hydra-build 0.18.0 → 0.18.1
raw patch · 3 files changed
+163/−6 lines, 3 filesdep ~hydra-kernelPVP ok
version bump matches the API change (PVP)
Dependency ranges changed: hydra-kernel
API changes (from Hackage documentation)
+ Hydra.Build.Modules: isLibModule :: Module -> Bool
+ Hydra.Build.Modules: partitionLibModules :: [Module] -> [Module] -> ([Module], [Module])
Files
- CHANGELOG.md +144/−4
- hydra-build.cabal +2/−2
- src/main/haskell/Hydra/Build/Modules.hs +17/−0
CHANGELOG.md view
@@ -15,6 +15,144 @@ --- +## [0.18.1] - 2026-11-20++Textual-sources release: the whole `hydra-kernel` is now authored as textual `.hydra` files, including the+primitive library `hydra.core.lib.*`, and the textual syntax gains the sugar, parsers, and tooling needed to carry it.+Also lands translingual tail-call optimization for the Scala and TypeScript coders, named provisions and+primitive parameter names, the property-graph encoding of term graphs, and a large body of host-specific bootstrap+and CI fixes.++### Breaking changes++- **Type renames** ([#837](https://github.com/CategoricalData/hydra/issues/837)): four types whose variant names clashed with other kernel types are renamed.+ Any code, JSON, or `.hydra` text that names them must be updated.+ - `hydra.core.regex.QuantifierRange` is now `hydra.core.regex.Range`.+ - `hydra.ext.azure.dtld.SchemaPrimitive` is now `PrimitiveSchema`, and `SchemaComplex` is now `ComplexSchema`.+ - `hydra.rust.syntax.TypePath` is now `PathType`.+- **New primitive signature syntax in `.hydra`** ([#837](https://github.com/CategoricalData/hydra/issues/837)): a `primitive` definition names its parameters,+ `primitive ifElse ∀x.(p: boolean, t: lazy x, f: lazy x) : x`, instead of the old arrow chain+ (`primitive ifElse : ∀t. boolean → lazy t → lazy t → t`).+ Parameter and result descriptions move into `@param` and `@return` doc lines.+ The arrow-chain form is for `term` signatures only.+- **Raw `case` keyword is now `cases`** ([#810](https://github.com/CategoricalData/hydra/issues/810)): the textual-syntax case statement keyword matches `Term.cases`.+- **`hydra.core.json.bootstrap` is now `hydra.core.bootstrap`** ([#791](https://github.com/CategoricalData/hydra/issues/791)).+- **`and` and `or` are lazy in their second operand** ([#802](https://github.com/CategoricalData/hydra/issues/802)): both are marked lazy, mirroring `ifElse`, and the+ Java and Python hosts add thunk-accepting overloads.++### Highlights++- **The whole kernel is `.hydra`-sourced** ([#837](https://github.com/CategoricalData/hydra/issues/837), [#711](https://github.com/CategoricalData/hydra/issues/711), [#710](https://github.com/CategoricalData/hydra/issues/710), [#497](https://github.com/CategoricalData/hydra/issues/497)):+ every `hydra-kernel` module, including all `hydra.core.lib.*` primitive modules and the test modules, is sourced from+ `packages/hydra-kernel/src/main/hydra/**/*.hydra`; Phase 1 reads that text instead of the Haskell DSL.+ The Haskell DSL modules remain as a twin that must stay byte-identical (the dual-edit rule).+ `bin/verify-hydra-sources.sh` is the identity guard (it is not run by CI),+ and `heads/haskell/bin/print-hydra-sources.sh` regenerates the text from the DSL.+ A package may textually source only modules it owns.+- **Named primitive parameters** ([#837](https://github.com/CategoricalData/hydra/issues/837)): all 292 primitives print with named parameters and `@param`/`@return`+ descriptions, carried through the kernel metadata.+- **`parseModules`** ([#842](https://github.com/CategoricalData/hydra/issues/842)): `hydra.core.parse.packaging.parseModules` reads `.hydra` text into modules and is the+ supported entry point for downstream code generation from textual sources+ (see `docs/recipes/downstream-codegen.md`).+- **Kernel JSON on the classpath** ([#786](https://github.com/CategoricalData/hydra/issues/786)): the published `hydra-kernel` jar bundles the kernel's JSON modules, so+ Java downstream projects no longer need a source checkout to load the kernel.