feat(cimcheck): add Intellij plugin#15
Draft
spah-soptim wants to merge 64 commits into
Draft
Conversation
- intellij: default javaExecutable to "" so resolveJavaExecutable falls through to IntelliJ's bundled JBR. The previous "java" default made the JBR auto-detection dead code, breaking the plugin where "java" is not on PATH (e.g. Windows). - intellij: document the deliberate ".ttl" file-type claim and its conflict/precedence implications in plugin.xml. - core: substitute the SHACL $PATH/?PATH placeholder only as a whole token (negative look-ahead) and via Matcher.quoteReplacement, so variables like $PATHOLOGY are left untouched and '$' in the IRI is not mistaken for a back-reference. - core: in validateAutoDetect, parse the query once with the analyzer's base URI and pass the parsed Query to the validator (new overload), removing a redundant re-parse and the probe/validate base-URI inconsistency. - core: locate property-path-chain diagnostics near the preceding segment (locateNear) so the squiggle lands on the right occurrence. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Extract ValidationConfig interface so CliConfig/LspConfig cannot silently diverge - Lift buildNamedGraphScope() into SparqlValidationApi to eliminate copy-paste between CLI and LSP - Delegate CLI SchemaLoader.buildIndex() to CgmesSchemaLoader so unparseable files are skipped-with-warn rather than causing a hard failure - Replace hand-rolled JSON serialisation in JsonFormatter with Jackson ObjectMapper - Unify SHACL path walkers into a single walkPath(onUri, onAlternativeGroup) callback traversal - Extract DefaultPrefixes.declaredPrefixes() and reuse it in SparqlTextDocumentService to eliminate the duplicate PREFIX_PATTERN field - Add BY/IN/TO to SPARQL_STATEMENT_KEYWORDS and guard findBadKeywordLine() against built-in function calls (e.g. COALESCE()) - Remove dead SchemaIndex variable in definition() handler - Extract positionFromMessage() helper to deduplicate SPARQL error-position parsing - Reuse tokenLengthInSource() in turtleParseErrorDiagnostic() instead of an inline scan Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…mpatibility settings
The repo-wide '*.jar' ignore also excluded gradle/wrapper/gradle-wrapper.jar,
so the wrapper jar was never committed and './gradlew' failed on a fresh
clone ("Unable to access jarfile gradle-wrapper.jar"). Add a negation rule
to re-include it and commit the regenerated wrapper.
Align the wrapper distribution to gradle-9.5.0 (a valid release) to match the
gradle-version pinned in the CI and release workflows.
Configure the IntelliJ Plugin Verifier (pluginVerification {} with
recommended() IDEs and a Marketplace-aligned failureLevel) and run it in
CI's build-intellij-plugin job. The verifier flagged two defects across the
2024.2–2026.1 IDE range that the prior "fixes" had only relocated:
- addBrowseFolderListener(Project, BrowseFolderActionListener) is deprecated
in 242, scheduled-for-removal in 243-261, and REMOVED in 262 (262 reported
a NoSuchMethodError compatibility problem). Replace with the stable
addActionListener + FileChooser.chooseFile pair.
- PluginManager.findEnabledPlugin is @ApiStatus.Internal in 262. Drop it and
read the plugin version from a build-time-generated classpath resource
(cimcheck-plugin.properties, expanded by processResources) instead.
verifyPlugin now reports all 7 verified IDEs (IC-2024.2 through IU-262 EAP)
as Compatible with no deprecated/internal/scheduled-for-removal usages.
…pe registration - Rewrite the plugin.xml Marketplace description with Features, Requirements, Getting started (.cgmes/validation.json), Settings, and Troubleshooting. - Add cimcheck/intellij/README.md mirroring the VS Code docs (diagnostics table, validation.json reference, IntelliJ shortcuts, LSP4IJ log location). - Implement SparqlFileType/ShaclFileType as plain FileType instead of LanguageFileType(PlainTextLanguage.INSTANCE). Binding the built-in PlainText language caused a file-type/language association conflict that could block extension registration (no highlighting / file not recognised). verifyPlugin passes against all 7 IDEs (IC-2024.2 through IU-262 EAP).
…s override
The IntelliJ plugin never managed to launch the language server ("Cannot start
server: Unable to start language server: ProcessStreamConnectionProvider
[commands=[...java.exe, -jar, ...jar]]") even though the command was correct.
Cause: LSP4IJ's ProcessStreamConnectionProvider.start() reads the private
`commands` field directly, but the provider only overrode getCommands(). The
field stayed empty, so start() threw — while toString() (which uses the getter)
printed the correct command, masking the real problem.
