wapdown/README.md
randogoth 3b728d9d74 Make --- divide cards, and stop drawing ASCII rules
WML has no horizontal rule element -- its entire %layout entity is
`<br>` -- so a thematic break was being rendered as `<p>------------</p>`,
a hardcoded twelve-dash guess at screen width that wraps into nonsense on
a narrow device and is short on a wide one. It was the one construct
where the renderer invented a width out of nothing.

A `---` already looks like a divider and reads as one in every Markdown
document ever written, so it now is one: it starts a new card, by
default. Rules are never drawn, whether or not they divide.

Thematic breaks and `{.card Title}` markers are now a single mechanism --
an explicit divider, optionally carrying a title -- and compose freely in
one document. Both still win over --split-level. A section takes its
title from its own first heading when the divider does not supply one, so
the common shape needs no titles at all:

    Intro on the menu.

    ---

    ## Weather

    Cold and clear.

`--no-split-on-rule` / `split_on_rule: false` opts out of the splitting;
it does not bring the dashes back.

Three edges this opened, each fixed here rather than left to bite:

- Setext headings. `Heading` over `---` is an H2 in Markdown, but the
  parser was ATX-only, so it produced a paragraph plus a rule. Once a
  rule divides cards, a setext document would have split at every heading
  and stranded each heading as body copy on the card before it. Setext
  headings are now parsed, so such a document keeps its structure.
- Empty sections. A trailing `---`, or two in a row, produced an empty
  "Untitled" card and a menu entry leading to it. Sections with no
  renderable content are dropped.
- The multi-card threshold was two sections, so a document with a single
  divider collapsed back into one card. Any explicit divider now makes a
  multi-card deck: `A --- B` asks for two screens.

Also fixes a latent frontmatter bug this made far more likely: a document
opening with `---` and no `key: value` lines had its opening swallowed as
if the break were a frontmatter fence. A fenced block with no keys in it
is not frontmatter.

222 tests. Rendering examples/trail.md still matches md2txt's committed
output byte for byte; no golden file contains an ASCII rule any more.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-23 09:31:51 +03:00

10 KiB

wapdown

Compile Markdown into a WML deck for WAP browsers.

wapdown trail.md -o trail.wml

Write a document the ordinary way; get back valid WML 1.3 that a 2001 handset can actually navigate.

Why this is not just another text exporter

Most Markdown converters transform a document: same content, different markup, one file in and one file out. A WML deck is not that shape. A WAP screen is a few lines tall, scrolling is painful, and the input device is a numeric keypad — so a deck is a set of short cards joined by softkeys and menus, not one long page.

So wapdown restructures. It decides where cards begin, builds a menu that exists nowhere in your source, wires up Back and More controls, and paginates anything still too big for the device's cache. That is why it has more knobs than a plain text exporter needs: the knobs are the tool.

Install

Install the CLI with uv:

uv tool install --from git+https://codeberg.org/randogoth/wapdown/ wapdown

Zero runtime dependencies, Python 3.13+.

Cards

A document with no dividers becomes a single card. There are three ways to ask for more.

Thematic breaks — a plain --- starts a new card. WML has no rule element to draw (its entire layout vocabulary is <br>), and a row of ASCII dashes is noise on a small screen, so the glyph is put to better use as the divider it already looks like:

Intro shown on the menu card...

---

## Weather

Cold and clear.

---

## Trail Status

- North Loop: open

Each card takes its title from its own first heading. *** and ___ work identically — any Markdown thematic break divides. A trailing ---, or two in a row, adds no empty card. --no-split-on-rule turns this off; breaks are then ignored rather than drawn, because there is still nothing to draw them with.

--- under a line of text is a setext heading, not a break, and is parsed as one — so a document written in that style keeps its headings instead of shattering into cards.

Named dividers — {.card Title} is the same mechanism with an explicit title, for when a section has no heading to borrow one from:

{.card Trail Status}
- North Loop: open

The two compose freely in one document.

Heading-level splitting — --split-level 2 starts a new card at every ##, which fits a document that is already structured by headings and has no dividers in it:

wapdown trail.md --split-level 2

Explicit dividers win over --split-level when both are present: a boundary written into the document is a stronger statement of intent than a rule passed on the command line.

Navigation

Once a document resolves to two or more cards, wapdown builds the classic WAP-portal shape — a menu card listing each section, each spoke reachable in one press and returning with Back:

wapdown trail.md --split-level 2 --menu-style select

--menu-style select renders those choices as a <select> rather than a list of links, so the destination is picked with a digit instead of scrolled to. On a keypad that is usually the better default; it is not the default here only because it changes how the deck looks.

--no-menu switches to a linear book instead: cards chained with More and Back, no hub.

--home-label Home adds an options softkey jumping straight back to the menu from anywhere, which <prev/> alone cannot do since it only pops history.

Shared navigation is hoisted into a deck-level <template> so it is transmitted once rather than repeated on every card — a real saving when the whole deck has to fit in a cache. --no-template-nav turns that off.

Size limits

WAP 1.x gateways cap how much deck a device will hold. --max-card-bytes (default 1400, 0 disables) is the safety net: any card still over budget after section splitting gets paginated with Prev/More. Some headroom is reserved for each card's navigation markup, which is why the effective budget is slightly below the number you pass.

