@mailwoman/corpus 9.4.0 → 10.1.0

package/lib/es/adapters/catastro/adapter.ts

@@ -0,0 +1,343 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ * `es-catastro`: the INSPIRE Addresses theme the Dirección General del Catastro publishes, read
+ * from its per-municipality zipped GML.
+ *
+ * The ATOM service at `https://www.catastro.hacienda.gob.es/INSPIRE/Addresses/ES.SDGC.AD.atom.xml`
+ * lists 55 entries, and 52 of them are the Dirección General del Catastro's own provinces. The other
+ * three link the foral cadastres of Bizkaia, Gipuzkoa and Navarra, which state their own terms and
+ * have an adapter each. Every DGC entry links a province feed listing one zipped GML per
+ * municipality, so a caller harvests municipality by municipality and this adapter reads one archive
+ * or a directory of them.
+ *
+ * Three properties of the file decide the reader, and each of the three would silently empty a row
+ * if it were read the way the other INSPIRE publishers are read.
+ *
+ * The archive carries two members. `A.ES.SDGC.AD.<code>.gml` is the data and
+ * `A.ES.SDGC.AD.MD.<code>.xml` is its ISO 19139 metadata record. A member selector matching `.xml`
+ * therefore selects the metadata record. That record carries no `AD:Address` at all, so
+ * {@linkcode GML_MEMBER} matches the extension that holds the features.
+ *
+ * The GML declares `<?xml version="1.0" encoding="ISO-8859-1"?>` and is not valid UTF-8: Ceuta's
+ * 11,861,975-byte member carries 26 bytes above 0x7F, one of which is the `Ñ` of a street name.
+ * `streamMarkupElements` decodes with a UTF-8 `TextDecoder`, which answers those bytes with U+FFFD,
+ * so the bytes pass through `decodeByteStream` first. {@linkcode assertLatin1Declaration} reads the
+ * declaration rather than trusting this note, because a publisher that re-encodes its output would
+ * otherwise have every accented name mojibaked under a reader that still claimed to be right.
+ *
+ * The file binds the INSPIRE namespaces to upper-case prefixes, `AD:Address` and `GN:text`, and
+ * states a designator type and an administrative level as element text rather than as a codelist
+ * URI. `#inspire/address` matches an element by its local name and accepts either form of a coded
+ * value, so the readers there serve this publisher. Only the element names this module passes to the
+ * markup stream carry the upper-case prefix.
+ *
+ * Each address references exactly three features by local `#` fragment, measured as 8,126 of 8,126
+ * addresses in Ceuta carrying one `ES.SDGC.PD`, one `ES.SDGC.TN` and one `ES.SDGC.AU` reference.
+ * Those features are written after the addresses — Ceuta's 511 of them sit in the last 530,000 bytes
+ * of the member — so the adapter reads the member twice through
+ * {@linkcode streamIndexedInspireRows} rather than holding every address until the document ends.
+ *
+ * The number is one designator, typed `1`. The publisher's own rendering fixes what that type means
+ * and what the value `S-N` means. `Consulta_DNPRC` on the OVC web service answers a cadastral
+ * reference with the address the Cadastre holds for it, as `<ldt>`, plus the fields it is built
+ * from: `tv` the street-type abbreviation, `nv` the street name, `pnp` the number, `plp` its letter,
+ * `dp` the postcode and `nm` the municipality. The cadastral reference is the tail of the address's
+ * own `gml:id`, so every row this adapter writes can be put beside the publisher's own line:
+ *
+ * | `gml:id` tail | designator | `<ldt>` |
+ * | --- | --- | --- |
+ * | `9744803TE8794S` | `13` | `CL SANTIAGO APOSTOL 13 51002 CEUTA (CEUTA)` |
+ * | `9745701TE8794S` | `13D` | `CL SANTIAGO APOSTOL 13(D) 51002 CEUTA (CEUTA)` |
+ *
+ * So the designator is `pnp` with `plp` appended, and the street is `tv` followed by `nv`, which is
+ * the `AD:ThoroughfareName` text verbatim once its leading space is trimmed. The abbreviation stays
+ * in the street component because the publisher writes it in its own rendered line. Expanding `CL`
+ * to `Calle` would state something the Cadastre did not.
