifier.class.attributes.key?(:rendering_style)`, and **no IEC class declares a `rendering_style` attribute** — verified by force-loading all 26 IEC identifier classes and inspecting `.attributes`. The whole block (builder.rb 155–200) is dead, so the read is doubly dead: never executed, and reading a field already nil on the parse path. The long spelling is already collapsed on `main` (`…/Amd 1:2017` renders back `…/AMD1:2017`). The block is **kept with an explicit UNREACHABLE comment** rather than deleted, so the next reader does not mistake it for a live consumer again. **The general lesson: before copying the JCGM delete to another flavor, check not just whether the field is read but whether the READ IS REACHABLE — a gated block whose gate is never true is not a consumer.** **(2) `Amendment`/`Corrigendum` redeclared `attribute :type`** with `default: -> { self.class.type[:key] }`. The redeclaration shadowed the derived reader `Iec::Identifier#type` (`typed_stage&.to_type`) on those two classes only; `type` is not serialized, so on `from_hash` the default assigned the **Symbol** `:amd`/`:cor` where the parse path assigns a `Components::Type`, and lutaml does not cast it. Both declarations are deleted so the two classes inherit the derived reader like every other IEC type; the class method `self.type` is untouched and still used. This is the AMCA lesson in another guise — **a default (or converter) that yields a different runtime shape from the parse path is a sign the attribute should not be there at all**. It also repairs a **parse-path** bug, not just the round trip: `Builder#wrap_with_consolidated` builds its supplement directly and never calls `assign_attributes`, so a consolidated amendment member (`IEC 60050-300:2001+AMD1:2005 CSV`) returned the bare Symbol `:amd` from `.type` on `main` even when parsed — pinned by a spec. **Latent, deliberately not changed:** `Builder#locate_typed_stage` (the parse-path lookup) hands out the shared frozen-array element **without** a `dup`, unlike the two from_hash paths; harmless today because nothing mutates a `TypedStage` in place, but it is the same "frozen array, unfrozen elements" hazard. Verified: `to_s`, `to_urn` and `to_hash` **byte-identical** across all 12,303 inputs in `spec/fixtures/iec/identifiers/pass/`, so nothing serialized moves and **no index migration follows**; `==` rises from 0 to **9,914 of 12,302**. **Verified against the published `relaton-data-iec` dataset too, which is the measurement that matters**: over all **32,817 `index-v2.yaml` rows**, the consumer shape `parse(row.to_s) == from_hash(row)` goes from **0 to 32,817** (and `matches?(ignore: [:year])` with it), while `from_hash(h).to_hash == h` holds for all 32,817 **both before and after** — so the published index is already valid and **needs no regeneration**. Over the 32,855 primary docidentifiers in `data/*.yaml`, 32,813 parse (the 42 failures are pre-existing title-bearing forms like `IEC Technology Report Cyber security:2019`, unchanged by this) and for all 32,813 `to_s`, `to_urn` and `root.number` are identical before and after, with **0 empty index keys**. The fixture corpus is the stricter test only because it contains WorkingDocument/PNW/TRF forms that the crawler does not publish. Locked by `spec/pubid/iec/typed_stage_equality_spec.rb` and by deleting the `"Pubid::Iec"` entry from `PENDING_EQUALITY`. **The 2,388 that remain have a THIRD, unrelated cause and are deliberately out of scope**: `WorkingDocument` (`technical_committee`/`wd_number`/`wd_stage`/`wd_language`, 2,318 ids — `to_hash` is `{"_type"=>"pubid:iec:working-document"}` and **`root.number` is nil**, so every WD row keys the empty string), the PNW `wp_stage`/`wp_type` pair (32) and `TestReportForm#cispr_identifier` (38) are **never serialized at all**. Mapping them ADDS keys to `to_hash` and forces a `relaton-data-iec` re-crawl, which is a separate decision — and a low-value one today, since the published index contains **zero** `working-document` and **zero** `test-report-form` rows. (hand-off: iec-typed-stage-original-abbr; follow-up hand-off: iec-unserialized-identity-attributes.)
+- **IEC `from_hash(to_hash) == parse` — two `from_hash` defaults that diverged from the parse path**: `==` was false for **every** IEC identifier, while `to_s`, `to_urn` and `to_hash` were all correct — so the relaton index gate (`from_hash(to_hash) == to_hash`) passed and a lookup (parsed reference vs `from_hash`-ed row) silently returned nothing. Two independent causes, one family. **(1) `original_abbr`** — the JCGM defect verbatim, with a **second** injector: `Identifier.published_typed_stage` (the attribute default) **and** `#stage_from_kv` both `dup`ped the registry entry and set `original_abbr = canonical_abbreviation`, where the parse path (`Builder#locate_typed_stage`, the `:type_with_stage` cast) leaves it nil. Both writes are deleted; the `dup` stays (`TYPED_STAGES` is frozen, its elements are not). **The hand-off warned this fix was rendering-sensitive because IEC reads the field at `iec/builder.rb:161-163` (long-vs-short `Amd 1` vs `AMD1`). It is not: that read is UNREACHABLE.** The block is gated on `identifier.class.attributes.key?(:rendering_style)`, and **no IEC class declares a `rendering_style` attribute** — verified by force-loading all 26 IEC identifier classes and inspecting `.attributes`. The whole block (builder.rb 155–200) is dead, so the read is doubly dead: never executed, and reading a field already nil on the parse path. The long spelling is already collapsed on `main` (`…/Amd 1:2017` renders back `…/AMD1:2017`). The block is **kept with an explicit UNREACHABLE comment** rather than deleted, so the next reader does not mistake it for a live consumer again. **The general lesson: before copying the JCGM delete to another flavor, check not just whether the field is read but whether the READ IS REACHABLE — a gated block whose gate is never true is not a consumer.** **(2) `Amendment`/`Corrigendum` redeclared `attribute :type`** with `default: -> { self.class.type[:key] }`. The redeclaration shadowed the derived reader `Iec::Identifier#type` (`typed_stage&.to_type`) on those two classes only; `type` is not serialized, so on `from_hash` the default assigned the **Symbol** `:amd`/`:cor` where the parse path assigns a `Components::Type`, and lutaml does not cast it. Both declarations are deleted so the two classes inherit the derived reader like every other IEC type; the class method `self.type` is untouched and still used. This is the AMCA lesson in another guise — **a default (or converter) that yields a different runtime shape from the parse path is a sign the attribute should not be there at all**. It also repairs a **parse-path** bug, not just the round trip: `Builder#wrap_with_consolidated` builds its supplement directly and never calls `assign_attributes`, so a consolidated amendment member (`IEC 60050-300:2001+AMD1:2005 CSV`) returned the bare Symbol `:amd` from `.type` on `main` even when parsed — pinned by a spec. **Latent, deliberately not changed:** `Builder#locate_typed_stage` (the parse-path lookup) hands out the shared frozen-array element **without** a `dup`, unlike the two from_hash paths; harmless today because nothing mutates a `TypedStage` in place, but it is the same "frozen array, unfrozen elements" hazard. Verified: `to_s`, `to_urn` and `to_hash` **byte-identical** across all 12,303 inputs in `spec/fixtures/iec/identifiers/pass/`, so nothing serialized moves and **no index migration follows**; `==` rises from 0 to **9,914 of 12,302**. **Verified against the published `relaton-data-iec` dataset too, which is the measurement that matters**: over all **32,817 `index-v2.yaml` rows**, the consumer shape `parse(row.to_s) == from_hash(row)` goes from **0 to 32,817** (and `matches?(ignore: [:year])` with it), while `from_hash(h).to_hash == h` holds for all 32,817 **both before and after** — so the published index is already valid and **needs no regeneration**. Over the 32,855 primary docidentifiers in `data/*.yaml`, 32,813 parse (the 42 failures are pre-existing title-bearing forms like `IEC Technology Report Cyber security:2019`, unchanged by this) and for all 32,813 `to_s`, `to_urn` and `root.number` are identical before and after, with **0 empty index keys**. The fixture corpus is the stricter test only because it contains WorkingDocument/PNW/TRF forms that the crawler does not publish. Locked by `spec/pubid/iec/typed_stage_equality_spec.rb` and by deleting the `"Pubid::Iec"` entry from `PENDING_EQUALITY`. **The 2,388 that remained had a THIRD, unrelated cause — fixed by #687, see the unserialized-identity bullet below**: `WorkingDocument` (`technical_committee`/`wd_number`/`wd_stage`/`wd_language`, 2,318 ids — `to_hash` was `{"_type"=>"pubid:iec:working-document"}` and **`root.number` was nil**, so every WD row keyed the empty string), the PNW `wp_stage`/`wp_type` pair (32) and `TestReportForm#cispr_identifier` (38) were **never serialized at all**. (hand-off: iec-typed-stage-original-abbr; follow-up hand-off: iec-unserialized-identity-attributes — landed as #687.)
+
+- **The unserialized-identity attributes are serialized (issue #687) — `wd_number` moved into the inherited `number`, five plain `key_value` maps and one nested one added, `wp_type` stripped at the construction site.** Three families kept their identity in attributes no `key_value` map covered, so `to_s` was right while `to_hash` was not: 2,388 of the 12,331 pass fixtures were not `==` after `from_hash(to_hash)`, all 2,318 working documents shared one hash and the type-only URN `urn:pubid:iec:wd`, `root.number` keyed the empty string, and the 38 CISPR-bearing TRFs hashed to publisher+year alone. **(1) `wd_number` is the inherited `number`** — the "prefer the inherited attribute over a parallel one" rule: the attribute is deleted and `Builder#cast` routes the parsed `:wd_number` onto `number`, so `root.number`, the index key and `to_hash` all read it; `technical_committee`/`wd_language`/`wd_stage`/`wp_stage`/`wp_type` take plain maps (primitives need no converters). **The two forms of the class key `number` differently, and that asymmetry is kept, not reconciled.** The #687 open-question-1 ruling ("`number` holds `wd_number`, the committee is a separate column — the ISO/IEEE bucket-by-document-number semantics") says to follow the existing code, and the existing code keys the work-programme form by committee: `PNW 1-1597` has carried number `1` (the TC) and part `1597` (the item) since before #360. So the wd form keys by document number (`1/2457/FDIS` → `2457`) and the wp form by committee (`root.number` = `1`); the committee column (`technical_committee`, serialized now) disambiguates.
+  **(2) `wp_type` carried a trailing space** (`"PWI TR 100-36"` captures `"TR "`); the renderer stripped it at print time, so the stored value differed from what rendered. `Builder#cast` strips it now (an all-space value drops to nil) and the renderer's strip is gone with it — `to_s` is byte-identical. **(3) `TestReportForm#cispr_identifier` maps as a nested identifier** with the converter shape of `VapIdentifier#base_to_kv`/`#base_from_kv` — `to_hash` embeds the nested polymorphic hash, `from_hash` dispatches back through the type map. **`#root` of a TRF stays the form** (#687 open-question-2 ruling): a TRF is a form ABOUT the CISPR document, not a wrapper of it, so those 38 ids key the empty string BY DECISION — the serialized `cispr_identifier` is the identity, not a re-rooted number.
+  **The pubid-scheme URN gained six atoms** — `tc-`/`wdlang-`/`wdstage-`/`wpstage-`/`wptype-` and `cispr-` (kind `:id`), after `vap-` and before `lang-`, so landed URNs do not move — and a working document URNs per document (`urn:pubid:iec:wd:2457:tc-1:wdstage-FDIS`). **One measured surface moved beyond `to_hash` and the pubid URN: the official `urn:iec:std:` URN.** The 2,318 wd-form ids emitted NO official URN on main (`to_urn` was nil — the generator refuses a document with no number); with `number` populated they emit `urn:iec:std:iec:2457::::`, the same generator rule the 32 wp-form rows already followed on main (`PNW 1-1597 ED1` → `urn:iec:std:iec:1-1597::stage-10.20:ed-1:`). No id that had an official URN changed (0 of the other 9,988), and no published `relaton-data-iec` row is touched (the index holds zero `working-document`/`test-report-form` rows) — the movement is 2,318 gains in the direction of more identity, the #360 precedent.
+  **Replay-verified against a main baseline, not just the suite.** Over all 12,331 pass-fixture inputs (12,306 parseable; the 25 parse failures pre-existing): `to_s` byte-identical for all 12,306; `to_urn` differs for exactly the 2,318 formerly URN-less rows above; `to_mr_string` differs for exactly the same 2,318 rows, each from the degenerate publisher-only `iec` (which every working document keyed, the shared-key defect verbatim) to `iec.<number>`; `to_hash` ADDS keys — never removes one, never changes a value — for exactly the 2,388 family rows and nothing outside them; `==` after the round trip rises 9,918 → 12,306; all 12,306 `to_pubid_urn` strings parse back to the same hash. Locked by `spec/pubid/iec/root_number_spec.rb` (the six references, the whole-corpus `==`/root-number sweeps, both rulings) and the per-document URN examples in `spec/pubid/iec/pubid_urn_spec.rb`. (hand-off landed: iec-unserialized-identity-attributes.)
 
