Pelican plugin to embed data tables in Markdown articles via a {% table %} shortcode.

uv add pelican-tabular

Add to pelicanconf.py:

PLUGINS = ["pelican.plugins.tabular"]

{% table data/books.yaml %}
{% table data/books.yaml fields="title,rating,year" %}
{% table data/books.yaml sort_by="rating" sort_order="desc" %}
{% table data/books.yaml hidden="internal_id" %}
{% table data/books.yaml group_by="genre" group_summary_at="genre" %}
{% table data/books.yaml group_by="author,year" aggregate="year:year" field_labels="year:Publication Year" %}
{% table data/books.yaml sort_by="date" date_format="%b %Y" %}
{% table data/books.yaml group_by="date:year" group_summary_at="date:year" %}
{% table data/books.yaml aria_columns="slide,recording" %}
{% table data/concerts.yaml fields="title,venue" %}       {# venue_ref → plain text #}
{% table data/concerts.yaml fields="title,venue:link" %}  {# venue_ref → link #}
{% table data/concerts.yaml group_by="venue.city" %}      {# group by a field of the referenced record #}

Supports YAML (list of dicts), JSON arrays, and CSV.

# data/books.yaml
- title: The Left Hand of Darkness
  author: Ursula K. Le Guin
  year: 1969
  rating: 5
- title: Piranesi
  author: Susanna Clarke
  year: 2020
  rating: 5

Link dicts also accept url as an alias for href, and label as an alias for text.

Link href values support Pelican's {filename} syntax to cross-reference other articles:

- title: My Post
  link: {href: "{filename}posts/my-post.md", text: "Read more"}

A field named <name>_ref lets a row point at a single record living in another YAML file, instead of duplicating that record's data (and letting the two copies drift apart). This is meant for the common case of a data table repeatedly citing the same handful of places/people/things that already have their own authoritative file elsewhere in content/ — e.g. a concert log citing venues that are also rendered on a map by pelican-osm.

Resolving <name>_ref replaces it with a <name> field holding the whole referenced record — every field the target has, not just a display string. A ref models a relationship between two rows of data; what that relationship looks like on the page (plain text, a link, grouped by one of the referenced record's own fields, …) is up to fields/group_by, not baked into the ref itself.

# content/places/venues/taiwan.yaml (pelican-osm locations-based shape)
locations:
  - id: zepp-new-taipei
    name: Zepp New Taipei
    city: 新北市
    lat: 25.059661
    lon: 121.449499

# content/data/concerts.yaml
- title: {text: "藥師寺寬邦「悟」", href: "https://example.com/goo"}
  date: 2024-10-25
  venue_ref: zepp-new-taipei

{% table data/concerts.yaml fields="title,date,venue" %}

renders a venue column (the _ref suffix is stripped) whose cell is the referenced record's name — Zepp New Taipei — as plain text. Want a link instead? Add the :link transform: fields="title,date,venue:link" (see Field paths and transforms). Want the venue's city? fields="title,venue.city", or group_by="venue.city" to group concerts by which city they were in — fields, group_by, sort_by, etc. all see the entire resolved venue record, not just venue_ref.

• Global id (venue_ref: zepp-new-taipei, shown above) — searches every YAML file under TABULAR_REF_ROOTS (default ["places"], relative to Pelican's PATH) for a record whose id (or, failing that, name) matches. This is the form to reach for by default: it doesn't encode where the target record currently lives, so moving/renaming files under TABULAR_REF_ROOTS never requires touching every row that cites them.
• <path>#<id> (venue_ref: places/venues/taiwan.yaml#zepp-new-taipei) — <path> is resolved relative to Pelican's content root (the PATH setting), not TABULAR_DATA_ROOT (which is where the referencing file was loaded from — these are often different roots, e.g. TABULAR_DATA_ROOT pointed at content/data while venues live under content/places/). Use this form to disambiguate when a global id would match more than one record (see Fail loud below), or to point at a file outside TABULAR_REF_ROOTS entirely.