+ *
+ * `S-N` is *sin número*, the publisher stating that the address has no number, and it is 856 of
+ * Ceuta's 8,126 designators. A row carrying it writes no `house_number`, because the string is a
+ * statement about the number rather than a number.
+ *
+ * One reference form states an absence the same way. The postal-descriptor reference's final segment
+ * is the postcode, and 70 of Ceuta's 8,126 addresses write `#ES.SDGC.PD.55.101.` with that segment
+ * empty, against five `AD:PostalDescriptor` features the file does publish. Those 70 are the only
+ * unresolvable references in the member: the thoroughfare and administrative-unit references resolve
+ * 8,126 of 8,126 each. A reference whose final segment is empty is therefore read as the publisher
+ * holding no value for that component, and every other unresolved reference raises.
+ */
+
+import { decodeByteStream } from "@mailwoman/core/fs/streams"
+import type { MarkupElement } from "@mailwoman/core/html/elements"
+import { stringifyJSON } from "@mailwoman/core/json"
+import { PathBuilder, type PathBuilderLike } from "path-ts"
+import { Globerator } from "spliterator/node/fs"
+
+import { UnsupportedCountryError } from "#adapters/errors"
+import { cadastralRow, referenceStatesNoValue, spanishHouseNumber } from "#es/cadastre"
+import {
+	adminUnitLevel,
+	componentHrefs,
+	componentJoinKey,
+	designatorsByType,
+	inspireNameIs,
+	placeName,
+	postalDescriptorCode,
+	thoroughfareName,
+	voidDesignatorTypes,
+} from "#inspire/address"
+import { inspireGMLChunks } from "#inspire/archive"
+import { UnresolvedComponentReferenceError, VoidDesignatorError } from "#inspire/errors"
+import { streamIndexedInspireRows } from "#inspire/stream"
+import { SourceRegister } from "#registers"
+import { AddressRole, type AdapterOptions, type CanonicalRow, type CorpusAdapter, SurfaceOrigin } from "#types"
+
+/**
+ * Registry id for this adapter.
+ *
+ * Stamped into every row it emits, so a corpus record can be traced back to the dataset it came from.
+ */
+export const ES_CATASTRO_ADAPTER_ID = "es-catastro"
+
+/**
+ * The jurisdiction this adapter emits.
+ *
+ * The 52 province feeds cover peninsular Spain, the Balearics, the Canaries, Ceuta and Melilla.
+ * The three Basque provinces and Navarra keep their own cadastres and their own adapters.
+ */
+export const ES_CATASTRO_COUNTRIES: readonly string[] = ["ES"]
+
+/**
+ * The license the address-source register elected for this publisher.
+ *
+ * The register elects the titled instrument, `Licencia de acceso y uso de los servicios y conjuntos
+ * de datos INSPIRE de la Dirección General del Catastro` at version 1.0, rather than the older
+ * `all rights reserved` template the feed still carries. It names no license instrument a
+ * Creative Commons label or an SPDX identifier would cover, so the row records the instrument's
+ * own title.
+ *
+ * That decision reads `refused` for `redistribute-data`: clause 1 withholds distribution
+ * of the information as supplied, and requires transformation before any public use.
+ * Ingest, transformation and training are permitted, which is what this adapter performs.
+ */
+export const ES_CATASTRO_LICENSE =
+	"Licencia de acceso y uso de los servicios y conjuntos de datos INSPIRE de la Dirección General del Catastro"
+
+/**
+ * The member of a municipality's archive that holds the GML.
+ *
+ * The archive ships two members and the other is `A.ES.SDGC.AD.MD.<code>.xml`,
+ * the ISO 19139 metadata record.
+ * A selector matching `.xml` reads that record, which carries no `AD:Address`.
+ */
+const GML_MEMBER = /\.gml$/iu
+
+/**
+ * The encoding the publisher declares and writes.
+ */
+const DECLARED_ENCODING = "iso-8859-1"
+
+/**
+ * The feature types an address references, indexed in the first pass over the member.
+ *
+ * `AD:AddressAreaName` is listed although Ceuta publishes none, because the schema admits one and a
+ * municipality that writes one would otherwise have its settlement part read as an unresolved reference.
