Repository navigation
HermitCrab conformance: 26 fixtures flagged fieldworks_producible still use a slot-rule outputPartOfSpeech FieldWorks never writes #612
Description
18 of 127 conformance fixtures and 8 of 16 stress fixtures are flagged fieldworks_producible: true but give a slot rule an outputPartOfSpeech, which FieldWorks never writes. The flag overclaims. A reader cannot yet trust it to mean that a real project can produce the grammar.
Status: open. The first retirement pass only acted on fixtures flagged false; it is on #480, and the stress half is on #578.
What we know, at e9569ec:
- Identity
outputPartOfSpeechon a slot rule: 18 conformance fixtures. Dropping it changed no parse in the fixtures already converted. edge-cases/morphotactic-attribute-breadthalso has apartial="true"entry with a part of speech, a partial rule in a slot, and a partial rule withoutputPartOfSpeech.- 52 conformance fixtures have a
linearstratum: 5 explicit and 47 by the DTD default oflinear. HCLoaderbuilds its Morphology and Clitics strata asUnordered. Surface sets no order, so a linear Surface stratum with no rules may be legitimate.
Not known: how many of the 52 have morphological rules in a linear stratum, and whether flipping them to unordered changes any parse.
Proposed change: a one-line drop in each slot-rule fixture, one rebuild of morphotactic-attribute-breadth, and a per-stratum count for linear, then convert or delete. Expected parses come from the engine. A check that fails on these patterns when a fixture is flagged true would stop them coming back.
Related: #480 (open), #578 (open), #552 (open).
Next: count the linear strata that hold morphological rules, then convert the fixtures.
Counting method
Parsed every grammar.xml at e9569ec: a fixture counts when a rule named in a Slot morphologicalRules attribute has outputPartOfSpeech, and a linear count when any Stratum has morphologicalRuleOrder="linear" or omits the attribute. HermitCrabInput.dtd:244 sets the default. XmlLanguageWriter.cs:586 writes the attribute only for unordered. MPR feature groups were reviewed and match the groups HCLoader hardcodes, so they are not in this list.
🤖 Generated with Claude Code
Activity
johnml1135 commentedon Nov 2, 2026
The grammars this issue asked to fix now live in sillsdev/fieldworks-parser-conformance, and the problem it described no longer exists there. No grammar in the suite gives a slot rule an
outputPartOfSpeech, and no stratum that lists a rule or holds a template islinear.What remains of the idea, that the
fieldworks_producibleflag only rules grammars out and cannot show that FieldWorks writes them, is tracked in the suite as sillsdev/fieldworks-parser-conformance#1 (generate witnesses and certify by round trip), #2 (tie each witness to its grammar), #3 (the round-trip rule and the unmeasured assumptions) and #11 (the shape guard cannot tell a plain affix process from a lone affix form).The guard test itself,
FieldWorksProducibleShapeTests, stays in machine and is in #615. Nothing in this issue is left open that those issues do not cover.