Modernize option defaults for the next major release

[schani]

Modernizes the Elm target from Elm 0.18 to Elm 0.19 and brings the Elm fixtures back into CI.

Breaking: Elm 0.18 output is removed

The renderer now emits Elm 0.19 code only. There was never a version option, so nothing was deleted from the options surface — the generated code simply targets 0.19 now, and 0.18 projects can stay on older quicktype releases.

Generated-output changes

• Decode pipelines are seeded with Jdec.succeed Foo instead of Jpipe.decode Foo (decode was removed in NoRedInk/elm-json-decode-pipeline 1.0; install instructions in the header comment updated accordingly).
• The 0.19 elm/json encoder API (Json.Encode.list/array take the element encoder, Json.Encode.dict exists) makes the generated makeArrayEncoder/makeListEncoder/makeDictEncoder helpers unnecessary — they are gone; only makeNullableEncoder remains.
• Imports no longer expose map/toList (ambiguous under 0.19 shadowing rules): import Dict exposing (Dict), plus import Array exposing (Array) only with --array-type array; List needs no import.
• Elm 0.19 forbids parameters shadowing top-level definitions, so the parameter names used by generated functions (x, y, r, f, m, str, somethingElse) are now forbidden as generated global identifiers (caught by priority/unions.json, which has a type named x).
• Generated identifiers can no longer start with _ (an 0.19 syntax error, caught by blns-object.json).
• String literals escape non-printables with 0.19's \u{XXXX} syntax instead of \uXXXX/\UXXXXXXXX (caught by identifiers.json / nst-test-suite.json).

array-type now defaults to list

Defaults-audit recommendation #13, deferred until this modernization: List is the idiomatic Elm collection, and with the 0.19 encoder API the list path needs no helper at all. --array-type array still works and stays covered in CI by a quicktest entry (quickTestRendererOptions).

Fixture revival

The driver in test/fixtures/elm/ is rewritten for 0.19:

• elm-package.json → elm.json (elm/core, elm/json, NoRedInk/elm-json-decode-pipeline 1.0.1).
• Main.elm uses Platform.worker (Platform.program is gone) and Jdec.errorToString for decode errors.
• runner.js uses the 0.19 embedding API (Elm.Main.init() instead of Elm.Main.worker()).
• A Warmup.elm module is compiled once by the setup command so all package dependencies are downloaded/built into the shared ELM_HOME cache before per-sample compiles run in parallel (elm 0.19.1 can corrupt its package registry on concurrent cold-cache downloads). The sysconfcpus hack for the 0.18 compiler is gone.

Skip lists (every entry re-verified against elm 0.19.1)

Un-skipped (fixed by the renderer changes above): identifiers.json, simple-identifiers.json, blns-object.json, nst-test-suite.json, keywords.json.

Still skipped, with verified reasons:

• recursive.json, direct-recursive.json, bug427.json, bug790.json, list.json and the recursion schemas (union-list, list, ref-remote, mutually-recursive, postman-collection, vega-lite, simple-ref, plus newly recursive-union-flattening): Elm type aliases cannot be recursive.
• constructor.schema, keyword-unions.schema: compile and run now, but elm/json's field decoder uses the JS in operator, which finds inherited Object.prototype members — decoding {} yields a non-null "constructor".
• nested-intersection-union.schema: the generated decoder accepts invalid union members because all class properties decode via Jpipe.optional (same limitation as documented for C++).
• The five newly enabled identifier-heavy JSON inputs are in skipDiffViaSchema (schema round-trip changes inferred names/types); all 20 pre-existing skipDiffViaSchema entries were re-checked and still differ.

CI

- elm,schema-elm is re-enabled in the test matrix; the workflow's elm 0.19.1 install step already existed. Locally validated with elm 0.19.1: the full corpus (215 JSON tests incl. misc, 67 schema tests) passes for both FIXTURE=elm and FIXTURE=schema-elm, as do npm run test:unit and npm run lint.

🤖 Generated with Claude Code