+ */
+const COMPONENT_ELEMENTS = [
+	"AD:ThoroughfareName",
+	"AD:PostalDescriptor",
+	"AD:AdminUnitName",
+	"AD:AddressAreaName",
+] as const
+
+/**
+ * The element each address is written as.
+ */
+const ADDRESS_ELEMENT = "AD:Address"
+
+/**
+ * The administrative level that is the municipality.
+ *
+ * The publisher writes `<AD:level>4</AD:level>` as text where the other INSPIRE publishers
+ * write the `AdministrativeHierarchyLevel/4thOrder` codelist URI, so both spellings
+ * are accepted and a reference to a coarser unit is left out of the row.
+ */
+const MUNICIPALITY_LEVELS = new Set(["4", "4thOrder"])
+
+/**
+ * The designator type the publisher writes on every address.
+ *
+ * Measured over Ceuta: 8,126 of 8,126 addresses carry exactly one designator
+ * and its `AD:type` is the text `1`.
+ * The value is the house number, which `Consulta_DNPRC`'s `pnp` and `plp` fields
+ * confirm for the two addresses tabulated in this module's header.
+ */
+const HOUSE_NUMBER_TYPE = "1"
+
+/**
+ * Raises unless the member declares the encoding this reader decodes it from.
+ *
+ * The declaration is the publisher's statement about its own bytes, and a reader that assumed ISO-8859-1
+ * through a change to UTF-8 would mojibake every accented street name while reporting success.
+ * A member that states no encoding is accepted, because XML's default is UTF-8
+ * and ISO-8859-1 decoding of ASCII bytes is the same text.
+ */
+async function* assertLatin1Declaration(
+	chunks: AsyncIterable<Uint8Array>,
+	archivePath: PathBuilderLike
+): AsyncIterable<Uint8Array> {
+	let first = true
+
+	for await (const chunk of chunks) {
+		if (first) {
+			first = false
+
+			const declaration = /<\?xml[^>]*encoding="([^"]+)"/iu.exec(Buffer.from(chunk.subarray(0, 512)).toString("latin1"))
+			const declared = declaration?.[1]?.toLowerCase()
+
+			if (declared && declared !== DECLARED_ENCODING) {
+				throw new TypeError(
+					`${ES_CATASTRO_ADAPTER_ID} adapter: ${archivePath.toString()} declares encoding ` +
+						`${stringifyJSON(declaration?.[1])}, and this reader decodes ${stringifyJSON(DECLARED_ENCODING)}`
+				)
+			}
+		}
+
+		yield chunk
+	}
+}
+
+/**
+ * One archive member's bytes, re-encoded into the UTF-8 the markup reader decodes.
+ */
+function memberChunks(archivePath: PathBuilderLike): AsyncIterable<Uint8Array> {
+	return decodeByteStream(
+		assertLatin1Declaration(inspireGMLChunks(archivePath, GML_MEMBER), archivePath),
+		DECLARED_ENCODING
+	)
+}
+
+export function createESCatastroAdapter(): CorpusAdapter {
+	return {
+		id: ES_CATASTRO_ADAPTER_ID,
+		defaultLicense: ES_CATASTRO_LICENSE,
+		addressRole: AddressRole.Premise,
+		register: SourceRegister.CatastroInspireAddresses,
+		surface: SurfaceOrigin.Rendered,
+		description:
+			"INSPIRE Addresses (Dirección General del Catastro): house-number-level addresses for Spain outside the foral cadastres, one zipped GML per municipality.",
+
+		async *rows(opts: AdapterOptions): AsyncIterable<CanonicalRow> {
+			if (opts.country && !ES_CATASTRO_COUNTRIES.includes(opts.country)) {
+				throw new UnsupportedCountryError(ES_CATASTRO_ADAPTER_ID, ES_CATASTRO_COUNTRIES, opts.country)
+			}
+
+			let emitted = 0
+
+			// One archive per municipality, each a document whose references resolve inside it.
