pubid 2.0.31 → 2.0.33

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: fe504b8e88ce92ba01d6969f2b3ee6ab295cf526e98be6be1ce2e599a48919fc
-  data.tar.gz: 2d3f32e02b8a11bc7e442be1f87182ecff153282d6c2f23f3f3793923976d846
+  metadata.gz: 8aef45a3d166689b823ce4ab85ab8651e066838a49ddf5b7d23e40345c1a9e0c
+  data.tar.gz: 0d9e256b0cfcdb8ba4354b7e16836e1a0f236f14f5f682bce0850504e4af16ae
 SHA512:
-  metadata.gz: f496ea7193892404ab19c6b294a8715159ce8c66ec1a68dc44f04b11b9d680dec5e8fc46cad783b23a404816aeeedf2ad0a5db7426dc7e9ff7c6a1121ecaa4e2
-  data.tar.gz: 11d90694f5985180a5f0319a2b91d76bce661f25a89afbaf175933ea364758e80b3466ab7f902bf13b6d2b7a39072a420c3b4ec647e662453af50d818627360e
+  metadata.gz: 7cdbc8c5709b74a0a3298a56b79ea954a689a95f94868c2c3a0973ad5912c07581a03f797944caf5f62414ad71cb6c27f9a4b0243351b883962adc67a5b5b2ff
+  data.tar.gz: 00debefc3b8438a48a158c4a706817807f75e2489cc8a3f8ceb247ebd7598592aa4927c6a0942730db12dba5d7e9af0ae875d126f7d7a321732eb2d866e7388d

data/data/iso/update_codes.yaml

@@ -2,6 +2,11 @@ ISO/R 657/IV: ISO/R 657-4:1969
 ISO/IEC/IEEE 8802-22.2:2015/Amd.2:2017(E): ISO/IEC/IEEE 8802-22:2015/Amd.2:2017(E)
 ISO/TR 17716.2: ISO/TR 17716
 ISO/TR 17716.2(E): ISO/TR 17716(E)
+# Unassigned-part placeholder: ISO's open data writes `-_` where a
+# multipart project has no part number yet ("ISO/NP TS 6041-_", title
+# "… Part _: _No part"). It carries no identity, so drop it and leave the
+# part nil (relaton/relaton-data-iso#43).
+"/(\\d)-_(?=[\\s:\\/(]|$)/": '\1'
 # Long-form draft typed stages (DIS TR, FDIS PAS, ...) → canonical short
 # typed-stage abbrs (DTR, FDPAS, ...). DIS/FDIS are typed-stages of the
 # IS class only — for TR/TS/PAS/ISP they have dedicated D<type>/FD<type>

data/lib/pubid/iec/CLAUDE.md

@@ -47,6 +47,8 @@ These notes were part of the root `CLAUDE.md`. Read them before you change `lib/
 
 - **The dotted supplement sub-number survives (issue #725 — the gap the #724 wave recorded).** The ballot faces print it — `/DAM 2.3:2022`, `/DCOR 1.3` on the base, and the dotted base designation `60038-2.3` in the ballot-prefix faces — and the ISO flavor has always kept its dot (its declared field is `stage_iteration`, a `Components::Iteration`; the issue's "`sub_number`" label is prose for this facet). The IEC losses were NOT grammar-shaped: the raw bound hash already carried `supp_stage_iteration`/`stage_iteration` on every dotted shape — they were **build-side and render-side**: `Builder#build_supplement` dropped the peeled `stage_iteration` (so `/DAM 2.3:2022` silently renumbered to `DAM 2`), `ensure_base`'s designation-move left the dot on the supplement (so `IEC/FDAM 60038-2.3` held it as a facet of the implied FDAM 1 instead of the base), and both `Base#number_portion` and `render_supplement_attached` dropped the facet at print time (so even the losslessly-parsed `IEC 60038-2.3` rendered as its sub-number-less twin). The facet rides the layer whose designation carries the dot: the SUPPLEMENT for the on-base draft faces, the BASE for the ballot-prefix faces and the plain dotted base (`parse("IEC/FDAM 60038-2.3") == parse("IEC 60038-2.3/FDAM 1")` still holds — the legacy rule now claims the dotted designation itself, so the ballot face and the on-base face are the same identifier). The grammar change is width only: the three dot-sites (base `stage_iteration`, `supp_stage_iteration`, `supp2_stage_iteration`) admit `1*digit` instead of one digit (`\.\d+`; nothing downstream starts with a digit, so no guard), baked at the CI-pinned compiler. **Surfaces:** `to_s` renders the dot on both layers (`/DAM 2.3:2022`, `IEC 60038-2.3`), `to_hash` serializes it (the flat `stage_iteration` scalar), the pubid-scheme URN already had the `iter-` atom, and the official URN rides the dotted number in the supplement's number slot (`amd:2.3:2022`; ISO keeps the facet in its version atom — this scheme has none, so the number slot carries the designation as printed). The MR string stays iteration-less, matching ISO (`iso.10646_amd.2.2022` is also iteration-less). **Unattested shapes deliberately stay raising:** a dotted ballot NUMBER in the prefix faces (`IEC/DCOR 1.3 60038-2`, `ISO/IEC PDAM 2.3 10646:2012`) — no source prints it, so no identity is guessed for it. **Measured by the replay gate:** over all 26,861 full-fixture inputs, `to_s`/`to_hash`/`to_urn`/`to_mr_string`/class are byte-identical to the 2.0.30 baseline — 0 diffs (no corpus row carried the facet). Locked by `spec/pubid/iec/dot_subnumber_spec.rb` (the on-base, ballot-prefix, plain-base and raising rows) and the testsuite corpus round (pass rows in `pass/`, never `full/`). (hand-off: iec-dot-subnumber.)
 