 - **PNW is a stage of its own (10.20), the stage separator is a space, and the URN is built from components — issue #360 items 1, 2 and 4.** The three are one change, because on `main` the URN was produced by a regex scraped over `identifier.to_s`, so the stage fix and the separator fix both moved it. Reference behaviour was taken from the installed `pubid-iec 1.15.21`, **except for the URN, where 1.x is the wrong oracle** — see below.
   **(1) PNW.** `identifiers/international_standard.rb` listed `PNW` inside the `code: :pwi` stage (harmonized `00.*`), so `IEC PNW 1000-1:2023 ED2` rendered back as `IEC/PWI 1000-1:2023 ED2`. IEC's portal and `pubid-iec-1.15.21/stages.yaml` both give `PNW: ["10.20"]`. PNW now has its own `code: :pnw` TypedStage with `harmonized_stages: ["10.20"]`, and **`10.20` was removed from `:np`** so a lookup by harmonized code is unambiguous — the URN now performs exactly that lookup. `Identifiers::WorkingDocument#stage` built a bare `Components::Stage` from a hand-written `{"PWI"=>"pwi","PNW"=>"pnw"}` map, giving a stage with no harmonized code and a `"pnw"` stage code that existed nowhere else; it resolves through `Pubid::Iec.locate_stage` instead. **The grammar needed no change for the reported form**: `working_programme_with_publisher` required a literal `/` after the publisher, so `IEC PNW 1000-1:2023` already fell through to `type_with_stage`. That rule was **deleted** (it had zero corpus inputs), and `render_working_document` no longer prefixes the publisher: a work-programme registration renders bare (`PNW 65-915 ED1`), because its number is TC-scoped (`86B-15`, `SyCCOMM-1`) rather than an IEC document number, and prefixing `IEC ` would break idempotence — `IEC PNW 65-915 ED1` re-parses as an International Standard. That alone repaired 32 fixture entries.
@@ -16,7 +21,7 @@ These notes were part of the root `CLAUDE.md`. Read them before you change `lib/
 