+			for (const archivePath of await archivePaths(opts.inputPath)) {
+				if (opts.signal?.aborted) break
+
+				if (opts.limit !== undefined && emitted >= opts.limit) break
+
+				for await (const row of streamIndexedInspireRows<MarkupElement>({
+					chunks: () => memberChunks(archivePath),
+					addressElement: ADDRESS_ELEMENT,
+					componentElements: COMPONENT_ELEMENTS,
+					// The subtree is kept, because one feature answers a street, a postcode
+					// or a place name depending on which type it is.
+					index: (feature) => feature,
+					compose: composeRow,
+					limit: opts.limit === undefined ? undefined : opts.limit - emitted,
+					signal: opts.signal,
+				})) {
+					yield row
+
+					emitted++
+				}
+			}
+		},
+	}
+}
+
+/**
+ * One address's row, or undefined when it carries too little to render.
+ *
+ * The reference index is complete before this is called, so a reference resolving to no
+ * feature is a value the reader asked for and could not read, and it raises.
+ */
+function composeRow(address: MarkupElement, referenced: ReadonlyMap<string, MarkupElement>): CanonicalRow | undefined {
+	const addressID = address.attributes["gml:id"]?.trim()
+
+	if (voidDesignatorTypes(address).has(HOUSE_NUMBER_TYPE)) {
+		throw new VoidDesignatorError(ES_CATASTRO_ADAPTER_ID, addressID, HOUSE_NUMBER_TYPE)
+	}
+
+	let street: string | undefined
+	let postcode: string | undefined
+	let locality: string | undefined
+	let settlement: string | undefined
+
+	for (const href of componentHrefs(address)) {
+		const key = componentJoinKey(href)
+
+		if (key && referenceStatesNoValue(key)) continue
+
+		const feature = key ? referenced.get(key) : undefined
+
+		if (!feature) throw new UnresolvedComponentReferenceError(ES_CATASTRO_ADAPTER_ID, addressID, href)
+
+		const name = feature.name
+
+		if (inspireNameIs(name, "ad:ThoroughfareName")) {
+			street = thoroughfareName(feature)
+		} else if (inspireNameIs(name, "ad:PostalDescriptor")) {
+			postcode = postalDescriptorCode(feature)
+		} else if (inspireNameIs(name, "ad:AddressAreaName")) {
+			settlement = placeName(feature)
+		} else if (inspireNameIs(name, "ad:AdminUnitName") && MUNICIPALITY_LEVELS.has(adminUnitLevel(feature) ?? "")) {
+			locality = placeName(feature)
+		}
+	}
+
+	return cadastralRow(
+		{
+			street,
+			house: spanishHouseNumber(designatorsByType(address), HOUSE_NUMBER_TYPE),
+			postcode,
+			locality,
+			settlement,
+			addressID,
+		},
+		{ adapterID: ES_CATASTRO_ADAPTER_ID, license: ES_CATASTRO_LICENSE }
+	)
+}
+
+/**
+ * Every archive under `inputPath`, which is one `.zip` or a directory holding them.
+ *
+ * A directory holding no archive raises rather than yielding zero rows, so a wrong
+ * path reports itself instead of reading as an empty publisher.
+ */
+async function archivePaths(inputPath: PathBuilderLike): Promise<readonly PathBuilderLike[]> {
+	const path = PathBuilder.from(inputPath)
+
+	if (path.basename().toLowerCase().endsWith(".zip")) return [path]
+
+	if (path.basename().toLowerCase().endsWith(".gml")) return [path]
+
+	const names = await Globerator.from("*.{zip,gml}", { cwd: path.toString(), absolute: false }).toSorted()
+
+	if (!names.length) {
+		throw new Error(`${ES_CATASTRO_ADAPTER_ID} adapter: ${path.toString()} holds no .zip archive`)
+	}
+
+	return names.map((name) => path(name))
+}
+
+/**
+ * The configured adapter instance registered with the corpus builder.
+ */
+export const esCatastroAdapter = createESCatastroAdapter()

package/lib/es/adapters/gipuzkoa/adapter.ts

@@ -0,0 +1,263 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ * `es-gipuzkoa`: the INSPIRE Addresses theme the Diputación Foral de Gipuzkoa publishes, read from
+ * the one zipped GML that carries the whole province.
