xslt/ holds an original set of XSLT 1.0 crosswalks for MARCXML (slim),
written for this project from the public Library of Congress mapping
specifications (see xslt/DISTINCTNESS.md for the originality check).
Every stylesheet accepts either a marc:collection or a bare
marc:record document element, and works with plain xsltproc.
| stylesheet | direction | output |
|---|---|---|
lib/marc-utils.xsl |
shared helpers | subfield joining, punctuation trimming, leader/008 slicing, $6 linkage parsing, JSON escaping, ISBN-13 normalization, year extraction |
marc21-to-unimarc-skeleton.xsl |
MARC21 → UNIMARC-shaped MARCXML | documented subset: label, 001, 100/101, 200, 210, 215, 606 (subdivision re-lettering), 700/701 |
marcxml-to-dcterms.xsl |
MARCXML → qualified Dublin Core (DCMI Terms) | qdcRecord of dcterms:* (issued, extent, medium, spatial, temporal, isPartOf, tableOfContents, alternative, …) |
marcxml-to-ead.xsl |
MARCXML → EAD3 | collection-level <ead> skeleton: control from 001/003/008, archdesc did + scopecontent + controlaccess |
marcxml-to-html.xsl |
MARCXML → HTML | one card per record: 245 heading + tag/ind/subfield table |
marcxml-to-mods.xsl |
MARCXML → MODS 3.7 | mods / modsCollection |
marcxml-to-oai_dc.xsl |
MARCXML → simple Dublin Core | oai_dc:dc (collections get a neutral dcRecords wrapper) |
marcxml-to-rdfdc.xsl |
MARCXML → RDF/XML with DCMI Terms | rdf:RDF of rdf:Description |
marcxml-to-ris.xsl |
MARCXML → RIS | text citations: TY from leader, AU/A2/TI/T2/CY/PB/PY/SN/UR/KW/N1, ER terminator |
marcxml-to-schemaorg.xsl |
MARCXML → schema.org JSON-LD | NDJSON text, one object per record; same key set and @type table as the marc_jsonld SQL macro |
mods-to-marcxml.xsl |
MODS 3.x → MARCXML | marc:collection with synthesized leader + 008 |
oai_dc-to-marcxml.xsl |
simple Dublin Core (oai_dc) → MARCXML | marc:collection; leader/008 synthesized, dc:identifier discriminated to 020/022/856/024 |
One-liners:
xsltproc xslt/marcxml-to-mods.xsl records.xml > records.mods.xml
xsltproc xslt/marcxml-to-oai_dc.xsl records.xml > records.dc.xml
xsltproc xslt/marcxml-to-rdfdc.xsl records.xml > records.rdf
xsltproc xslt/marcxml-to-html.xsl records.xml > records.html
xsltproc xslt/mods-to-marcxml.xsl records.mods.xml > roundtrip.xmlThe natural pipeline with this extension: use COPY (FORMAT marcxml) to
get slim MARCXML out of any supported source, then transform:
-- in the duckdb shell, marc21 extension loaded
COPY (SELECT leader, fields FROM read_marc('input.mrc'))
TO 'records.xml' (FORMAT marcxml);xsltproc xslt/marcxml-to-mods.xsl records.xml > records.mods.xmlBecause read_marcxml() reads slim MARCXML back in, mods-to-marcxml.xsl
also closes the loop for MODS sources:
xsltproc xslt/mods-to-marcxml.xsl records.mods.xml > back.xmlSELECT * FROM read_marcxml('back.xml');Tests: sh test/xslt/run.sh (uses test/data/sample.xml and the
fixtures in test/data/xslt/; requires xsltproc and xmllint).
The stylesheets ship inside the extension. xslt/registry.tsv is the
catalog — one row per transform: alias, repo-relative path, source format,
target format, one-line description — and the build reads it, every
stylesheet it names, and the shared helper library into the binary. An
installed extension therefore resolves a crosswalk with no files on disk,
in a plain LOAD marc21 session and in WASM alike.
