pubid 2.0.29 → 2.0.30

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c56349e19d87e6f18a3704f1f1ee4256572485aa3917660803ce6265131a782b
-  data.tar.gz: 40baabfdee09da5b4bf2aec91082aacb1ba729bcc2f6b5d89d1e9d784c9997a7
+  metadata.gz: 34bcdf345b40aa5e1bc95e74b7c110e474c2019bda1e16e1cbbbe9988677427f
+  data.tar.gz: a2fa7df1b0c06f2ba455e55b88bf0e1b4891f15e6ae3e73fab71e698f3e88570
 SHA512:
-  metadata.gz: 27ae4a98098399556651770b46bece23f167906b6313050f06a5539ba269809d578442e6b47ace060f03db3fd0f3c63ec553461deb544ddb1051af455a631ef5
-  data.tar.gz: fbc0dfbdcb34f5d563071d6d74d5607b2545620d9491a42a8387b4f9da829020f0590c9391057dfbb8147cd28b6dea2056d1952535f6ad48c28891b7129ce673
+  metadata.gz: 5dbd5dffdb3a62b01d6d0692bb3b9bbe3c0cf1fda27c6425b4cafb9a393be1c81c550cda411d60a2888d2234e16f3820ed9253b057de2ca705cb19893ae15da8
+  data.tar.gz: f9c0a0ebf03670237ad1da8f21e575ce03e81eb73a8ed8b623bc98fcbca27130bbf23c69d72e98a51128dfd29185c325ed8d1c7fc193db97db87f9af62a14c8f

data/lib/pubid/iec/CLAUDE.md

@@ -41,8 +41,8 @@ These notes were part of the root `CLAUDE.md`. Read them before you change `lib/
 
 - **A standalone supplement is `==` its `from_hash(to_hash)` round trip — the cause was the synthetic base's stage, not the ivar swap.** A supplement written without its base (`IEC/FDAM 60038-1`, `IEC/DCOR 60038-1`, `IEC/DISH 60038-1`) gets a bare `Pubid::Iec::SingleIdentifier` base from `SupplementIdentifier#ensure_base`; `to_s`, `to_urn`, `to_hash` and `#hash` agreed on both paths, but `==` was false, so `#matches?` failed silently. **The hand-off blamed the swapped `@number`/`@part` ivars; that was wrong.** `EqualityIgnored#==` compares through `public_send`, so the reader overrides hid the swap from it. The real difference was `base.typed_stage`: `ensure_base` set an IS/undated `TypedStage` with no `code`, which `stage_to_kv` cannot serialize, so `from_hash` rebuilt the base with none. **Diagnose an `==` failure by diffing the attributes through their readers, recursively, before trusting a hypothesis about ivars.** The stage write is deleted (the JCGM "delete the write" shape); the one visible change is `root.to_s` of the synthetic base on the parse path (`IEC IS60038` → `IEC60038`, the value `from_hash` already gave). The ivar swap was cleaned up too, because it was a hazard: `ensure_base` now moves the supplement number from `@part` into `@number`, so the stored layout is what `from_hash` rebuilds, and the `#number`/`#part` overrides, `@supplement_number`, `@synthetic_base` and `mark_synthetic_standalone!` are gone. `synthetic_base?` reads `base.instance_of?(SingleIdentifier)`, so it also survives `#exclude`, which copies `base` into a new object (on `main` the copy lost the flag and rendered in attached form, `IEC IS60038/FDAM1`). **The move must not depend on which reader runs first**, and a first version got that wrong: it called `ensure_base` only in the slash-form branch of `Builder#build`, but the canonical spelling `IEC FDAM 60038-1` (what `to_s` prints) reaches a different grammar branch, so a fresh parse kept the grammar layout until something read `base` — `.number` answered `60038`, `to_mr_string` gave `iec.60038_amd.60038`, and `#exclude` (which copies through the readers) copied the unmoved numbers onto a base synthesised afterwards, rendering `IEC FDAM 60038-60038`. So `#number`/`#part` keep a thin override that only TRIGGERS the move (`settle_standalone_layout`: `ensure_base` when there is a `part` and no base) and then return the stored value, and `Iec::Identifier.parse` calls `ensure_base` on a top-level supplement so a fresh parse already stores the moved layout. Supplements without a part (consolidation members, `IEC/AMD 1:2017`) are untouched by either. Two renderings outside the fixture corpus move, both repairs: `exclude(:date)` of a standalone supplement keeps the standalone form, and `IEC FDAM 60038-1 (all parts)` renders as itself instead of `IEC IS60038/FDAM1 (all parts)`. One hazard is left, unreachable from the grammar: reading `part` on a hand-built supplement BEFORE its base is assigned synthesises a base and moves the number, so a later `base =` renders the moved number. **A second hand-off claim was also wrong: an equal `#hash` for two unequal objects is legal** (the rule is `a == b ⇒ a.hash == b.hash`); `compute_hash` does not read `base`, which is a collision, not a defect. Measured: all 12,331 pass-fixture ids are byte-identical in `to_s`, `to_urn`, `to_mr_string`, `root.number`, `to_hash` and class; all 32,949 `relaton-data-iec` `index-v2.yaml` rows keep `from_hash(row).to_hash == row` and `parse(from_hash(row).to_s) == from_hash(row)`, and none has a `pubid:iec:single-identifier` base, so **no index migration follows**. Locked by `spec/pubid/iec/standalone_supplement_round_trip_spec.rb` and by the six standalone fixtures now in `full/iec-pubid.txt`, which the whole-corpus `==` sweep in `spec/pubid/iec/root_number_spec.rb` covers. **Residue, deliberately unchanged:** the self-base form renders its number twice — `IEC/AMD 1:2017` → `IEC Amd 1-1:2017` — because `render_supplement_standalone` prints `base.number` and then the supplement's own number, which are the same value there. (hand-off: iec-standalone-supplement-ivar-swap.) **CORRECTED 2026-11-12 (owner ruling, reverses this bullet's identity):** the part-to-number move this bullet introduced is the DEFECT the legacy-decode bullet below removes. `IEC/FDAM 60038-1` is not "base 60038 with FDAM 1": the whole `60038-1` is the base designation. The equality machinery this bullet built (the stored layout `from_hash` rebuilds, the reader-triggered settle, `synthetic_base?` reading the class) is kept and holds on the corrected identity.
 