When a deck as a whole is too large, paginating inside one file does not help — the file still has to be fetched whole. --deck-per-card writes one .wml file per section into an output directory, cross-linked by filename, so a device only ever downloads the card it asked for:

wapdown trail.md --split-level 2 --deck-per-card -o ./decks/
# decks/index.wml  decks/weather.wml  decks/trail-status.wml  ...

In menu mode the hub lands in index.wml. In linear mode there is no hub, so the entry point is the first section's file.

Images

wapdown performs no image conversion. Most WAP 1.x browsers render only WBMP, so referencing a PNG generally yields a broken-image placeholder on the device. --images decides what to do about that:

Value Behaviour
keep (default) Emit <img> as written, whatever the format.
alt Replace non-WBMP images with their alt text, and warn.
drop Remove non-WBMP images entirely, and warn.

.wbmp sources always pass through untouched.

Configuration

Every option can be set on the command line or in the document's frontmatter. The frontmatter key is the long flag name with underscores; command-line flags win.

---
title: Trail Conditions
split_level: 2
menu_style: select
cache_control: 0
---

# Trail Conditions
...
Flag Frontmatter Default Meaning
--title title (none) Deck title, used on the menu card.
--split-level split_level 0 Heading level that starts a new card; 0 disables. Ignored when {.card} markers are present.
--max-card-bytes max_card_bytes 1400 Per-card byte safety net; 0 disables.
--menu / --no-menu menu true Hub-and-spoke menu vs. linear More/Back chaining.
--menu-style menu_style links links or select for the menu's choices.
--split-on-rule / --no-split-on-rule split_on_rule true Start a new card at every thematic break.
--deck-per-card deck_per_card false Write one file per section into the output directory.
--nav-next-label nav_next_label More Forward label.
--nav-prev-label nav_prev_label Prev Label between pagination parts.
--nav-back-label nav_back_label Back Label back to the menu or previous section.
--home-label home_label (none) Adds an options softkey to the menu card.
--template-nav / --no-template-nav template_nav true Hoist shared nav into a <template>.
--images images keep keep, alt, or drop.
--cache-control cache_control (none) Emit a Cache-Control: max-age=N meta element.
--access-domain access_domain (none) Emit <access domain=...>.
--access-path access_path (none) Emit <access path=...>.

Markdown support

Markdown WML
Headings <p><b><big> for the first on a card, <p><b> after
Paragraphs, lists <p> — WML has no list element, so bullets and numbers become literal text
--- / *** / ___ a card divider; never drawn, as WML has no rule element
Setext headings (=== / --- underlines) same as # / ##
**bold**, *italic* <b>, <i>
~~strike~~, `code` markers stripped; WML has no equivalent
Links <a href>
Images <img src alt>, subject to --images
Fenced/indented code <p> with <br/> between lines
Blockquotes <p><i>
Pipe tables <p><table columns="N"> with a bold header row
{.card Title} a card divider carrying a title
{.include file.md} inlined before conversion

Two consequences of WML's own grammar are worth knowing:

  • A literal $ is written $$. WML uses $ to introduce a variable substitution, so wapdown doubles it for you. Permit fee: $5 comes out as Permit fee: $$5, which the device renders as $5.
  • Link labels cannot carry emphasis. WML declares <a> as (#PCDATA | br | img)*, so there is nowhere valid to put a <b> inside one. [**go** now](url) becomes go now, with the markers stripped rather than translated.

WML 1.3 is the only supported DTD, by design rather than by flag.

Serving a deck

Serve .wml as text/vnd.wap.wml — a browser that receives text/html will usually refuse to render it. --cache-control 0 is worth setting for content that changes, since WAP gateways cache aggressively.

Development

uv sync
uv run pytest                  # unit, golden, and well-formedness tests
uv run --extra dtd pytest      # ...plus validation against the real WML 1.3 DTD

The DTD is vendored at tests/dtd/wml13.dtd so the suite never depends on fetching a twenty-year-old URL. DTD validation needs lxml and is skipped when it is absent.

Golden files under tests/golden/ pin whole-document output. After an intentional renderer change:

uv run python tests/regenerate_golden.py
jj diff tests/golden          # the diff is the change, stated in output

Architecture

Markdown is parsed into a stream of block events, which a renderer turns into output. The pieces are registered as plugins (src/wapdown/plugins/), so another WAP-era target — HDML, i-mode cHTML — can be added alongside wml without restructuring anything.

  • parsers/markdown.py — Markdown to BlockEvents
  • renderers/wml/inline.py — inline markup and WML's escaping rules
  • renderers/wml/deck.py — sectioning, byte packing, navigation topology
  • renderers/wml/renderer.py — element emission
  • config.py — defaults, frontmatter, and CLI layering

Unlike a streaming text renderer, the WML renderer buffers the whole event stream: card boundaries, menu choices, and nav targets all depend on knowing the document's full structure up front.

Credit

The WML renderer began life inside md2txt, a suite of plain-text Markdown exporters, and was extracted here because a deck compiler is a different kind of thing from a text exporter.

License

MIT