 - **`edition` serializes flat, and the date does not need to** (pubid#360 item 3): IEC's edition emitted a one-key hash, `edition: {number: "2"}`, for a value that is only ever a scalar. `lib/pubid/iec/builder.rb` has a **single** edition construction site and it writes `Edition.new(number: value&.to_s)` and nothing else, so the component is degenerate by construction — confirmed over all **12,331** parseable IEC fixtures, where 607 carry an edition and `number` is the only field ever populated (`year` and `original_text`: zero). It now emits `edition: "2"`, and `from_hash` reads both that and the stored nested form, so no published `relaton-data-iec` row stops deserializing. The mapping at `lib/pubid/iec/identifier.rb` stays a plain `map "edition", to: :edition`; the flattening is the shared rule on `::Pubid::Identifier`, not an IEC converter. **The date half needed no work at all** — IEC already serializes it flat as top-level `year`/`month`/`day`/`undated` through its own converters, so no `date` key ever reaches the shared flattener. **That is not a coincidence worth ignoring: IEC is the flavor that proves the flat rule has to be per identifier rather than per flavor.** Of its 9,730 dated identifiers, 9,728 carry a year alone, but **10 carry a month** (`IEC CA 01:2025-10`, `IECEE 02:2025-06(en)`, `IECEE AD-001:2024-06(en)`) and **2 are undated** (`IEC 60050:--`, `IEC 60050:—`). A flavor-wide "IEC dates are year-only" switch would have thrown those twelve away silently. `spec/pubid/flat_scalar_serialization_spec.rb` asserts both directions on IEC for exactly this reason.
 