-- **The legacy prefix faces decode onto the modern identity (owner ruling 2026-11-12; evidence: `data/iec/legacy-notation.md`; reverses #693's resolution).** The old notation prints the stage token BEFORE the base designation — the JTC1 ballot faces `IEC/FDAM 60038-1`, `ISO/IEC PDAM 12207`, `ISO/IEC PDAM 2 10646:2012`, `ISO/IEC FPDAM 9075/1`, `ISO/IEC FDAM 1 14496-12:2004`, and the IEC draft faces `IEC/DCOR 60038-1`, `IEC/DISH 60038-1`, `IEC CDISH 60050-191`. #693's model had resolved those by STEALING the designation's part for the supplement (`ensure_base` moved `@part` into `@number`), so the old face and the modern on-base face (`IEC 60038-1/FDAM 1`) disagreed on the same document's identity. The decode now: **the designation moves WHOLE onto the base** (number, part, subpart, and the date — old notation prints the base's year, and a draft supplement carries no publication date of its own), **the supplement's number defaults to 1** (the no-number rule: an unnumbered DAM/PDAM/FDAM/FPDAM/DCOR/DISH means Amendment/Corrigendum/Interpretation-Sheet 1), and **the base is a concrete `InternationalStandard`**, not the synthetic `SingleIdentifier` — that concreteness is what makes the legacy face hash-match (and equal) the modern parse: `parse("IEC/DCOR 60038-1") == parse("IEC 60038-1/DCOR 1")`. The render is the modern on-base slash face (`IEC 60038-1/FDAM1`), so the legacy face is not a spelling facet — it is decoded, IHO-style.
-  **Two grammar halves.** The DAM-family prefix faces (`DAM`/`PDAM`/`FDAM`/`FPDAM`, spelled literally in the rule so the typed_stages table stays put for them) raised on main; the new `legacy_supplement` rule in `pubid-grammar` captures `legacy_stage`, the optional `legacy_supplement_number` (N before the base), and the base designation — including the old slash-part spelling (`FPDAM 9075/1` = base 9075, part 1; captured as `legacy_base_number` + `legacy_base_part`, and ordered BEFORE the dash-part alternative because `number_with_part` also matches a bare "9075", which would have left the "/1" dangling). The rule sits ahead of `standalone_supplement` in the ordered choice so both the slash and space faces reach it, and its date rides with the designation. PDAM/FPDAM joined the `iec_typed_stages` table and `Amendment::TYPED_STAGES` (stage codes mirror ISO's entries: PDAM at the CD level, FPDAM at the FDIS level; the LEAD harmonized codes are the circulation slots 30.20/50.20 — ISO's 30.00/50.00 are CDAM's and FDAM's (type, leading code) URN keys). The non-DAM draft faces (DCOR, FDCOR, CDISH, DISH, and any typed standalone with a designation-shaped token) already parsed; they are corrected by `ensure_base`, so the two paths converge on one decode. **The discriminator between a legacy face and a true standalone is the PART:** a designation token (`60038-1`, `9075/1`) moves whole onto the base; a bare ordinal (`IEC/AMD 1:2017`) is the supplement's own number and keeps the synthetic self-base (`IEC Amd 1-1:2017`, unchanged). Under the JTC1 reading a bare number after a DAM-family token is a base designation (`IEC FDAM 60038` → `IEC 60038/FDAM1`), not a self-base mangle — that residue from the #693 bullet is gone.
+- **The legacy prefix faces decode onto the modern identity (owner ruling 2026-11-12; evidence: `data/iec/legacy-notation.md`; reverses #693's resolution).** The old notation prints the stage token BEFORE the base designation — the JTC1 ballot faces `IEC/FDAM 60038-1`, `ISO/IEC PDAM 12207`, `ISO/IEC PDAM 2 10646:2012`, `ISO/IEC FPDAM 9075/1`, `ISO/IEC FDAM 1 14496-12:2004`, and the IEC draft faces `IEC/DCOR 60038-1`, `IEC/DISH 60038-1`, `IEC CDISH 60050-191`. #693's model had resolved those by STEALING the designation's part for the supplement (`ensure_base` moved `@part` into `@number`), so the old face and the modern on-base face (`IEC 60038-1/FDAM 1`) disagreed on the same document's identity. The decode now: **the designation moves WHOLE onto the base** (number, part, subpart, and the date — old notation prints the base's year, and a draft supplement carries no publication date of its