Both forms prefer an id match and fall back to name only when no ID matches. Global lookup applies that preference across all indexed files; <path>#<id> applies it within the target file. OSM's own shortcodes accept either an ID or a name, but return all matches rather than prioritizing IDs. Use unique IDs to avoid ambiguity.

The target file may use locations: [...] (shown above) or a bare top-level YAML list of dicts.

Requires tabular 0.8.1 or later: ID-keyed mappings, inheritance of file defaults and exclusion of underscore-prefixed files from global discovery. These fixes are included in this source checkout; see the release status before upgrading from PyPI. For example:

defaults:
  country: 日本
toho-kawasaki:
  name: TOHOシネマズ 川崎
  city: 神奈川
  lat: 35.5313838
  lon: 139.7002316

Here toho-kawasaki becomes the record's id unless it has an explicit id. File defaults follow OSM's rules: top-level fields alongside locations, a defaults mapping alongside keyed records, or a standalone defaults entry applying to subsequent records in a bare list. Record values override defaults. When defaults contain a non-empty tags list and the record's tags are also a list, they are combined without duplicates, defaults first. Missing or falsy tags (including [], null, "", false and 0) retain the default tags. Non-empty, non-list tags use normal override behavior; without a non-empty default tag list, the record's tags remain unchanged. Nested items remain part of the referenced record.

Global discovery skips underscore-prefixed YAML files such as _schema.yaml and _common.yml, matching OSM's metadata-file convention. An explicit <path>#<id> reference can still load such a file. Malformed data files continue to produce warnings; unresolved references still fail the build.

A ref that doesn't resolve — target file not found, id not found, or a global id that matches more than one record, even within one file under TABULAR_REF_ROOTS — fails the build by default (TABULAR_REF_STRICT = True). A ref pointing at nothing is data corruption, not something worth burying as a log.warning line in a 250-article build log. A path-qualified reference can disambiguate records in different files; duplicate IDs within one file should be corrected in the data.

This happens once, at the end of the build, not per-page. Pelican wraps each page's processing in a try/except that logs an ERROR and skips that one page on any exception, continuing the rest of the build with exit code 0 — raising immediately when a bad ref is found would only delete that page from the site, silently, while everything else (including the CI step) reports success. So instead: every <field>_ref failure encountered while processing any page is collected (with full locator info) as pages are generated, and once the whole build reaches Pelican's finalized stage, a single exception is raised listing every failure found, across every page, in one shot — not just the first one, so fixing a data file with several bad refs doesn't take several build-fix-rebuild cycles:

pelican-tabular: 2 ref error(s); failing the build:
  - pelican-tabular: global ref id 'moondog' is ambiguous: matches records in
2 file(s): ['content/places/venues/japan.yaml',
'content/places/venues/taiwan.yaml']; use the '<path>#<id>' form to
disambiguate (field 'venue_ref', row 3 of content/data/concerts.yaml)
  - pelican-tabular: ref id 'nangang-exhibiton-hall-1' not found under
TABULAR_REF_ROOTS ['places'] (searched relative to content) (field
'venue_ref', row 7 of content/data/concerts.yaml)

Because this raises at the very end, the build's output directory may still contain pages generated before the failure was surfaced (with the offending cell showing the raw, unresolved ref string) — the exit code is what a CI/deploy step should gate on, not the presence of output/.

Set TABULAR_REF_STRICT = False (globally only — there's no per-shortcode override) to degrade every failure to a log.warning instead, leaving the raw ref string in the field (rendered as plain text) and never failing the build, for setups that would rather not fail outright on one bad row.

Each target file is read at most once per build, and the global-id index (built by scanning TABULAR_REF_ROOTS) is built at most once regardless of how many <field>_ref values use the global-id form.

