Skip to content

predicate-grammar.md

Latest commit

 

History

History
75 lines (59 loc) · 3.69 KB

predicate-grammar.md

File metadata and controls

75 lines (59 loc) · 3.69 KB
Raw

Predicate grammar

A Processing-Model @predicate decides when a <model> applies: an element takes the first of its models, in document order, whose predicate holds. Params (<param value>) are XPath too. How far each target reads them:

XSLT — odd-to-xslt.mjs, odd-to-xsl.xsl

Any XPath 3.1. Both generators copy predicates and params verbatim into the generated XSLT 3.0 stylesheet, where the XSLT processor evaluates them. The stylesheet sets xpath-default-namespace to TEI's, so the ODD's unprefixed names are TEI elements. A prefix the ODD binds where the expression stands is resolved when the ODD is read, to Q{uri}local, or to the unprefixed name for TEI's namespace, so the expression needs no declaration where it is carried.

unified / xast and CETEIcean — odd-to-unified.mjs, odd-to-ceteicean.mjs, odd-to-ceteicean.xsl

A subset of XPath, implemented in runtime/xpath.mjs. It ships beside the generated handlers, which evaluate predicates and params with it, and is inlined in the CETEIcean pages, which evaluate them in the browser:

  • Paths: ., .., @name, name, *, node(), text(), /, //, absolute paths, and [predicate] (a number is a position).
  • Axes: self, child, parent, ancestor, ancestor-or-self, descendant, descendant-or-self, preceding-sibling, following-sibling, preceding, following.
  • Names: unprefixed or tei:, TEI's; Q{uri}local another namespace's, as a prefix the ODD binds becomes. A prefix it does not bind is outside the subset. x:*, *:name and namespace-uri() are too.
  • Operators: or, and, the general comparisons (= != < <= > >=), the value comparisons (eq ne lt le gt ge), + -, mod, if (…) then … else …, and sequences (a, b).
  • Functions: position, last, count, not, exists, empty, string, concat, contains, contains-token, starts-with, ends-with, matches, replace, translate, normalize-space, string-length, local-name, true, false.

What it accepts follows XPath 3.1: a path gives each node once, in document order; a general comparison is numeric only against a number; a value comparison with more than one item on a side is an error when evaluated, as is a call with the wrong number of arguments when compiled. Two known differences: regular expressions are JavaScript's, and @n eq 1 compares numerically where XPath raises a type error.

odd-to-unified.mjs and odd-to-ceteicean.mjs compile every expression once. A CETEIcean page compiles them again when it loads.

Outside the subset: degrade, don't fail

An expression the subset cannot evaluate — an eXist function, an XQuery let chain, a union — is compiled to what a failing evaluation would have returned anyway: false() for a predicate, so the model never fires and the next one decides, and the empty sequence for a param. Someone else's ODD therefore runs here, degraded where it has to be, instead of being rejected.

What was degraded is recorded, never silent: each generator writes odd-unsupported.json beside what it generates —

{ "element": "p", "model": "p-1", "where": "predicate", "expression": "@a | @b",
  "reason": "unexpected \"|\"", "odd": "bad.odd", "line": 3 }

— prints the count in the build log, and findings.mjs lists the entries with the rest of what the ODD does not foresee.

--strict fails the build instead, naming the element, the model and the line. Our own ODDs are compiled with it (see package.json), so a regression in them stays a build error.

CSS

The CSS needs no predicates: the renderers put the selected model's classes on the element, and edition.css styles those classes.