own), **the supplement's number defaults to 1** (the no-number rule: an unnumbered DAM/PDAM/FDAM/FPDAM/DCOR/DISH means Amendment/Corrigendum/Interpretation-Sheet 1), and **the base is a concrete `InternationalStandard`**, not the synthetic `SingleIdentifier` — that concreteness is what makes the legacy face hash-match (and equal) the modern parse: `parse("IEC/DCOR 60038-1") == parse("IEC 60038-1/DCOR 1")`. The render is the modern on-base slash face, and for a draft stage it is the SPACED ballot face (`IEC 60038-1/FDAM 1`; owner ruling 2026-11-14 — the stored/citation twins are spaced: the frozen corpus carries the spaced `/FDAM 1`-family rows against only the decode-wave glued rows, and published amendments stay glued `/AMD1:2019`), so the legacy face is not a spelling facet — it is decoded, IHO-style. The discriminator is the typed stage's own published mark (`stage_code == :published` keeps glued `Amd`/`Cor`/`ISH`; every draft stage prints token + space + number in `Renderer#render_supplement_attached`).
+  **Two grammar halves.** The DAM-family prefix faces (`DAM`/`PDAM`/`FDAM`/`FPDAM`, spelled literally in the rule so the typed_stages table stays put for them) raised on main; the new `legacy_supplement` rule in `pubid-grammar` captures `legacy_stage`, the optional `legacy_supplement_number` (N before the base), and the base designation — including the old slash-part spelling (`FPDAM 9075/1` = base 9075, part 1; captured as `legacy_base_number` + `legacy_base_part`, and ordered BEFORE the dash-part alternative because `number_with_part` also matches a bare "9075", which would have left the "/1" dangling). The rule sits ahead of `standalone_supplement` in the ordered choice so both the slash and space faces reach it, and its date rides with the designation. PDAM/FPDAM joined the `iec_typed_stages` table and `Amendment::TYPED_STAGES` (stage codes mirror ISO's entries: PDAM at the CD level, FPDAM at the FDIS level; the LEAD harmonized codes are the circulation slots 30.20/50.20 — ISO's 30.00/50.00 are CDAM's and FDAM's (type, leading code) URN keys). The non-DAM draft faces (DCOR, FDCOR, CDISH, DISH, and any typed standalone with a designation-shaped token) already parsed; they are corrected by `ensure_base`, so the two paths converge on one decode. **The discriminator between a legacy face and a true standalone is the PART:** a designation token (`60038-1`, `9075/1`) moves whole onto the base; a bare ordinal (`IEC/AMD 1:2017`) is the supplement's own number and keeps the synthetic self-base (`IEC Amd 1-1:2017`, unchanged). Under the JTC1 reading a bare number after a DAM-family token is a base designation (`IEC FDAM 60038` → `IEC 60038/FDAM 1`), not a self-base mangle — that residue from the #693 bullet is gone.
   **Measured by the replay gate, not the suite alone:** over all 26,838 full-fixture inputs, `to_s`/`to_urn`/`to_mr_string`/`to_hash`/class/`root.number` are byte-identical to the `main` baseline EXCEPT exactly the 6 #693-era standalone rows (both faces of FDAM/DCOR/DISH `60038-1`), which re-resolve to the corrected identities; 0 inputs stopped parsing; the prefix faces go raise→parse (12,491 full entries → 12,359 pass, fail steady at 132). #693's equality invariant (`parse == from_hash(to_hash)`) holds on every corrected identity. Locked by `spec/pubid/iec/standalone_supplement_round_trip_spec.rb` (rewritten: legacy faces decode and equal their modern parses, the self-base standalone keeps its form), `stage_separator_spec.rb`, `amendment_spec.rb`, `corrigendum_spec.rb`, `interpretation_sheet_spec.rb`, and the pass fixtures re-pinned by `validation:classify`. **Remaining gaps, deliberately open:** the dotted on-base `/Amd.1` spelling still raises in the IEC flavor (ISO accepts it), and the stamped prefix faces (`ISO/IEC FPDAM 10373-6/Amd 9:2001`, `ISO/IEC PDAM 9541-1/Amd.4`, `ISO/IEC FDAM 19776-2:2005/Am1:2006`, `ISO/IEC PDAM TR 9573-11/Amd.1`) stay raising — their on-base amendment stamps need the dotted/ISO stamp spellings the IEC flavor does not carry (gap 4 of the draft); the ISO-side agent owns that study.
 