+ *
+ * The ATOM service at `https://b5m.gipuzkoa.eus/inspire/download/addresses.xml` is 4,469 bytes and
+ * holds a single entry for the province rather than one per municipality. That entry links its
+ * dataset through `rel="alternate"` and not `rel="enclosure"`, so a fetcher that looks only for an
+ * enclosure finds no dataset here. The adapter is handed the archive and the feed is acquisition's
+ * concern. The archive `ES.GFA.AD.zip` is 7,959,103 bytes and its one member `ES.GFA.AD.gml` is
+ * 313,162,472 bytes, a ratio of 39, so the member is read as a chunk sequence and never buffered.
+ *
+ * The member is self-contained and its order decides the reader. Counted over all 313,162,472 bytes:
+ * 68,744 `ad:Address`, 4,250 `ad:ThoroughfareName`, 178 `ad:PostalDescriptor` and 88
+ * `ad:AdminUnitName`, which is every one of the 73,260 `gml:featureMember` elements. The 4,516
+ * reference features are written last, beginning at byte 306,440,816. Holding an address until the
+ * document ends would therefore hold all 68,744, so the adapter reads the member twice through
+ * {@linkcode streamIndexedInspireRows} and indexes 4,516 features instead.
+ *
+ * A reference is a WFS stored-query URL whose `ID` parameter equals the target's `gml:id`, written
+ * against the internal host `http://b5mdev/` and with the id repeated as the URL's fragment:
+ * `…?…&ID=ES.GFA.TN.056.1110#ES.GFA.TN.056.1110`. `componentJoinKey` reads the parameter and the
+ * fragment belongs to the URL rather than to the id, so the key is `ES.GFA.TN.056.1110`. The
+ * unreachable host is not this reader's problem, because every reference resolves inside the member.
+ *
+ * A postal-descriptor reference whose final segment is empty, `…&ID=ES.GFA.PD.059. #…`, states that
+ * the publisher holds no postcode, and `referenceStatesNoValue` separates it from a broken join.
+ * The Cadastre writes the same form, which is why the predicate has one home.
+ *
+ * The locator vocabulary is one type. 68,743 of 68,744 addresses write exactly one designator, typed
+ * `LocatorDesignatorTypeValue/addressIdentifierGeneral`; one address writes none. The municipality is
+ * the only `ad:AdminUnitName` level published, `4thOrder`, and the province publishes 88 of them for
+ * its 88 municipalities.
+ *
+ * The publisher writes its street names in Basque, with the street type as a trailing word rather
+ * than a leading abbreviation: `Samikolla ibilbidea`. The name is taken as written.
+ *
+ * `ad:LocatorName` is populated on 46,365 of 68,744 addresses and holds a building name as a Basque
+ * and Spanish pair, `Gasolinera` with `Gasolindegia` and `Agirre, baserria` with `Agirre, caserío`.
+ * The adapter writes none of them. The address-source register's personal-data review reads `absent`
+ * on the ground that these name the building rather than its occupant, and it records that a Basque
+ * *baserri* name is commonly also a surname. Keeping them out of the corpus keeps that reading from
+ * resting on the stronger claim that the name is safe to publish as address text.
+ *
+ * No publisher-side rendering of a whole address was found for this source. `ad:alternativeIdentifier`
+ * is populated on all 68,744 addresses and holds a serial number such as `36798` rather than an
+ * address. What the file does state twice is the join: the address `gml:id`
+ * `ES.GFA.AD.056_1110_003` concatenates the municipality code `056`, the thoroughfare code `1110`
+ * and the number `003`, and the references it carries are `AU_ADMINISTRATIVEUNIT_34162020056`,
+ * `ES.GFA.TN.056.1110` and `ES.GFA.PD.056.20830`. The adapter's test checks the composed street and
+ * number against that identifier, which is a check that the right features were joined rather than a
+ * check of the address the publisher would print.