Three functions expose it:
| function | returns |
|---|---|
marc_xslt_functions() |
the catalog: alias, path, source_format, target_format, description |
marc_xslt_library() |
the shared lib/marc-utils.xsl text every sheet includes |
marc_xslt_stylesheet(alias_or_path) |
the stylesheet text (NULL when the key is not registered) |
Pick a crosswalk the way a cataloger would — by what it converts:
SELECT alias, target_format, description
FROM marc_xslt_functions()
WHERE source_format = 'MARC'
ORDER BY alias;| alias | stylesheet | source → target |
|---|---|---|
BIBFRAME2=>MARC |
xslt/bibframe2-to-marcxml.xsl |
BIBFRAME2 → MARC |
MARC=>BIBFRAME2 |
xslt/marcxml-to-bibframe2.xsl |
MARC → BIBFRAME2 |
MARC=>DCTERMS |
xslt/marcxml-to-dcterms.xsl |
MARC → DCTERMS |
MARC=>EAD3 |
xslt/marcxml-to-ead.xsl |
MARC → EAD3 |
MARC=>EDM |
xslt/marcxml-to-edm.xsl |
MARC → EDM |
MARC=>HTML |
xslt/marcxml-to-html.xsl |
MARC → HTML |
MARC=>MADS |
xslt/marcxml-to-mads.xsl |
MARC-AUTHORITY → MADS |
MARC=>MODS |
xslt/marcxml-to-mods.xsl |
MARC → MODS |
MARC=>OAI_DC |
xslt/marcxml-to-oai_dc.xsl |
MARC → OAI_DC |
MARC=>ONIX3 |
xslt/marcxml-to-onix3.xsl |
MARC → ONIX3 |
MARC=>RDFDC |
xslt/marcxml-to-rdfdc.xsl |
MARC → RDF-DC |
MARC=>RIS |
xslt/marcxml-to-ris.xsl |
MARC → RIS |
MARC=>SCHEMA.ORG |
xslt/marcxml-to-schemaorg.xsl |
MARC → JSON-LD |
MARC=>UNIMARC |
xslt/marc21-to-unimarc.xsl |
MARC → UNIMARC |
MARC=>UNIMARC-SKELETON |
xslt/marc21-to-unimarc-skeleton.xsl |
MARC → UNIMARC |
MODS=>MARC |
xslt/mods-to-marcxml.xsl |
MODS → MARC |
OAI_DC=>MARC |
xslt/oai_dc-to-marcxml.xsl |
OAI_DC → MARC |
UNIMARC=>MARC |
xslt/unimarc-to-marc21.xsl |
UNIMARC → MARC |
An alias is matched case-insensitively; the registry path works as a key
too, so marc_xslt_stylesheet('xslt/marcxml-to-mods.xsl') and
marc_xslt_stylesheet('marc=>mods') return the same text. An unregistered
key returns NULL.
The extension hands out stylesheet text; writing it to a file is a plain
COPY. A stylesheet is full of newlines and quotation marks, so the CSV
writer must be told to do nothing to it — QUOTE '' and ESCAPE '' turn
off quoting and escaping, and HEADER false keeps the column name out of
the file. The writer still terminates the row with a newline, so trim the
text's own final newline and let the row terminator put exactly one back:
the file is then byte-for-byte the stylesheet.
COPY (SELECT rtrim(marc_xslt_stylesheet('MARC=>MODS'), chr(10)))
TO 'work/marcxml-to-mods.xsl' (FORMAT csv, HEADER false, QUOTE '', ESCAPE '');
COPY (SELECT rtrim(marc_xslt_library(), chr(10)))
TO 'work/lib/marc-utils.xsl' (FORMAT csv, HEADER false, QUOTE '', ESCAPE '');Every stylesheet includes the helpers as lib/marc-utils.xsl, so export the
library once into a lib/ subdirectory beside the sheets and any number of
crosswalks can run from that directory. Verify a copy at any time:
SELECT content = marc_xslt_stylesheet('MARC=>MODS')
FROM read_text('work/marcxml-to-mods.xsl');This extension does not execute XSLT — it registers, catalogs and hands out
the stylesheets, and the transform itself runs in xsltproc, Saxon, or a
browser (the composition rule in ECOSYSTEM.md).