-- **`WorkingDocument` is the one IEC type that still renders plain under `annotated: true`.** The cross-flavor annotator (`Renderers::Annotator`, see the root `CLAUDE.md`) wraps the identifier's own component values where they appear in the rendered string; a WorkingDocument's identity lives in `technical_committee` / `wd_number` / `wd_stage` / `wd_language`, none of which is one of those components. This is the same unserialized-identity gap already recorded above (`to_hash` is `{"_type"=>"pubid:iec:working-document"}` and `root.number` is nil), seen from the rendering side. Nothing raises — the flag is accepted and simply finds no token — so it is a missing span, not a failure. Fixing it means mapping those attributes, which forces a `relaton-data-iec` re-crawl, and remains the separate decision it already was. (hand-off: iec-unserialized-identity-attributes.)
+- **`WorkingDocument` annotates, and since #687 it serializes too.** The cross-flavor annotator (`Renderers::Annotator`, see the root `CLAUDE.md`) wraps the identifier's own component values where they appear in the rendered string; a WorkingDocument's identity lives in `technical_committee` / `number` / `wd_stage` / `wd_language`, so `annotation_tokens` names them explicitly (the wd form's document number IS the inherited `number` since #687 — it used to sit in the unserialized `wd_number`). The rendered annotated form is byte-identical before and after #687 (`<span class="docnumber">1</span>/<span class="docnumber">2457</span>/<span class="stage">FDIS</span>` both sides) — the gap #687 closed was the serialization one recorded above, not the rendering one. (hand-off: iec-unserialized-identity-attributes — landed.)
 