+- **The standalone no-part dotted face is rejected at parse (issue #727).** `IEC/AMD 1.2:2017` parsed and then LOST the dot at render: `render_supplement_standalone` prints the base number and the supplement's own number only, so the stored `stage_iteration` was dropped and `parse(to_s) != parse` — the URN (`amd:1.2:2017`) and `to_hash` carried a facet the canonical face cannot print. No IEC source prints a no-part standalone face with a dotted number (the facet is attested only on the on-base draft faces; 0 rows in the gem fixtures, the testsuite ledger or `relaton-data-iec`), so the shape is **rejected at parse** — the #726 ruling for unattested dotted shapes — rather than built unrenderable. **The crash the issue reported does not reproduce on the Ruby side, and that finding shaped the fix**: the Ruby renderer nil-guards `copublishers` (`copublishers&.any?`) and the parse path always materializes the publisher (the attribute default fills it; the only `publisher_portion` crash needs a hand-set `publisher = nil`, unreachable from parse and from_hash — verified at 2.0.29, the released 2.0.31 and main over a ~40-shape probe matrix); the crash in the issue came from pubid-ts#184's note, where the TS `publisherPortion` reads `copublishers` off a supplement whose table does not materialize the empty collection. So the nil-guard branch of the issue's suggested direction had nothing to guard, and the live Ruby defect was the lossy render on the same face. **The check is the SHAPE, not a grammar route**: `Identifier.parse` walks the built object's `base` chain after `ensure_base` and raises `ParseError` (with `input:`/`flavor:` set, `parse_failure_cause` nil — the synthetic-raise shape) when a `SupplementIdentifier` layer holds both a `stage_iteration` and a synthetic self-base. One condition covers all three parse routes — the `supp_*` standalone branch (`IEC/AMD 1.2:2017`), the publisher-less plain route (`AMD 1.2:2017`, bound as `type_with_stage` + `stage_iteration`, no `supp_` keys) and a dotted layer under an undotted second supplement (`IEC/AMD 1.2:2017/COR 1:2018`) — and none of the neighbours that keep the dot on a printable layer: the single-document draft faces (`IEC CD 1.2:2017`, `IEC/DTR 1.2:2017` — not supplements, the dot rides the base and `parse(to_s) == parse` holds), the DAM-family decodes (`IEC/DAM 1.2` → `IEC 1.2/DAM 1`, dot on the base designation) and the on-base faces (a real base, `synthetic_base?` false). The undotted standalone faces keep parsing with their #693 residue renders (`IEC/AMD 1:2017` → `IEC Amd 1-1:2017`). BSI's delegation (`bsi/single_identifier.rb` → `Pubid::Iec.parse`) and `Pubid.parse` routing raise through the same entry. **`from_hash` deliberately still deserializes the shape** (rendering the undotted face as it always did): parse can no longer produce the hash, no stored row carries it, and rejecting deserialization would hard-break a consumer cache holding a pre-#727 parse — the residue is pinned, not chased. **No corpus round follows** (the shape is unattested; nothing moves to `pass/`). Locked by the new blocks in `spec/pubid/iec/dot_subnumber_spec.rb`: the 11 raising rows with the message, the input/flavor/cause shape of the synthetic raise, the keep-parsing rows (render + `from_hash(to_hash) ==` + re-parse idempotence), the undotted residue rows, and the deserialization residue. (hand-off: iec-standalone-dotted-reject.)
+
 ## pubid-scheme URN (issue #670; gaps closed by #687)
 