By default, the plain-text rendering and the :link transform's link text both use the target record's name field, and :link's href is built from lat/lon in OpenStreetMap's URL format — this matches what real place data already looks like, so the common case (a table citing pelican-osm place records) needs no configuration. Override per-shortcode with ref_text_field and ref_href_template, or globally with TABULAR_REF_TEXT_FIELD / TABULAR_REF_HREF_TEMPLATE:

{% table data/concerts.yaml fields="title,venue:link" ref_text_field="label" ref_href_template="https://maps.example/{lat},{lon}" %}

ref_href_template is a str.format template evaluated against the resolved record's fields, used only by the :link transform. Both settings apply to every ref-resolved field in a given shortcode call — this plugin does not support per-column overrides in a single invocation. If a table cites two different kinds of referenced records that need different text/href rules, split it into two {% table %} calls (one per data file) rather than mixing ref types in one row shape.

Real place data is rarely uniform. Some records may have a stable identifier (e.g. an OSM node id, once you've looked one up by hand) while older or less-curated records only have raw coordinates. ref_href_template accepts multiple templates separated by |, tried in order — the first template whose placeholders are all present and non-empty in the resolved record wins:

{% table data/concerts.yaml fields="title,venue:link" ref_href_template="https://www.openstreetmap.org/{osm_type}/{osm_id}|https://www.openstreetmap.org/?mlat={lat}&mlon={lon}#map=17/{lat}/{lon}" %}

Here, a venue record with osm_type/osm_id set links straight to that OSM element; a venue with only lat/lon falls back to the coordinate-based map link; a template is only considered applicable when every placeholder it references resolves to a real, non-empty value — a template is never partially filled with blanks. If a record satisfies none of the templates in the chain, the cell falls back further still: no link is rendered at all, just the plain text (never a malformed href like https://www.openstreetmap.org//). | was chosen as the separator because it essentially never appears literally inside a URL template (browsers and YAML both treat it as unremarkable text, but it's not a URL-safe character you'd expect in a real template — a template that genuinely needs a literal | would percent-encode it as %7C). A single template with no | behaves exactly as before.

Before this plugin resolved <name>_ref to the whole record, fields="venue" always rendered a link (or nothing, if the record had no usable coordinates) — that was the only way to display a ref field. If you're upgrading:

• fields="venue" now renders plain text (the record's ref_text_field). Add :link — fields="venue:link" — to keep the old link rendering.
• Nothing else about existing <path>#<id> refs needs to change; that form still works exactly as before.

Anywhere a field name is accepted — fields, group_by, sort_by, group_summary_at, field_labels, aggregate — it may be:

• A dotted path into a nested field, e.g. venue.city (most useful against a resolved <field>_ref, but works against any nested YAML data). A missing or null value anywhere along the path renders as a blank cell rather than erroring.
• Suffixed with :transform, e.g. date:year or venue:link — and the two compose: venue.date:year walks the dotted path first, then applies the transform to what it finds there.

field_labels and hidden key off the bare path (no :transform suffix) — field_labels="venue:場館" labels the column whether it's shown as venue or venue:link, since the transform is a rendering choice, not a different field.

TABULAR_COUNT_TEMPLATE and TABULAR_GROUP_COUNT_TEMPLATE have built-in defaults for zh ({n} 筆資料 / {n} 筆) and ja ({n} 件), derived from Pelican's DEFAULT_LANG setting.

group_by reorders rows so that rows sharing the same key values are contiguous. group_summary_at adds a collapsible header row above each group that shows the group value and a row count.

{% table data/books.yaml group_by="genre,author" group_summary_at="genre" %}

This renders a genre-level header row for each genre, with all books listed beneath it. The header is collapsible via this plugin's shared table core — see CSS / JS below.

A group_by (and group_summary_at) field may use a field:transform form to group by a value derived from the field rather than its raw value. The only transform currently supported is year, which extracts the year from a date/datetime:

{% table data/talks.yaml sort_by="date" sort_order="desc" group_by="date:year" group_summary_at="date:year" %}

This groups rows under a collapsible header per year while keeping the original date column intact. Sorting still operates on the underlying date.

When aggregate is set, rows sharing a group_by key are collapsed into a single row. Currently supported operations:

sum/avg/min/max coerce numeric-looking strings (e.g. "8.5") but skip anything that can't be parsed as a number; if no value in the group is numeric, the aggregated cell is empty.