 - **The attribute path takes a plain scalar for language and copublisher — the residue of issue #360 item 3.** Items 1, 2 and 4 gave `IEC PNW 1000-1:2023 ED2` the right stage, separator and URN, and the constructor contract gave it `edition: 2` and `year: 2023`. The URN the issue asked for also carries a language, and that half still failed: `languages: ["en"]` stored raw Strings, so `to_s` raised `wrong number of arguments` at `lib/pubid/iec/single_identifier.rb:59` (`lang.to_s(lang_single:)`) and `to_urn` raised `undefined method 'code'` at `lib/pubid/iec/urn_generator.rb:190`. `copublishers: ["ISO"]` raised the same way on `body`. Both are now coerced by the shared `DEGENERATE_COMPONENT_FIELDS` rule on `::Pubid::Identifier` (see the root `CLAUDE.md`), not by an IEC converter — and IEC is the flavor that proves the wrap has to key on the component **class**, because its copublisher is the subclass `Pubid::Iec::Components::Publisher`, which an attribute-name or exact-class table would miss. So `Identifiers::InternationalStandard.new(number: "1000", part: "1", year: 2023, edition: 2, languages: ["en"])` now renders `IEC 1000-1:2023 ED2(en)` and yields `urn:iec:std:iec:1000-1:2023::ed-2:en`. The Hash and component forms are unchanged and still agree with the scalar form. **`number` was deliberately NOT coerced** — and the bullet below resolves that tension the other way round, by making the attribute a `:string` so a String is the type rather than a tolerated intruder. Note the copublisher renders `IEC/ISO 1000`, IEC's own ordering — the coercion does not touch order. (hand-off: iec-degenerate-component-coercion.)
 
@@ -28,19 +33,19 @@ These notes were part of the root `CLAUDE.md`. Read them before you change `lib/
   **Verified by replaying a captured baseline, not by the suite alone.** Over all **12,331** IEC pass-fixture ids, `to_s`, `to_urn`, `to_mr_string`, `root.number`, the identifier class, the deep-key-sorted `to_hash`, the `from_hash(to_hash) == to_hash` flag and the `from_hash(to_hash) == id` flag are **byte-identical, 0 diffs in every field**. The verbatim `to_hash` differs for 9,950 rows and for **one reason only** — `number`/`part`/`subpart` moved to the end of the key order, because lutaml *appends* a subclass's `key_value` maps to the parent's; removing those three keys leaves the two key sequences identical for every row, and `Hash#==` ignores order, so **0 rows differ in any other way**. Over all **7,573** ISO pass-fixture ids every field including the verbatim hash is byte-identical, so neither shared-file change moved ISO. Against the published `relaton-data-iec` `index-v2.yaml` (**32,847** rows) `from_hash(row).to_hash == row` holds for 32,845 and `parse(from_hash(row).to_s) == from_hash(row)` for all 32,847, **both before and after** — the 2 misses are the pre-existing rows storing the old nested `edition: {number: …}` form. **No index migration follows**, and that is measured rather than assumed. Locked by `spec/pubid/iec/number_string_spec.rb` (structural tripwire over the whole class tree, the constructed-vs-parsed `==`, the wrapper hashes, and the absence of the deleted constant) and by IEC's entry in `spec/pubid/number_string_retype_spec.rb`.
   **Known residue, deliberately unchanged:** after the map moved, `Pubid::Iec::Identifier.from_hash({"number" => "60038"})` — a hand-written hash with **no `_type`**, on the bare base class — silently drops the key instead of building an `Identifier` that holds it. That object is already degenerate (`to_s` on it raises `NoMethodError` for `publisher_portion`) and nothing in this repo or in `relaton-iec` constructs one. (hand-off: iec-number-string.)
 
-  **Two parent-slot gaps recorded by the cross-flavor `base` survey** (`lib/pubid/bsi/CLAUDE.md` has the measured version). `Identifiers::TestReportForm#cispr_identifier` is a polymorphic `Identifier` parent slot with no `#root` override, so `root` returns the form itself. And `Identifiers::ConsolidatedIdentifier#base_document` returns `identifiers&.first` **without** the recursive `&.base_document` that its BSI, CSA and CEN counterparts use — a one-level peel where the others peel fully, so a consolidated edition of a supplement does not reduce to the bare standard. Neither is chased here; both are pinned by nothing, which is why they are written down.
+  **Two parent-slot gaps recorded by the cross-flavor `base` survey** (`lib/pubid/bsi/CLAUDE.md` has the measured version). `Identifiers::TestReportForm#cispr_identifier` is a polymorphic `Identifier` parent slot with no `#root` override, so `root` returns the form itself — ratified as DELIBERATE by #687 (a TRF is a form about the CISPR document, not a wrapper of it; see the unserialized-identity bullet). And `Identifiers::ConsolidatedIdentifier#base_document` returns `identifiers&.first` **without** the recursive `&.base_document` that its BSI, CSA and CEN counterparts use — a one-level peel where the others peel fully, so a consolidated edition of a supplement does not reduce to the bare standard. Neither is chased here; both are pinned by nothing, which is why they are written down.
 