+ */
+
+import type { MarkupElement } from "@mailwoman/core/html/elements"
+import { PathBuilder, type PathBuilderLike } from "path-ts"
+import { Globerator } from "spliterator/node/fs"
+
+import { UnsupportedCountryError } from "#adapters/errors"
+import { cadastralRow, referenceStatesNoValue, spanishHouseNumber } from "#es/cadastre"
+import {
+	adminUnitLevel,
+	componentHrefs,
+	componentJoinKey,
+	designatorsByType,
+	inspireNameIs,
+	placeName,
+	postalDescriptorCode,
+	thoroughfareName,
+	voidDesignatorTypes,
+} from "#inspire/address"
+import { inspireGMLChunks } from "#inspire/archive"
+import { UnresolvedComponentReferenceError, VoidDesignatorError } from "#inspire/errors"
+import { streamIndexedInspireRows } from "#inspire/stream"
+import { SourceRegister } from "#registers"
+import { AddressRole, type AdapterOptions, type CanonicalRow, type CorpusAdapter, SurfaceOrigin } from "#types"
+
+/**
+ * Registry id for this adapter.
+ *
+ * Stamped into every row it emits, so a corpus record can be traced back to the dataset it came from.
+ */
+export const ES_GIPUZKOA_ADAPTER_ID = "es-gipuzkoa"
+
+/**
+ * The jurisdiction this adapter emits.
+ *
+ * Coverage is the province of Gipuzkoa, which the feed states as the dataset's whole extent.
+ */
+export const ES_GIPUZKOA_COUNTRIES: readonly string[] = ["ES"]
+
+/**
+ * The license the address-source register elected for this publisher.
+ *
+ * Gipuzkoa states four things, and the register elects the one addressing this dataset:
+ * the `otherConstraints` of its ISO 19139 dataset record, `CC BY-SA 4.0`.
+ * Its feed states attribution alone and its website states CC BY 4.0, and the register
+ * records why the stricter reading is the one to err towards: the share-alike filter reads
+ * this label, so recording CC BY 4.0 would hide an obligation the publisher may hold.
+ *
+ * The value is the SPDX identifier the register elected rather than the license's title,
+ * because `licenseVerdict` resolves an identifier and reads a title as unrecognized.
+ * Under `LicensePolicy.ShareAlikeFree` the title returns `{refusal: null, mentionsShareAlike: false}`
+ * and `CC-BY-SA-4.0` returns `{refusal: "share-alike-carried"}`, so the title admits
+ * these rows to a corpus assembled to hold no share-alike source.
+ */
+export const ES_GIPUZKOA_LICENSE = "CC-BY-SA-4.0"
+
+/**
+ * The member of the archive that holds the GML.
+ *
+ * The archive ships one member, so the extension selects it without the adapter naming it.
+ */
+const GML_MEMBER = /\.gml$/iu
+
+/**
+ * The feature types an address references, indexed in the first pass over the member.
+ *
+ * `ad:AddressAreaName` is listed although the province publishes none, because the schema admits one
+ * and a future edition that wrote one would otherwise read as an unresolved reference.
+ */
+const COMPONENT_ELEMENTS = [
+	"ad:ThoroughfareName",
+	"ad:PostalDescriptor",
+	"ad:AdminUnitName",
+	"ad:AddressAreaName",
+] as const
+
+/**
+ * The administrative level that is the municipality.
+ *
+ * The only level the province publishes, on 88 of 88 `ad:AdminUnitName` features.
+ */
+const MUNICIPALITY_LEVEL = "4thOrder"
+
+/**
+ * The designator type the publisher writes.
+ */
+const HOUSE_NUMBER_TYPE = "addressIdentifierGeneral"
+
+export function createESGipuzkoaAdapter(): CorpusAdapter {
+	return {
+		id: ES_GIPUZKOA_ADAPTER_ID,
+		defaultLicense: ES_GIPUZKOA_LICENSE,
+		addressRole: AddressRole.Premise,
+		register: SourceRegister.GipuzkoaInspireAddresses,
+		surface: SurfaceOrigin.Rendered,
+		description:
+			"INSPIRE Addresses (Diputación Foral de Gipuzkoa): house-number-level addresses for the province, as one zipped GML.",
+
+		async *rows(opts: AdapterOptions): AsyncIterable<CanonicalRow> {
+			if (opts.country && !ES_GIPUZKOA_COUNTRIES.includes(opts.country)) {
+				throw new UnsupportedCountryError(ES_GIPUZKOA_ADAPTER_ID, ES_GIPUZKOA_COUNTRIES, opts.country)
+			}
+
+			let emitted = 0
+
+			for (const archivePath of await archivePaths(opts.inputPath)) {
+				if (opts.signal?.aborted) break
+
+				if (opts.limit !== undefined && emitted >= opts.limit) break
+
+				for await (const row of streamIndexedInspireRows<MarkupElement>({
+					chunks: () => inspireGMLChunks(archivePath, GML_MEMBER),
+					componentElements: COMPONENT_ELEMENTS,
+					index: (feature) => feature,
+					compose: composeRow,
+					limit: opts.limit === undefined ? undefined : opts.limit - emitted,
+					signal: opts.signal,
+				})) {
+					yield row
+
+					emitted++
+				}
+			}
+		},
+	}
+}
+
+/**
+ * One address's row, or undefined when it carries too little to render.