{% table data/books.yaml group_by="author" aggregate="year:year" %}
{% table data/books.yaml group_by="genre" aggregate="rating:avg,pages:sum" %}

Every <th> element has an id attribute derived from the column label (e.g., id="osm-col--title"). These can be used as fragment links within the page. IDs are unique within a table; duplicate slugs get a numeric suffix (-2, -3, …).

The plugin copies its assets to output/static/pelican_tabular/. A shortcode with view="..." includes the optional view assets automatically, once per content item, using that site's SITEURL. No theme change is needed to opt in.

Plain tables do not include database controls or new asset tags. When using the migrated OSM plugin, keep your existing osm-map.js / osm-map.css tags: its build step bundles the shared table core at those same URLs. For plain tables without OSM, include only the lightweight table assets in your theme:

<link rel="stylesheet" href="{{ SITEURL }}/static/pelican_tabular/css/tabular-legacy.css">
<script src="{{ SITEURL }}/static/pelican_tabular/js/tabular-core.js" defer></script>

The optional tabular.js bundle is assembled from tabular-core.js and tabular-views.js; tabular.css imports tabular-legacy.css. Deploy the whole generated directory. Plain tables retain the existing selectors, anchors, localized count settings and data format. Starting with tabular 0.7.0 and OSM 0.16.1, tabular owns shared table behavior and styling. Use OSM 0.16.1 or later when both plugins are installed; older OSM versions contain their own table controller.

Without JavaScript, table content and links remain readable. Interactive views show their details and hide inactive controls until initialization succeeds.

Define a named view in pelicanconf.py and select it in the shortcode:

TABULAR_VIEWS = {
    "works": {
        "fields": ["title", "format", "genres", "rating"],
        "field_labels": {"title": "作品", "format": "形式", "rating": "評分"},
        "search_fields": ["title", "creator", "note"],
        "sort_fields": ["title", "rating", "released_at"],
        "field_types": {"rating": "number", "released_at": "date"},
        "filters": {
            "status": {
                "options": [
                    {"value": "ongoing", "label": "正在看", "tone": "amber"},
                    {"value": "completed", "label": "已看完", "tone": "green"},
                    {"value": "planned", "label": "想看", "tone": "blue"},
                ]
            },
            "format": {"match": "any"},
            "genres": {"match": "all"},
            "released_at": {"control": "year_range"},
        },
        "display": {
            "layout": "responsive",
            "filters_expanded": "auto",
            "title_field": "title",
            "meta_fields": ["format", "genres", "rating"],
            "detail_fields": ["creator", "released_at", "status", "note"],
        },
        "presets": [{"label": "正在看", "filters": {"status": ["ongoing"]}}],
        "query_sync": True,
    }
}

{% table data/works.yaml view="works" id="works" sort_by="rating" sort_order="desc" %}

The complete example includes rating descriptions, multiple date ranges, all reading statuses, links to reviews and a standalone theme. Its fictional data needs no external service.

Year ranges include both endpoints. Rows without a usable date do not match an active range; a list of dates matches if any date is in the range. Reversed ranges are rejected. Clearing filters also clears search, while keeping the sort order. Grouped tables sort within groups; collapsed rows remain part of the matched count. New interactive views currently reject aggregate; plain tables and OSM's existing aggregation remain supported.

URLs use repeated parameters for multiple selections, for example ?works.status=ongoing&works.genres=科幻&works.genres=日常&works.released_at.from=2020. Search uses works.q, sorting uses works.sort / works.order. Unrelated query parameters and fragments are preserved. Changes use replaceState, so each keystroke does not add a history entry; navigation restores the URL's state. Detail and filter-panel expansion are not stored in the URL.

Only fields needed for display, details, search, filters and sorting are embedded in the page. A field used for searching is public even if it is not a visible column. Use hidden for presentation, not for protecting confidential data.

Colors can be themed through scoped .tabular-view CSS variables such as --tabular-bg, --tabular-text, --tabular-border and --tabular-accent. The defaults use Attila's --brand and --color-background-* / --color-content-* tokens when present, with standalone fallbacks. System dark mode, .theme-dark / .theme-light, .dark and data-theme are supported.