+- **Tail-call optimization on more hosts** ([#709](https://github.com/CategoricalData/hydra/issues/709)): shared `TcoLoopPlan`/`TcoStep` kernel types and a+ `lowerSelfTailRecursion` pass, consumed by the new Scala (`@tailrec`) and TypeScript lowerings.+- **Named provisions** ([#725](https://github.com/CategoricalData/hydra/issues/725)): primitives and tests carry named requirements and recommendations,+ spec prose is ported into module and primitive metadata, and the generated spec pages render comments and lifecycle.+- **Term graphs as property graphs** ([#738](https://github.com/CategoricalData/hydra/issues/738)): `hydra.pg.subtermGraph` encodes a term graph as a property graph,+ with specification pages (`docs/specification/pg/subterm-graph.md`).++### New features++- **Textual syntax sugar** ([#809](https://github.com/CategoricalData/hydra/issues/809), [#811](https://github.com/CategoricalData/hydra/issues/811)): nominal construction and elimination sugar+ (`record`/`inject`/`wrap {...}`, `project`/`unwrap [...]`), applied `case` sugar, and monadic `do` blocks+ (`do either`, `do list`, `do optional`, `do effect`).+- **Source spans** ([#815](https://github.com/CategoricalData/hydra/issues/815)): the parsers track source offsets and record term spans through the new+ `hydra.core.positions` module; spans are round-trip-exempt surface trivia.+- **Parsers and printers** ([#710](https://github.com/CategoricalData/hydra/issues/710), [#497](https://github.com/CategoricalData/hydra/issues/497)): `hydra.core.parse.core` (term, type, and type scheme parser),+ parse/print for packaging, doc-comment parsing into `EntityMetadata`, and `line:col` parse errors.+ A signature-level `lazy` keyword round-trips `Parameter.isLazy` ([#800](https://github.com/CategoricalData/hydra/issues/800)),+ and raw `cases(...)` accepts the `_` default field ([#822](https://github.com/CategoricalData/hydra/issues/822)).+- **JSON** ([#776](https://github.com/CategoricalData/hydra/issues/776), [#777](https://github.com/CategoricalData/hydra/issues/777)): `JsonWriterOptions` and `printJsonWith` for configurable JSON layout,+ and `\uXXXX` escapes with surrogate pairs in the JSON parser.+- **Error printers** ([#778](https://github.com/CategoricalData/hydra/issues/778)): real printers for inference and extraction errors replace placeholder constants.+- **Spec generator** ([#723](https://github.com/CategoricalData/hydra/issues/723)): the `hydra.RegenerateSpec` driver and `bin/regenerate-spec.sh` now exist,+ and the module-reference pages under `docs/specification/` are generated by them.+ The module reference is still partial (only the primitives and four kernel type modules are covered).+ Primitive pages show named signatures (`isAlpha(char: int32) → boolean`) with Parameters and Returns entries,+ type pages list fields and variants, and headings use local names so in-page links work.+- **Build system in Hydra** ([#788](https://github.com/CategoricalData/hydra/issues/788)): the lib-pass partition (`isLibModule`, `partitionLibModules`) moves into the+ translingual `hydra.build.modules`, consumed by the Java, Scala, and Haskell drivers.+- **Bootstrap matrix** ([#769](https://github.com/CategoricalData/hydra/issues/769)): `docs/bootstrap-matrix.md` persists the N×N cross-host generation matrix.+- **Build slot tooling** ([#730](https://github.com/CategoricalData/hydra/issues/730), [#814](https://github.com/CategoricalData/hydra/issues/814)): `with-stack-slot.sh` gains a second lane, priorities, pinning, and a memory guard,+ and scripts that run bare `stack` re-exec under it.+- **CI** ([#821](https://github.com/CategoricalData/hydra/issues/821), [#700](https://github.com/CategoricalData/hydra/issues/700)): an inference-cost guard (time and peak memory) and a dedicated haskell-tests job.