+ *
+ * The reference index is complete before this is called, so a reference resolving to no
+ * feature is a value the reader asked for and could not read, and it raises.
+ */
+function composeRow(address: MarkupElement, referenced: ReadonlyMap<string, MarkupElement>): CanonicalRow | undefined {
+	const addressID = address.attributes["gml:id"]?.trim()
+
+	if (voidDesignatorTypes(address).has(HOUSE_NUMBER_TYPE)) {
+		throw new VoidDesignatorError(ES_GIPUZKOA_ADAPTER_ID, addressID, HOUSE_NUMBER_TYPE)
+	}
+
+	let street: string | undefined
+	let postcode: string | undefined
+	let locality: string | undefined
+	let settlement: string | undefined
+
+	for (const href of componentHrefs(address)) {
+		const key = componentJoinKey(href)
+
+		if (key && referenceStatesNoValue(key)) continue
+
+		const feature = key ? referenced.get(key) : undefined
+
+		if (!feature) throw new UnresolvedComponentReferenceError(ES_GIPUZKOA_ADAPTER_ID, addressID, href)
+
+		const name = feature.name
+
+		if (inspireNameIs(name, "ad:ThoroughfareName")) {
+			street = thoroughfareName(feature)
+		} else if (inspireNameIs(name, "ad:PostalDescriptor")) {
+			postcode = postalDescriptorCode(feature)
+		} else if (inspireNameIs(name, "ad:AddressAreaName")) {
+			settlement = placeName(feature)
+		} else if (inspireNameIs(name, "ad:AdminUnitName") && adminUnitLevel(feature) === MUNICIPALITY_LEVEL) {
+			locality = placeName(feature)
+		}
+	}
+
+	return cadastralRow(
+		{
+			street,
+			house: spanishHouseNumber(designatorsByType(address), HOUSE_NUMBER_TYPE),
+			postcode,
+			locality,
+			settlement,
+			addressID,
+		},
+		{ adapterID: ES_GIPUZKOA_ADAPTER_ID, license: ES_GIPUZKOA_LICENSE }
+	)
+}
+
+/**
+ * Every archive under `inputPath`, which is one `.zip`, one unpacked `.gml`, or a directory holding either.
+ *
+ * The publisher ships one archive, and the unpacked member is accepted because an
+ * operator who inflated 313,162,472 bytes once should not have to inflate them again,
+ * and because a committed fixture is reviewable only as text.
+ * A directory holding neither raises rather than yielding zero rows, so a wrong path
+ * reports itself instead of reading as an empty publisher.
+ */
+async function archivePaths(inputPath: PathBuilderLike): Promise<readonly PathBuilderLike[]> {
+	const path = PathBuilder.from(inputPath)
+	const basename = path.basename().toLowerCase()
+
+	if (basename.endsWith(".zip") || basename.endsWith(".gml")) return [path]
+
+	const names = await Globerator.from("*.{zip,gml}", { cwd: path.toString(), absolute: false }).toSorted()
+
+	if (!names.length) {
+		throw new Error(`${ES_GIPUZKOA_ADAPTER_ID} adapter: ${path.toString()} holds no .zip archive or .gml member`)
+	}
+
+	return names.map((name) => path(name))
+}
+
+/**
+ * The configured adapter instance registered with the corpus builder.
+ */
+export const esGipuzkoaAdapter = createESGipuzkoaAdapter()