   **Stage codes — the SSOT and what is settled.** The authoritative source for IEC stage abbreviations is committed in this repo: **`data/iec/stage-codes.md`** — the 52-row IEC/ISO stage-code correspondence table transcribed from IEC's official *IEC/ISO Harmonized APIs* documents (v0.0.8 2023-06, api-portal.iec.ch; v0.0.7 2021-05, iec-apiportal.apigee.io — the successors that replace the IEC Data Feeds), the legacy Data-Feeds field examples, and the IS-lifecycle ladder from `StageCodes_IS_v2.pdf` (2023-06). `www.iec.ch` itself is fully WAF-gated to non-browsers and the Wayback Machine holds no snapshots of its PDFs, so the extract — not the unreachable original — is the reference. **`Pubid::Iec.locate_stage` reads `TYPED_STAGES` only** (`lib/pubid/iec.rb`); `PROJECT_STAGES` (AFDIS, TCDV, PRVC…) are not queryable through it. Every one of metanorma-iec's 55 `STAGE_CODES` — the list quoted in pubid#237 — resolves, locked by `spec/pubid/iec/typed_stages_spec.rb` in the issue's own order; aliases are always **appended, never prepended**, so `abbr.first` (what renders) is unchanged. Notable rulings from that pass: **ADIS** is the legacy Data-Feeds spelling of AFDIS ("Approved for FDIS circulation", ISO_LEVEL 40.99) and is an alias on `cdv` (the typed stage claiming 40.99); **CDISH/CDPAS** are the 50.20 rows of the DISH/DPAS deliverables, not "committee draft ISH/PAS" (30.xx); **Reconciliation items recorded in the extract, deliberately not applied:** DELPUB sits at 99.60 there while the `is` ladder does not claim 99.60; CAN carries two codes (30.98 IS/TR/TS, 50.98 PAS/ISH/COR); the rows CDISH-table-siblings DECDISH/MERGED/SPLIT/PCC/PRVN/DEL/RPUB/TPUB and the per-deliverable translation/preparation families are not registered as typed-stage abbreviations.
 
-## pubid-scheme URN (issue #670)
+## pubid-scheme URN (issue #670; gaps closed by #687)
 
 `lib/pubid/iec/pubid_urn_schema.rb` completes the partial schema the IEEE
 and GOST waves landed; rulings in `docs/URN-SCHEMES.adoc` (`[[urn-iec]]`).
 Invariant: every atom added after those waves sits after `database-` and
 before `lang-`, so a landed URN encodes identically (`database` became the
-bool the corpus serializes before any URN carried it). The URN inherits
-the unserialized-identity gaps pinned above: a WorkingDocument's URN is
-the type-only `urn:pubid:iec:wd` (all WD rows share one hash — a shared
-hash, not a collision), and a TestReportForm's CISPR document rides
-`cispr_identifier`. Mapping them stays the wire-format decision it
-already was.
+bool the corpus serializes before any URN carried it). The two identity
+gaps the wave inherited are closed (#687, atoms `tc-`/`wdlang-`/`wdstage-`/
+`wpstage-`/`wptype-`/`cispr-` after `vap-`, still before `lang-`): a
+WorkingDocument URNs per document instead of the shared type-only
+`urn:pubid:iec:wd`, and a TestReportForm's CISPR document nests under
+`cispr-`. All landed URNs are unchanged.

data/lib/pubid/iec/builder.rb

@@ -484,10 +484,28 @@ module Pubid
             end
           end
 
-        when :technical_committee, :subcommittee, :wd_number, :wd_stage, :wd_language, :wp_stage, :wp_type
+        when :technical_committee, :subcommittee, :wd_stage, :wd_language, :wp_stage
           # Working document/programme fields - just return as string
           value.to_s
 
+        when :wd_number
+          # The working document's own number IS the inherited `number`
+          # (issue #687, the "prefer the inherited attribute over a parallel
+          # one" rule): `root.number`, the index key and `to_hash` all read
+          # `number`, so a parallel `wd_number` attribute left every working
+          # document keyed on the empty string. The returned hash routes the
+          # value through assign_attributes onto `number`.
+          { number: value.to_s }
+
+        when :wp_type
+          # The grammar captures a trailing space ("PWI TR 100-36" → "TR ");
+          # the renderer used to strip it at print time. Strip at the
+          # construction site instead, so the stored value — and therefore
+          # to_hash and the URN — is what renders. An all-space value drops
+          # out entirely (nil is skipped by assign_attributes).
+          stripped = value.to_s.strip
+          stripped.empty? ? nil : stripped
+
         when :cispr_identifier
           # CISPR identifier object for TRF
           value

data/lib/pubid/iec/identifiers/test_report_form.rb

@@ -11,6 +11,35 @@ module Pubid
         attribute :cispr_identifier, Identifier, polymorphic: true, default: -> {
         }
 