marc_xslt_command() prints the command line for a registered crosswalk,
with all three paths quoted for a POSIX shell:
SELECT marc_xslt_command('MARC=>MODS', 'records.xml', 'records.mods.xml');
-- xsltproc --output 'records.mods.xml' 'marcxml-to-mods.xsl' 'records.xml'The sheet is named by its bare filename — marc_xslt_sheet_file(alias) —
because the command is meant to run in the directory holding the exported
sheet and its lib/. An unknown alias yields NULL rather than a command
that would fail at the shell. A whole batch is one query:
SELECT marc_xslt_command(alias, 'records.xml', 'records.' || lower(alias) || '.out')
FROM marc_xslt_functions() WHERE source_format = 'MARC';End to end, from records to MODS:
COPY (SELECT leader, fields FROM read_marc('input.mrc')) TO 'work/records.xml' (FORMAT marcxml);cd work && xsltproc --output 'records.mods.xml' 'marcxml-to-mods.xsl' 'records.xml'The shipped catalog is read-only; local stylesheets belong in a table beside
it. A macro cannot do this job: scalar macros bind when the extension loads,
and a body that named marc_xslt_user would fail to bind for every database
that has not created that table. So the union is a two-statement recipe, run
once per database, and it stays yours:
CREATE TABLE marc_xslt_user (
alias VARCHAR PRIMARY KEY,
path VARCHAR, -- where the sheet lives on disk
source_format VARCHAR,
target_format VARCHAR,
description VARCHAR
);
INSERT INTO marc_xslt_user VALUES
('MARC=>ALMA-LOAD', '/srv/xslt/marcxml-to-alma.xsl', 'MARC', 'ALMA',
'Local vendor load profile: 9XX stripping and 949 stamping.');
CREATE VIEW marc_xslt_all AS
SELECT alias, path, source_format, target_format, description, 'shipped' AS origin
FROM marc_xslt_functions()
UNION ALL
SELECT alias, path, source_format, target_format, description, 'local' AS origin
FROM marc_xslt_user;marc_xslt_all is then the one list a cataloger picks from, and
marc_xslt_stylesheet() still answers for the shipped half while the local
half is already a path on disk. Keep local aliases distinct from the shipped
ones (a WHERE alias NOT IN (SELECT alias FROM marc_xslt_functions()) check
on insert is enough) so a pick is never ambiguous.
Because the catalog and the stylesheet text live in the extension binary,
a browser app that loads marc21 in duckdb-wasm gets the identical list from
the identical query, with no files to fetch: marc_xslt_functions() fills
the crosswalk picker, marc_xslt_stylesheet() supplies the sheet, and the
browser's own XSLTProcessor runs it client-side. xsl:include is
resolved against the document the browser parsed, so the app either serves
marc_xslt_library() at the lib/marc-utils.xsl the sheet asks for, or
splices that text into the sheet in place of the include before importing
it.
test/sql/marc_xslt.test holds the registry to its promises: the catalog
against xslt/registry.tsv row for row, every stylesheet in the tree
registered, every alias resolving to a document, the byte-exact export and
the command shape.
| MARC | MODS |
|---|---|
| 001/003, 008/00-05, 040 | recordInfo |
| 008/35-37, 041$a | language/languageTerm (deduplicated against 008) |
| 010/020/022/024/035 | identifier typed; $z marked invalid="yes" |
| 050/082/084 | classification (lcc/ddc/$2) |
| 100/110/111, 700/710/711 | name (type + usage="primary" for 1XX), namePart, date namePart, roles from $e/$4 |
| 130/240 | titleInfo type="uniform" |
| 245 (ind2 nonfiling) | titleInfo with nonSort/title/subTitle/partNumber/partName; $c → note type="statement of responsibility" |
| 246 | titleInfo type="alternative" (displayLabel from $i) |
| 260, 264 (ind2 1 publication, ind2 4 copyright), 250, 310, 008/06-14, Leader/07 | originInfo: place/publisher/dateIssued, dateIssued encoding="marc" (+start/end points for ranged 008 date types), copyrightDate, edition, frequency, issuance |
| 300, 856$q | physicalDescription: extent, internetMediaType |
| 440/490/800/810/811/830 | relatedItem type="series" |
| 520 / 505 / 521 | abstract / tableOfContents / targetAudience |
| 600/610/611/630/648/650/651 | structured subject (name/titleInfo/topic/temporal/geographic heading + $x$y$z$v subdivisions in order; authority from ind2 or $2) |
| 655 | genre with authority |
| 760–787 | relatedItem typed (773 host, 776 otherFormat, 780 preceding, 785 succeeding, …) with title, $x$z$w identifiers, $g part |
| 856$u, 852 | location/url (displayLabel, usage="primary display" for ind2 0), physicalLocation |
| 880 linked to 245 | paired titleInfo sharing altRepGroup; other 880s → note type="alternate-script" |
| Leader/06-07 | typeOfResource (+collection="yes") |
| other 5XX | note (typed for 504/511/518/546/561) |
245/246 → dc:title; 1XX → dc:creator; 7XX → dc:contributor;
600–653 → dc:subject (subdivisions joined with --); 500/505/520 →
dc:description; 260/264$b → dc:publisher; imprint $c else
008/07-10 → dc:date; Leader/06-07 (DCMI vocabulary) + 655 → dc:type;
300$a + 856$q → dc:format; 001, 010/020/022/024, 856$u →
dc:identifier; 008/041 → dc:language; series + 760–787$t →
dc:relation; 651$a/650$z/752 → dc:coverage; 506/540 → dc:rights.