++### Bug fixes++- [#799](https://github.com/CategoricalData/hydra/issues/799): shadow-renaming of variables was decoupled across three layers (`unshadowVariables`,+ `analyzeFunctionTermWithGather`, the Java coder's `dedupBindings`).+- [#803](https://github.com/CategoricalData/hydra/issues/803), [#804](https://github.com/CategoricalData/hydra/issues/804), [#805](https://github.com/CategoricalData/hydra/issues/805): printed import aliases and bare names now round-trip.+- [#798](https://github.com/CategoricalData/hydra/issues/798): a `Term.variable` encodes and decodes as `{"@var": "name"}` at any JSON position.+- [#817](https://github.com/CategoricalData/hydra/issues/817): `printFloat32`/`printFloat64` for subnormals in the Common Lisp, Emacs Lisp, and Clojure hosts.+- [#779](https://github.com/CategoricalData/hydra/issues/779): the Python coder no longer breaks on nested module names.+- [#752](https://github.com/CategoricalData/hydra/issues/752): `PropertyType.required` was hardcoded to true.+- [#751](https://github.com/CategoricalData/hydra/issues/751), [#832](https://github.com/CategoricalData/hydra/issues/832), [#838](https://github.com/CategoricalData/hydra/issues/838): the SHACL serializer emits blank nodes for anonymous shape references, RDF lists for+ `sh:in`/`sh:and`/`sh:or`/`sh:xone`, definition references under `sh:node`/`sh:not`/`sh:and`/`sh:or`/`sh:xone`,+ and qualified value shapes with their counts.+- [#754](https://github.com/CategoricalData/hydra/issues/754): IRI strings are escaped in the rdf4j serializer.+- [#748](https://github.com/CategoricalData/hydra/issues/748), [#781](https://github.com/CategoricalData/hydra/issues/781): `inferInGraphContext` routes through the discharge-and-finalize tail, and+ `inferModules` seeds its graph from both universe and target modules.+- [#764](https://github.com/CategoricalData/hydra/issues/764): `refreshModule` fails loudly on dropped bindings; binding dedup and primitive shadowing are fixed.+- [#649](https://github.com/CategoricalData/hydra/issues/649): the Python coder merges a type and a term that share a module path.+- [#780](https://github.com/CategoricalData/hydra/issues/780): generated Javadoc escapes backslashes and `*/`.+- [#782](https://github.com/CategoricalData/hydra/issues/782): simple Unicode case mapping on the Common Lisp and Emacs Lisp hosts.+- [#775](https://github.com/CategoricalData/hydra/issues/775): `mapSet` is unregistered on hosts that never implemented it.+- [#289](https://github.com/CategoricalData/hydra/issues/289), [#762](https://github.com/CategoricalData/hydra/issues/762), [#801](https://github.com/CategoricalData/hydra/issues/801): stale post-rename paths in the Go overlay imports, the TypeScript bootstrap driver,+ and the Common Lisp and Emacs Lisp test loaders.+- [#741](https://github.com/CategoricalData/hydra/issues/741), [#747](https://github.com/CategoricalData/hydra/issues/747), [#757](https://github.com/CategoricalData/hydra/issues/757), [#771](https://github.com/CategoricalData/hydra/issues/771), [#787](https://github.com/CategoricalData/hydra/issues/787), [#823](https://github.com/CategoricalData/hydra/issues/823), [#831](https://github.com/CategoricalData/hydra/issues/831): sync and cold-seed fixes+ (target assembly order, `hostOverrides` immunity, overlay hashing in the freshness gate, downstream assembly after+ overlay re-apply, the generator stamp's coder-copy component).+- [#736](https://github.com/CategoricalData/hydra/issues/736): CI pins JDK 11 for the Haskell job after `ubuntu-latest` moved to JDK 25.+- [#755](https://github.com/CategoricalData/hydra/issues/755): effects fixes (Type.effect encoder arms, Python `map_list`, Emacs Lisp loading).+- [#812](https://github.com/CategoricalData/hydra/issues/812): the Clojure/Scheme `testEnv` redirect applies to generated test trees only.+- [#768](https://github.com/CategoricalData/hydra/issues/768): the cold-seed driver step fails fast when its local fallback cannot succeed.