+        # The embedded CISPR document is part of the form's identity, so it
+        # serializes under its own key (issue #687 — before it, a
+        # CISPR-bearing TRF hashed to just its publisher and year and never
+        # matched after a from_hash round trip). The converter shape is
+        # VapIdentifier#base_to_kv / #base_from_kv: the nested identifier
+        # carries its own polymorphic `_type`, so to_hash embeds the nested
+        # hash and from_hash dispatches back through the type map.
+        key_value do
+          map "cispr_identifier",
+              with: { to: :cispr_identifier_to_kv,
+                      from: :cispr_identifier_from_kv }
+        end
+
+        def cispr_identifier_to_kv(model, doc)
+          cispr = model.cispr_identifier
+          return unless cispr
+
+          doc.add_child(
+            Lutaml::KeyValue::DataModel::Element.new("cispr_identifier",
+                                                     cispr.to_hash),
+          )
+        end
+
+        def cispr_identifier_from_kv(model, value)
+          if value
+            model.cispr_identifier = ::Pubid::Iec::Identifier.from_hash(value)
+          end
+        end
+
         # TRF has type abbreviation for parsing
         TYPED_STAGES = [
           Pubid::Components::TypedStage.new(

data/lib/pubid/iec/identifiers/working_document.rb

@@ -11,12 +11,28 @@ module Pubid
       #   Working Document: "100/3705(F)/FDIS"
       class WorkingDocument < Base
         attribute :technical_committee, :string, default: -> {}
-        attribute :wd_number, :string, default: -> {}
         attribute :wd_language, :string, default: -> {}
         attribute :wd_stage, :string, default: -> {}
         attribute :wp_stage, :string, default: -> {}
         attribute :wp_type, :string, default: -> {}
 
+        # The working document's identity attributes are plain strings, so no
+        # converter is needed: lutaml writes the scalar, and
+        # ::Pubid::Identifier#to_hash drops a nil (the `-> {}` defaults are
+        # nil). This is the serialization that issue #687 restored — before
+        # it, `to_hash` was `{"_type" => "pubid:iec:working-document"}` alone,
+        # all 2,318 working documents shared one hash and one URN
+        # (`urn:pubid:iec:wd`), and no row of this class ever matched after a
+        # from_hash round trip.
+        key_value do
+          map "technical_committee", to: :technical_committee,
+                                     render_default: false
+          map "wd_language", to: :wd_language, render_default: false
+          map "wd_stage", to: :wd_stage, render_default: false
+          map "wp_stage", to: :wp_stage, render_default: false
+          map "wp_type", to: :wp_type, render_default: false
+        end
+
         # Working documents have no TYPED_STAGES - they use PROJECT_STAGES only
         TYPED_STAGES = [].freeze
 
@@ -29,14 +45,15 @@ module Pubid
         end
 
         # A working document prints its committee, number, language and stage —
-        # `1/2457(F)/FDIS` — and none of those is the `number` the shared token
-        # table looks for, so it had no annotatable token. `wd_stage` is the
-        # printed stage token ("FDIS"); the derived `#stage` reader covers only
-        # the PWI/PNW work-programme form and is nil here.
+        # `1/2457(F)/FDIS` — and the document number is the inherited
+        # `number` (issue #687: it used to sit in a parallel `wd_number`
+        # attribute that nothing serialized). `wd_stage` is the printed stage
+        # token ("FDIS"); the derived `#stage` reader covers only the PWI/PNW
+        # work-programme form and is nil here.
         def annotation_tokens
           super + [
             [:technical_committee, "docnumber"],
-            [:wd_number, "docnumber"],
+            [:number, "docnumber"],
             [:wd_language, "language"],
             [:wd_stage, "stage"],
             [:wp_stage, "stage"],