Same source fields, refined targets: dcterms:title/alternative
(includes 880-linked 245), creator/contributor, subject plus
spatial (651$a, 650$z) and temporal (648$a, 650$y), abstract,
tableOfContents, description, publisher, issued, extent,
type as an rdf:resource pointing into the DCMI type vocabulary,
ISBN/ISSN as urn: identifiers, isPartOf (440/490/830/773),
replaces/isReplacedBy (780/785), accessRights/rights. First
856$u becomes rdf:about.
Reverses the core of the forward mapping: leader (type/issuance) and
40-character 008 (creation date, date1, language) are synthesized;
001/003 from recordIdentifier; 010/020/022 (invalid → $z), 050/082/084,
100/110/111/700/710/711 with $e/$4 roles, 240, 245 (ind2 from
nonSort length, : before subtitle, $c from the statement-of-
responsibility note), 246, 250, 264 (ind2 1; copyrightDate → ind2 4),
041 for multilingual records, 300, 310, 490, 5XX notes (typed notes back
to 504/511/546/561), 505, 520, 521, 600/610/611/630/648/650/651 with
subdivision codes and $2, 655, 773 ($x$z$w$g), 856 (usage → ind2 0).
- Punctuation: trailing ISBD punctuation is trimmed with a simple
recursive rule (strip
,;:/=runs and one final period unless it ends an ellipsis or single-letter initial). The LC chop rules have more special cases (abbreviations like "etc."); those survive here with the period trimmed or kept per the simple rule. The reverse direction regenerates only the:before a 245 subtitle, not full ISBD. - 880 handling is title-centric: only 880s linked to 245 become
paired
titleInfoelements with a sharedaltRepGroup; other 880s are preserved asnote type="alternate-script"rather than duplicatedname/originInfo/etc. withscriptattributes. The reverse transform does not reconstruct 880s. - Fixed-field breadth: 008 material-specific positions (audience,
form of item, literary form, government publication…) and 006/007 are
not mapped to
targetAudience/form/genre; only dates and language are read. The synthesized 008 on the way back fills places 15-17 withxxand leaves material positions blank. - Subjects:
$e/$4relators inside 6XX,$g/$3and linkage subfields are ignored; 630 keeps only$a$d$f$p; 653 is mapped in the DC outputs but not in MODS (no LC-defined structured home). - Names: 100
$bnumeration is folded into the singlenamePartrather than split;affiliation,description,etalare not produced. The reverse maps every non-primary name to 700/710/711 (no 720s) and cannot restore$qversus$adistinctions. - originInfo: only the first 260/264-ind2-1 field is mapped in full;
production/distribution/manufacture statements (264 ind2 0/2/3) are
not emitted (the copyright 264 ind2 4 is).
placecarries only the text form, nocodeplaceTerm from 008/15-17. - relatedItem: linking entries carry title, ISSN/ISBN/
$wcontrol numbers and$gparts only — no full name/originInfo sub-records; 490 loses$vnumbering on the way back (it returns as part of the title string when present in the MODS title). - Round-trip fidelity is asserted for the core fields exercised in
test/xslt/run.sh(titles, names, imprints, subjects, identifiers, series/host links, notes); a full MARC → MODS → MARC pass is lossy by design — MODS itself does not keep every MARC distinction. - DC outputs:
dcRecords(oai_dc collection wrapper) is a convenience element in no namespace; single records are schema-validoai_dc:dc. In the RDF output, literals are untyped and agents are plain labels, not structuredfoaf:Agentnodes.
Every stylesheet lives directly in xslt/, including the shared
marc-utils.xsl each one includes — a flat directory, so exporting a sheet
and its library means copying two files into one place. Two groups have
their own documentation and their own runner under test/xslt/: BIBFRAME
2.x in both directions (BIBFRAME.md), and the international
set — field-level UNIMARC both ways, the national formats, ONIX 3,
Europeana EDM, MADS (INTERNATIONAL.md).