@marko/runtime-tags@6.3.3

Patch Changes

• #3391 9bb8b3d Thanks @DylanPiercey! - Keying a <for> by its own loop parameter (<for|city| of=cities by=city>) is now a guided compile error suggesting the property-string and function forms (by="id" / by=(city) => key), instead of dying at render time with a bare undefined-variable error — the by= expression is evaluated before the loop runs, so the loop parameter is not in scope there.

• #3393 ef4f23f Thanks @DylanPiercey! - Expose cheatsheet.md through the package exports map so module resolvers can locate it as marko/cheatsheet.md.

• #3388 642c68d Thanks @DylanPiercey! - Make common syntax errors self-guiding. Call-style control flow (<if(cond)>, <await(x) p>, <show(x)>) now says how to write the condition/value as an attribute; bare JS declarations at the template root (let count = 0;) and slashless <let count=0>/<const x=1> point at the tag-variable form and static; unknown tags from other frameworks get curated pointers (<slot> → <${input.content}/>, <state> → <let>) before falling back to nearest-name suggestions; JSX-style brace-wrapped attribute values (onClick={handler} and values that only parse once unwrapped) explain that attribute values are plain JavaScript expressions; attribute tags on <await> point at <try> with <@placeholder>/<@catch>. Also fixes a typo in the nested-attribute-tags assertion message ("Tag not support" → "Tag does not support").

• #3332 5fc3bce Thanks @DylanPiercey! - Reads recorded after their owning expression was merged into another
  node's extra (eg a $global read tracked at identifier-visit time after
  an <if>'s analyze merged its test into the tag's extra) now land on the
  merge target instead of splitting the expression's references.

• #3390 ec97955 Thanks @DylanPiercey! - Report every analyze-stage error in a template at once instead of stopping at the first. Analyze failures are now recorded through the compiler's diagnostics channel (diagnosticError), and the compiler mirrors the parse layer by throwing all error diagnostics together at the end of the analyze stage (duplicates deduped, capped at 8) — or, when compiling with errorRecovery, keeping them as recoverable diagnostics for editors instead of throwing. The tags translator reports tag-level analysis failures this way, skipping the failed tag's subtree to avoid cascading follow-on errors. Templates with a single error produce byte-identical output.

• #3332 5fc3bce Thanks @DylanPiercey! - $signal abort-ids are now allocated once during analyze and stamped on
  each expression root's extra, so every output/entry translate of a cached
  file reads the same id — agreement across compiles is structural instead
  of a property of identical visit order. Previously the ids were allocated
  at translate time from a module-level map keyed by section; sections are
  cached-analysis objects that outlive compiles — the compiler cache
  shares one analyzed file across every output/entry kind and each compile
  translates a clone — so the old map leaked allocations across compiles of
  the same cached file and drifted the ids ($signal(scope, 0) → 1) on
  the second dom-mode translate of the same cached file.

• #3392 d734bd0 Thanks @DylanPiercey! - Ship an LLM-optimized syntax reference (cheatsheet.md) inside the published marko package, giving tooling a stable in-package path (node_modules/marko/cheatsheet.md) to point coding agents at when they write or fix Marko code.

• Updated dependencies [642c68d, ec97955]:

  • @marko/compiler@5.40.2