data/lib/pubid/iec/pubid_urn_schema.rb

@@ -16,13 +16,15 @@ module Pubid
     # published URNs are unchanged. `database` was a string atom before any
     # URN carried it — the corpus holds only `true`, so it is a bool.
     #
-    # Known identity gaps the URN inherits, because it carries exactly what
-    # to_hash carries (pinned in lib/pubid/iec/CLAUDE.md): a WorkingDocument
-    # serializes only `_type`, and a TestReportForm's CISPR document sits in
-    # the unserialized `cispr_identifier` — such rows emit type-only URNs
-    # (`urn:pubid:iec:wd`) and are one hash, so they are not a collision.
-    # Mapping them is a wire-format change and stays the separate decision
-    # it already was. The official RFC 5141-style urn:iec:std: form stays
+    # The identity gaps this schema used to inherit are closed (issue #687):
+    # the working document's committee/number/language/stage and work-programme
+    # stage/type serialize (`tc-`, `wdlang-`, `wdstage-`, `wpstage-`,
+    # `wptype-`, with the document number in the positional core), and a
+    # TestReportForm's CISPR document rides `cispr-` as a nested id — so a
+    # working document URNs per document instead of the shared type-only
+    # `urn:pubid:iec:wd` it used to be. The new atoms sit after `vap-` and
+    # before `lang-`, per the invariant below, so every landed URN encodes
+    # identically. The official RFC 5141-style urn:iec:std: form stays
     # on #to_urn.
     PubidUrnSchema = Pubid::PubidUrn.define("iec") do
       atom "publisher", "pub"
@@ -41,6 +43,12 @@ module Pubid
       atom "sheet_number", "sheet"
       atom "sheet_year", "sheetyear"
       atom "vap", "vap", kind: :list
+      atom "technical_committee", "tc"
+      atom "wd_language", "wdlang"
+      atom "wd_stage", "wdstage"
+      atom "wp_stage", "wpstage"
+      atom "wp_type", "wptype"
+      atom "cispr_identifier", "cispr", kind: :id
       atom "languages.*.code", "lang"
       atom "identifiers", "ids", kind: :ids
 

data/lib/pubid/iec/renderer.rb

@@ -208,8 +208,10 @@ module Pubid
           # International Standard, so rendering would not be a fixed point.
           parts = []
           parts << id.wp_stage
-          if id.wp_type && !id.wp_type.strip.empty?
-            parts << " #{id.wp_type.strip}"
+          if id.wp_type && !id.wp_type.empty?
+            # wp_type is stripped at the construction site (Builder#cast),
+            # so the stored value is what renders.
+            parts << " #{id.wp_type}"
           end
 
           if id.number
@@ -226,8 +228,10 @@ module Pubid
         parts = []
         parts << id.technical_committee if id.technical_committee
 
-        if id.wd_number
-          num_part = id.wd_number.to_s
+        # The working document's own number is the inherited `number`
+        # (issue #687).
+        if id.number
+          num_part = id.number.to_s
           num_part += "(#{id.wd_language})" if id.wd_language
           parts << num_part
         end

data/lib/pubid/iso/CLAUDE.md

@@ -143,3 +143,18 @@
     (`data/parg/iso.parg`) must move together — the twin is the embedded
     source's provenance copy.
 
+- **`typed_stage.original_abbr` is spelling, so `==` skips it (issue #656).**
+  The parse path stores the input token (`Amd.`, `COR`, `DAmd`, `pDCOR`) and
+  `stage_from_kv` stores the canonical one (`Amd`, `Cor`, `DAM`); the value is
+  not serialized. Before the fix, 90 pass ids (supplements, Directives,
+  Guides, `ISO/NWIP`) were not `==` after `from_hash(to_hash)`, so
+  `#matches?` against an index row returned false with no error. ISO cannot
+  delete the write as JCGM and IEC did, because `detect_stage_format_long`
+  reads it. `Components::TypedStage` now declares
+  `equality_ignored :original_abbr`. Since `to_s` reads the value, two `==`
+  ids may render differently, which the #656 rule allows. Locked by
+  `spec/pubid/iso/typed_stage_round_trip_spec.rb`.
+  **Still open, and not covered by that spec**: 14 pass ids stay unequal for
+  another reason — `DirectivesSupplement#supplement_publisher` is an
+  `Iso::Components::Publisher` after `parse` and a `Pubid::Components::Publisher`
+  after `from_hash` (11 supplements, 3 bundles).

data/lib/pubid/subset_match.rb

@@ -169,9 +169,7 @@ module Pubid
     # Exactness composes rather than flattening: a nested value that
     # includes this module is matched in BOTH directions with its own
     # `===`, so the class that owns it keeps its rule. A plain `==` would
-    # override it — `Components::TypedStage` ignores `original_abbr`
-    # (the input spelling, `Amd` against `AMD`), and CEN/CENELEC declares
-    # `typed_stage` strict.
+    # override it — CEN/CENELEC declares `typed_stage` strict.
     # @return [Boolean]
     def self.exact_match?(mine, theirs)
       return true if blank?(mine) && blank?(theirs)

data/lib/pubid/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Pubid
-  VERSION = "2.0.18"
+  VERSION = "2.0.20"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: pubid
 version: !ruby/object:Gem::Version
-  version: 2.0.18
+  version: 2.0.20
 platform: ruby
 authors:
 - Ribose Inc.
@@ -38,6 +38,20 @@ dependencies:
     - - "~>"
       - !ruby/object:Gem::Version
         version: 1.3.122
+- !ruby/object:Gem::Dependency
+  name: benchmark
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
 description: PubID is a Ruby library for parsing and generating international publication
   identifiers (PubIDs). It provides a simple and flexible API for working with PubIDs,
   allowing developers to easily create, manipulate, and validate PubIDs in their applications.