Design notes
ADR 146: the :name sugar

ADR 146: the `:name` sugar

Status: accepted 2026-10-04 (decision 146 and its addendum in the decisions log). Depends on: ADR 145 (defaultTag). Implementation: three PRs (parser patch, core, tooling and docs).

Context

Marko has two tag-adjacent shorthands borrowed from CSS selectors: <div#main> sets id, <div.card> sets class. A great many tags, in HTML (<input>, <select>, <button>, <slot>, <meta>) and in data vocabularies (attributes, fields, resources, actions), carry a name attribute, and there is no shorthand for it. The data target made the gap visible: a vocabulary that declares attributes writes <attribute name="title" type="string"/> for every entry.

MX is a language first (its targets exist to test it), so the question was treated as one of language design: which spelling, in which positions, at what cost in parser work and divergence from Marko.

Decision

:val sets name="val", the way #val sets id and .val sets class. All three sugars work in three positions:

position	example	meaning
tag-adjacent	`<input:email>`, `<:email>`	`name="email"`; a bare `<:email>` is the unnamed tag of ADR 145 with a name
first attribute	`<input :email>`	same
after any attribute	`<input type="email" :email>`, `<input x="1" #main .big>`	same, in HTML and concise mode

Tag-adjacent sugars compose in any order (<a.c:b>, <a#d:b.c>, <:b.c>). The value is the same identifier-like token Marko’s shorthand accepts; :x=1 is a positioned error. After the sugar is resolved, name is an ordinary attribute: contracts, duplicate rules and attribute typing apply unchanged.

What was measured

All claims below were measured on @marko/compiler 5.42.5 (htmljs-parser 5.15) with a capture translator, 2026-10-04.

Marko’s parser is permissive about tag names: only whitespace, /, >, ., #, ,, =, ( and | end or change a name. <a:b> is the tag a:b, <:b> is the tag :b. So the tag-adjacent split is a post-parse rewrite; no parser work.
#b in attribute position already parses as an attribute named #b, first, after a quoted value, after an unquoted value, in concise mode.
:b first or after , parses as the default attribute with a modifier, { name: "value", modifier: "b" }, i.e. value:b. After any attribute value it is the parse error “Expected a single expression, but found : after it”: the parser reads : as a ternary continuation of the previous value.
.b after any value is swallowed as member access (x="1" .b is "1".b).
Bare :mod meaning value:mod is undocumented. It falls out of two generic parser rules (split an attribute name at its last :; fill an empty head with value), and the only use found anywhere was MX’s own oracle fixture.
The expression-continuation rule is general: after a value, any token that can continue a JavaScript expression continues it (- + * / % & | ^ < ., ?/:, in, instanceof, =>, a ( call, across newlines). It never bites today because every real attribute name starts with a letter, _, $, @ or #, none of which continue an expression.

Alternatives considered

option	works in every position without parser change	reads as “name”	rejected because
`:`	no (`:` after a value is an error)	yes	chosen; the parser rule it needs is non-breaking
`$`	yes	no: reads as “variable” in JS, shell and PHP	sugar that is misread is sugar people avoid
`@`	yes	partly	`<@x>` is an attribute tag; one sigil, two positions, two meanings
`&`	no (`x=a &b` is a valid bitwise expression today)	YAML anchor reading is nice	needs a breaking parser change; real ambiguity, not just an error
`::`	no (same `:` error)	yes	same parser problem; two spellings for one thing once `<input:email>` exists
wildcard attribute tags (`<@title type="string"/>`)	yes	different thing	forces a tag’s whole contract into attribute-tag shape; attribute tags do not have every tag capability
replace the modifier operator `:` with `\|`	n/a	n/a	changes a documented Marko feature (`class:x`, `style:x`, `value:fn:=x`) and buys nothing the chosen rule does not

Consequences

Four divergences from Marko, each recorded in divergences.md:

A tag name may not contain :; tag:rest is tag plus name. No HTML element has a : in its name, so nothing real is lost.
Bare :x in attribute position is name="x", not value:x. The named modifier forms (class:x, style:x, value:fn:=x) are untouched.
A shorthand class or id cannot contain : (<div.hover:x> is class hover, name x). Tailwind-style variants use class="hover:x", which is unchanged.
Parser: after whitespace inside an attribute value, a : with no open ? in that value, and a . followed by an identifier start, begin a new attribute instead of continuing the previous value. The : half breaks nothing (every input it affects is a parse error today) and is offered upstream to htmljs-parser. The . half changes x=a.b .c (member access across whitespace), which must now be written without the space or in parentheses.

MX carries rule 4 as a bun patchedDependencies hunk on htmljs-parser until upstream takes the : half. Marko’s own language tools will flag :x after a value in .marko files; .mx files use MX’s tools.

What stays true: # and . keep their Marko meaning and merge rules; every existing golden is byte-identical; the sugar never produces anything a user could not have written as name="…".

ADR 146: the :name sugar