+- [#825](https://github.com/CategoricalData/hydra/issues/825): `dist/haskell/hydra-kernel` is reseeded when stale relative to `dist/json`.+- [#839](https://github.com/CategoricalData/hydra/issues/839): the assemble freshness gate refreshes a stale input digest after a rebase instead of failing, and never rewrites a current one.+- [#826](https://github.com/CategoricalData/hydra/issues/826): Lisp coders reference nullary effect primitives by call.+- [#827](https://github.com/CategoricalData/hydra/issues/827): overlay variant-order tables are checked against `model.json`.+- [#772](https://github.com/CategoricalData/hydra/issues/772), [#756](https://github.com/CategoricalData/hydra/issues/756): several silent failures in the sync scripts and vitest/Scala drivers now fail loudly.++### Improvements++- Python DSL phantom types are threaded through combinator signatures ([#731](https://github.com/CategoricalData/hydra/issues/731), [#732](https://github.com/CategoricalData/hydra/issues/732), [#818](https://github.com/CategoricalData/hydra/issues/818)).+- Hackage and PyPI batch publish sets are derived from the package registry ([#573](https://github.com/CategoricalData/hydra/issues/573), [#577](https://github.com/CategoricalData/hydra/issues/577)).+- Pre-existing namespace leftovers from the 0.18.0 rename are cleaned up in Lisp, Scala, TypeScript, and Java bootstrap+ drivers ([#767](https://github.com/CategoricalData/hydra/issues/767), [#729](https://github.com/CategoricalData/hydra/issues/729)).+- `/bootstrap`, `/sync-default`, `/test`, and the pre-push gate default to Java and Python, with Haskell opt-in+ ([#773](https://github.com/CategoricalData/hydra/issues/773)); the bootstrap demo holds the build slot only for its own `stack` calls ([#797](https://github.com/CategoricalData/hydra/issues/797)).+- `bin/sync.sh` prints per-phase timing ([#824](https://github.com/CategoricalData/hydra/issues/824)), and the CI Haskell job is split to stay within the runner time limit.+- `with-stack-slot.sh` fails fast on low disk before a build ([#789](https://github.com/CategoricalData/hydra/issues/789)).+- `bin/verify-hydra-sources.sh --dsl` runs the dual-edit guard in reverse, regenerating from the Haskell DSL twins and+ comparing with the committed JSON ([#837](https://github.com/CategoricalData/hydra/issues/837)).+- `docs/hydra-lexicon.txt` prints primitives with named parameters, matching the generated spec pages+ ([#723](https://github.com/CategoricalData/hydra/issues/723)).++### Documentation++- `docs/specification/syntax.md` gains the module-level grammar, doc comments, import aliasing, file layout,+ `@param`/`@return`, and the named-parameter primitive signature.+- The kernel README, `docs/implementation.md`, `docs/build-system.md`, and the test and primitive recipes describe the+ `.hydra` sources and the dual-edit rule.+- The hand-written primitive spec pages use the `hydra.core.lib.*` namespace.+- Recipes and READMEs document the decoder/encoder fixpoint guard ([#760](https://github.com/CategoricalData/hydra/issues/760)), keeping derived modules out of+ re-inference ([#758](https://github.com/CategoricalData/hydra/issues/758)), the hydra-ext coder testing pattern ([#699](https://github.com/CategoricalData/hydra/issues/699)), scripts with no automated caller+ ([#714](https://github.com/CategoricalData/hydra/issues/714)), overlay-versus-prune ordering ([#645](https://github.com/CategoricalData/hydra/issues/645)), the Phase 1 memory envelope ([#592](https://github.com/CategoricalData/hydra/issues/592)),+ and the Rust coder's home in `hydra-rust` ([#382](https://github.com/CategoricalData/hydra/issues/382)).++---+ ## [0.18.0] - 2026-09-29 Module-naming systematization release: every kernel and package module moves onto a uniform@@ -48,9 +186,11 @@ JSON-Schema/Avro/GraphQL/GraphSON/Protobuf coders (extended integer widths, map-key handling), the kernel's encode/decode generators (reversed type-parameter order on 2+-param polymorphic types), and the Python coder (silently dropping a module's types when types and terms share a file path).