The view inherits the theme's font family and defaults to 18px work/item titles, 16px content and controls, and 15px labels. Override --tabular-font-size, --tabular-control-size and --tabular-label-size on a view to adjust them. Toolbar and filter controls have a minimum 44px height. Mobile rows show the configured title first, followed by bold field labels and values; link lists wrap without the indentation used by ordinary tables. Mobile and desktop share the same rows, preserving links, selections and expanded details when resizing.

Cells expose data-field for site-specific column styling, for example #works td[data-field="rating"]. Page headers, introductions, category names, field widths and domain-specific badge choices remain the site's responsibility. These defaults apply only to named views; ordinary {% table %} output and OSM place lists keep their existing appearance and assets. See the release and adoption notes when replacing site overrides.

Control labels and live counts are marked data-pagefind-ignore; table content and details remain available to the site's search index. Built-in messages use generic data terminology; the works example supplies its own reading vocabulary.

Components use shortcode/view lang, then the article/page's Lang, then DEFAULT_LANG. English, Traditional Chinese and Japanese UI catalogs are included. Matching preserves scripts: zh-Hans falls back to English instead of silently using Traditional Chinese.

Labels, filter options and presets accept language-to-text mappings. Optional TABULAR_TRANSLATIONS and TABULAR_REF_TRANSLATIONS project explicitly allowlisted text fields without changing IDs or source data. Counts support CLDR plural forms, and existing count templates remain supported.

See configuration, fallback rules and migration examples.

The dependency direction is pelican-osm → pelican-tabular. The core never imports OSM or Leaflet.

• pelican.plugins.tabular.i18n provides Catalog, component_locale, localized_text, project_record and plural message helpers. Each package owns its JSON catalog.
• pelican.plugins.tabular.core.collapse_rows(rows, group_by, aggregate, union_fields=()) handles grouping/aggregation; consumers can request union semantics for fields such as OSM tags.
• pelican.plugins.tabular.rendering.render_table_body(...) accepts prepared rows, a render_row callback and optional group_suffix/group_key_value callbacks (the latter defaults to core.group_key_value; pass your own for dotted-path or reference-aware group keys). It owns group boundaries, counts, anchors and headers; OSM retains its coordinate links, schema hints and image cells. Callbacks return trusted HTML and must escape their data.
• pelican.plugins.tabular.views.render_view(rows, config, table_id=..., lang=...) renders a complete interactive view without shortcode initialization.
• pelican.plugins.tabular.assets.register_assets() registers a shared, idempotent Pelican asset-copy hook. Consumer plugins call this from their own register() even when the tabular shortcode plugin is not enabled.
• pelican.plugins.tabular.assets.bundle_legacy_assets(script, stylesheet) adds the shared engine and legacy CSS to a consumer's freshly copied output files. Call it after each copy so rebuilds do not append repeated bundles. OSM uses this to preserve its existing asset URLs without including database controls.
• window.Tabular.initTable(table, {formatCount}) attaches the shared controller idempotently. OSM supplies its localized count formatter and attaches its own lightbox handler. window.Tabular.init() initializes tables and, when tabular-views.js is loaded, newly added views; the tabular:ready event handles asset loading order.

uv sync --group dev
uv run poe lint
uv run poe cover
npm ci
npm test
npm run example
uv run python scripts/build_browser_fixtures.py
npx playwright install chromium
npm run test:browser

Serve examples/database/output and open database.html to explore the example. Browser tests cover light/dark layouts at 360, 768 and 1280 pixels and save screenshots under test-results/. Compatibility fixtures exercise unchanged OSM asset URLs, Attila article styles, manual theme selection and subsite paths. Integration fixtures include multiple views and legacy OSM markup; pass --osm-source ../pelican-osm to the fixture builder to test a live migrated OSM checkout, including its lightbox. The fixture builder also creates 1,000/5,000-row performance pages; browser tests report initialization and filtering timings.

See the implementation verification record for results, measurement limits and the coordinated OSM release order.