 `lib/pubid/iec/pubid_urn_schema.rb` completes the partial schema the IEEE

data/lib/pubid/iec/identifier.rb

@@ -275,9 +275,44 @@ module Pubid
         # parse holds that layout before any reader runs
         # (SupplementIdentifier#number).
         id.ensure_base if id.is_a?(SupplementIdentifier)
+        reject_standalone_dotted_supplement(id, string)
         id
       end
 
+      # A supplement's OWN dotted sub-number has nowhere to print on the
+      # standalone (self-base) face: render_supplement_standalone prints the
+      # base number and the supplement number only, so the parsed dot was
+      # dropped at render and parse(to_s) != parse. No IEC source prints a
+      # no-part standalone face with a dotted sub-number — the facet is
+      # attested only on the on-base draft faces ("IEC 60038-2/DAM 2.3:2022")
+      # — so the shape is rejected here instead of being built unrenderable
+      # (pubid#727; the same ruling as the prefix-face dotted ballot numbers,
+      # which the grammar keeps raising). The walk reaches the dotted layer
+      # when it sits under an undotted second supplement ("/AMD 1.2:2017" as
+      # the base of a "/COR 1" chain). The single-document draft faces
+      # ("IEC CD 1.2:2017") and the DAM-family decodes ("IEC/DAM 1.2") keep
+      # their dot on a non-supplement layer and pass; the on-base faces have
+      # a real base, so synthetic_base? is false and they pass too.
+      STANDALONE_DOTTED_SUPPLEMENT_MESSAGE =
+        "the standalone no-part supplement face carries a dotted sub-number " \
+        "(e.g. \"IEC/AMD 1.2:2017\"): no IEC source prints one, and the " \
+        "standalone face has no layer to print the dot on, so the stored " \
+        "sub-number would be dropped at render — the dotted sub-number is " \
+        "attested only on the on-base draft forms"
+
+      def self.reject_standalone_dotted_supplement(id, input)
+        node = id
+        while node.is_a?(SupplementIdentifier)
+          if node.stage_iteration && node.synthetic_base?
+            raise Pubid::Errors::ParseError.new(
+              STANDALONE_DOTTED_SUPPLEMENT_MESSAGE, nil,
+              input: input, flavor: "iec"
+            )
+          end
+          node = node.base
+        end
+      end
+
       # from_hash is the shared polymorphic dispatch on Pubid::Identifier.
       # IEC_TYPE_MAP remains as the key_value polymorphic_map.
     end

data/lib/pubid/iso/CLAUDE.md

@@ -207,3 +207,24 @@
     `spec/pubid/iso/ballot_face_decode_spec.rb` (all 13 SSOT anchors, the
     no-number rule, base-whole guards, Am alias, stored-twin agreement, and
     the website-face-vs-stored-face drift rows the SSOT §5 records).
+
+- **The unassigned-part placeholder `-_` is dropped before the parse
+  (relaton/relaton-data-iso#43).** ISO's open data
+  (`iso_deliverables_metadata.jsonl`) writes `-_` where a multipart project
+  has no part number yet — record 95616 is `ISO/NP TS 6041-_`, stage 10.20,
+  title "… Part _: _No part". The grammar's part body is word characters and
+  digits, so the id failed, and the relaton crawler left the file unindexed.
+  A `data/iso/update_codes.yaml` regex removes `-_` after a digit when a
+  separator (`:`, `/`, `(`, space) or the end follows, so the id parses as
+  `ISO/NP TS 6041` with a nil part. A nil part is the honest model (the part
+  is unknown), and `Pubid::Iso.parse` of the `-_` form (the call relaton's
+  crawler and lookup make) gives the same identifier as the index row. The
+  render is normalized, not byte-exact. No grammar or artifact change.
+  **Known gap:** the rule is ISO-only, and IEC's grammar accepts `_` as a
+  part, so `Pubid.parse("ISO 6041-_")` routes to IEC (an exact round trip
+  wins) with part `"_"`. The reported `ISO/NP TS 6041-_` still reaches ISO,
+  because IEC cannot read the NP stage. BSI and CSA delegate to ISO and get
+  the normalization. Over the whole open-data file (81,625
+  non-`Withdrawn` references, 2026-11-14) this was the only failure; the
+  `Withdrawn …` rows fail too, but the crawler skips them. Locked by
+  `spec/pubid/iso/update_codes_spec.rb`.

data/lib/pubid/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Pubid
-  VERSION = "2.0.31"
+  VERSION = "2.0.33"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: pubid
 version: !ruby/object:Gem::Version
-  version: 2.0.31
+  version: 2.0.33
 platform: ruby
 authors:
 - Ribose Inc.