-- **Self-hosted specification docs** ([#723](https://github.com/CategoricalData/hydra/issues/723)):- the module-reference pages under `docs/specification/` are now generated directly from the kernel- (`bin/regenerate-spec.sh`) via a new metadata-only doc coder, rather than hand-maintained.+- **Specification-docs generator, kernel side** ([#723](https://github.com/CategoricalData/hydra/issues/723)):+ a new metadata-only doc coder (`generateModuleDoc`) and `hydra.markdown` model, the translingual core+ for generating the module-reference pages under `docs/specification/`. *Erratum:* the Java driver that+ `bin/regenerate-spec.sh` invokes (`hydra.RegenerateSpec`) did not ship in 0.18.0, so the pages are still+ hand-maintained; the driver is pending under #723. ### New features @@ -58,7 +198,7 @@ suite covering the minimal-core grammar (literals, classes, quantifiers, alternation, anchors, grouping) across all 9 hosts. - **[#723](https://github.com/CategoricalData/hydra/issues/723)** `hydra.markdown` type module and- `generateModuleDoc`, the metadata-only doc coder backing the self-generated spec pages.+ `generateModuleDoc`, the metadata-only doc coder intended to back self-generated spec pages. - **[#382](https://github.com/CategoricalData/hydra/issues/382)** the Rust coder is split out of `hydra-ext` into its own first-class `hydra-rust` package.
hydra-build.cabal view
@@ -5,7 +5,7 @@ -- see: https://github.com/sol/hpack name: hydra-build-version: 0.18.0+version: 0.18.1 synopsis: Hydra's build system, promoted into Hydra description: Hydra is an implementation of the LambdaGraph data model, which takes advantage of an isomorphism between labeled hypergraphs and typed lambda calculus: in Hydra, "graphs are programs, and programs are graphs". This package is Hydra's build system, promoted into Hydra itself: the manifest-derived module-to-package router, kernel/host reconciliation utilities, and pure module-list helpers, expressed as Hydra modules and generated into every self-hosting dialect. It builds on hydra-kernel. category: Data@@ -64,6 +64,6 @@ build-depends: base >=4.19.0 && <4.22 , containers >=0.6.7 && <0.8- , hydra-kernel ==0.18.0+ , hydra-kernel ==0.18.1 , scientific >=0.3.7 && <0.4 default-language: Haskell2010
src/main/haskell/Hydra/Build/Modules.hs view
@@ -5,6 +5,7 @@ module Hydra.Build.Modules where import qualified Hydra.Core.Ast as Ast+import qualified Hydra.Core.Codegen as Codegen import qualified Hydra.Core.Coders as Coders import qualified Hydra.Core.Diff as Diff import qualified Hydra.Core.Docs as Docs@@ -68,6 +69,22 @@ Lists.filter (\m -> Lists.foldl (\acc -> \d -> Logic.or acc (case d of Packaging.DefinitionType _ -> True _ -> False)) False (Packaging.moduleDefinitions m)) mods++-- | Whether a module is one of the hydra.core.lib.* primitive-registry modules+isLibModule :: Packaging.Module -> Bool+isLibModule m =++ let name = Packaging.unModuleName (Packaging.moduleName m)+ parts = Strings.splitOn "hydra.core.lib." name+ in (Optionals.match (Lists.uncons parts) False (\uc -> Logic.and (Strings.isEmpty (Pairs.first uc)) (Logic.not (Lists.isEmpty (Pairs.second uc)))))++-- | Lower hydra.core.lib.* modules and partition (lowered lib mods, full lowered universe)+partitionLibModules :: [Packaging.Module] -> [Packaging.Module] -> ([Packaging.Module], [Packaging.Module])+partitionLibModules allMainMods modsToGenerate =++ let libMods = Lists.map (\m -> Codegen.lowerPrimitiveDefinitions m) (Lists.filter (\m -> isLibModule m) modsToGenerate)+ libUniverse = Lists.map (\m -> Logic.ifElse (isLibModule m) (Codegen.lowerPrimitiveDefinitions m) m) allMainMods+ in (libMods, libUniverse) -- | The first two forward-slash-separated segments of a relative path, joined by / secondLevelDir :: String -> Maybe String