 ## pubid-scheme URN (issue #670; gaps closed by #687)

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

@@ -46,9 +46,10 @@ module Pubid
             type_code: :cor,
             # "CDCOR" is the all-caps spelling the attached-supplement
             # renderer emits (abbr.first.upcase) when a legacy prefix face
-            # decodes onto a CDCor draft ("IEC/CDCor 60038-1" renders
-            # "IEC 60038-1/CDCOR1"); appended as an alias so the canonical
-            # render re-parses. abbr.first stays "CDCor" (spelling-only).
+            # decodes onto a CDCor draft ("IEC/CDCor 60038-1" renders the
+            # spaced draft face "IEC 60038-1/CDCOR 1"); appended as an
+            # alias so the canonical render re-parses. abbr.first stays
+            # "CDCor" (spelling-only).
             abbr: ["CDCor", "CDCOR"],
             short_abbr: "CDCor",
             long_abbr: "CD Cor",

data/lib/pubid/iec/renderer.rb

@@ -10,7 +10,7 @@ module Pubid
     #   "IEC 60038:2009"
     #   "IEC/CD 60038"
     #   "IEC 60038:2009/AMD1:2011"
-    #   "IEC/FDAM 60038-1"
+    #   "IEC 60038-1/FDAM 1"  (draft supplement: spaced ballot face)
     #   "IEC 60529:1989+AMD1:1999 CSV"
     #
     # The renderer is registered as the `:human` format in the IEC format
@@ -119,7 +119,7 @@ module Pubid
         parts << id.base.to_s(**opts)
 
         abbr = id.typed_stage.abbr.first.upcase
-        supp_part = "/#{abbr}#{id.number}"
+        supp_part = "/#{abbr}#{ballot_number_separator(id)}#{id.number}"
         supp_part += ":#{id.date.render(context:)}" if id.date
         parts << supp_part
 
@@ -128,6 +128,22 @@ module Pubid
         parts.join
       end
 
+      # Draft-stage supplements carry the ballot face: stage token, SPACE,
+      # number ("60038-1/FDAM 1", "60038-1/DCOR 1.3") — the spaced
+      # citation form the stored twins carry (the ISO corpus holds
+      # /FDAM 1, /PDAM 1, /FPDAM 2, /DAM 2.3:2022, /DCOR 1.3, /DISH 1 …;
+      # the published faces have no spaced twin). The published stages
+      # (stage_code :published — Amd/Cor/ISH) keep the glued
+      # corpus-dominant face ("60038-1/AMD1:2011"). The separator only
+      # rides when there is a number, so a numberless supplement renders
+      # the bare stage token on either side.
+      def ballot_number_separator(id)
+        return "" unless id.number
+        return "" if id.typed_stage.stage_code.to_s == "published"
+
+        " "
+      end
+
       def render_consolidated(id, opts, _context)
         id.identifiers.map.with_index do |sub_id, idx|
           if idx.zero?

data/lib/pubid/iec/supplement_identifier.rb

@@ -103,7 +103,7 @@ module Pubid
       # 2026-11-12, reversing #693's part-to-number move, which stole the
       # base's part for the supplement). The concrete base is what makes the
       # decode hash-match (and equal) the modern on-base parse of the same
-      # document: `parse("IEC/DCOR 60038-1") == parse("IEC 60038-1/DCOR1")`.
+      # document: `parse("IEC/DCOR 60038-1") == parse("IEC 60038-1/DCOR 1")`.
       #
       # A supplement with no part (a true standalone like "IEC/AMD 1:2017",
       # or a bare "+AMD1:2008" in a consolidation) keeps its own number and

data/lib/pubid/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Pubid
-  VERSION = "2.0.29"
+  VERSION = "2.0.30"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: pubid
 version: !ruby/object:Gem::Version
-  version: 2.0.29
+  version: 2.0.30
 platform: ruby
 authors:
 - Ribose Inc.