Populate the command list and working directory via setCommands()/
setWorkingDirectory() in an init block. Also clarify in the plugin description
and README that LSP4IJ auto-installs only on a Marketplace install, not an
install-from-disk.
verifyPlugin passes against all 7 IDEs (IC-2024.2 through IU-262 EAP).
…etons Plain FileType is not guaranteed to trigger the editorHighlighterProvider pipeline — IntelliJ's highlighter factory expects LanguageFileType for the provider lookup path to fire reliably. Introduce SparqlLanguage and ShaclLanguage singletons (extending Language directly with unique IDs) so each file type has its own Language, eliminating the PlainTextLanguage conflict that caused the original revert while restoring LanguageFileType and fixing syntax highlighting.
…ugin.xml IntelliJ validates that the language= attribute matches what the LanguageFileType instance returns; omitting it logs a SEVERE error and aborts file type registration, breaking highlighting. With dedicated SparqlLanguage/ShaclLanguage singletons the IDs are stable, so declaring language="SPARQL" and language="SHACL" is both correct and required.
…ctory editorHighlighterProvider is a low-level hook for custom EditorHighlighter overrides; the standard IntelliJ highlighting pipeline for LanguageFileType files goes through lang.syntaxHighlighterFactory instead. Replace the two editorHighlighterProvider registrations with lang.syntaxHighlighterFactory entries (one per language) backed by the new CimcheckSyntaxHighlighterFactory, and remove the now-unused CimcheckEditorHighlighterProvider class.
…Mapping LSP4IJ resolves the server by Language when the file type is a LanguageFileType; fileTypeMapping only fires for plain FileType files. Since SparqlFileType and ShaclFileType now extend LanguageFileType with dedicated Language singletons, the server was no longer being activated for SPARQL/SHACL files. Replace the two fileTypeMapping entries with languageMapping entries keyed on Language IDs "SPARQL" and "SHACL".
…ine range Without a ParserDefinition, IntelliJ has no PSI token structure for SPARQL/SHACL files. When Ctrl is held, CtrlMouseHandler finds the entire file as the single PSI element and underlines it all. Register minimal SparqlParserDefinition and ShaclParserDefinition — flat parsers backed by the existing CimcheckLexer that produce a token-level PSI tree without any real AST — so IntelliJ knows the boundaries of each token and underlines only the one under the cursor.
Replace all hardcoded version strings with 0.0.0-SNAPSHOT / 0.0.0 placeholders. Two new scripts derive and apply versions from git tags: - scripts/compute-version.sh <component>: emits X.Y.Z on a release tag push, or X.Y.Z-SNAPSHOT (patch+1 of last tag) on any other ref. - scripts/set-versions.sh <cimxml-ver> [<cimcheck-ver>]: applies versions to all Maven poms, gradle.properties, and package.json in one call. Release workflows no longer validate pom versions against tags; the validate-tag jobs only check the tag format and extract the version. CI snapshot-publish jobs compute the snapshot version from git before deploying. The duplicated versions:set / SNAPSHOT-strip blocks and the "TODO: remove once cimxml is hardcoded as a release version" workarounds are removed.
…elease Add a publishing token to build.gradle.kts (read from PUBLISH_TOKEN env var) and a new publish-jetbrains-marketplace job in cimcheck-release.yml. The job runs in the "JetBrains Marketplace Deployment" environment, depends on validate-tag and publish-maven-central (gates on a valid release), builds the LSP fat JAR + plugin, and calls gradle publishPlugin. The marketplace token must be stored as the JETBRAINS_MARKETPLACE_TOKEN secret in that environment.
… push cimcheck: switch gh release create to --draft and update artifacts to cimcheck-core jar, CLI jar, VS Code vsix, and IntelliJ plugin zip (LSP jar removed — it is bundled inside the extensions, not a standalone download). cimxml: add a new publish-github-release job that builds the cimxml jar and creates a draft GitHub Release, gated on publish-maven-central. Both releases use --generate-notes and --draft so they can be reviewed and published manually after the automated jobs complete.
Add a rename step that copies cli/lsp JARs to include the release version in the filename before gh release create, so all artifacts have consistent versioned names (cimcheck-X.Y.Z.jar, cimcheck-lsp-X.Y.Z.jar). The copy happens after the extension-bundling steps that reference the original fixed names, so no other build step is affected.
- cimcheck-cli: finalName cimcheck-cli-\${project.version}
- cimcheck-lsp: finalName cimcheck-lsp-\${project.version}
- VS Code vsix: vsce package --out cimcheck-vscode-X.Y.Z.vsix
- IntelliJ zip: already cimcheck-intellij-X.Y.Z.zip (unchanged)
- cimcheck-core jar: already cimcheck-core-X.Y.Z.jar (unchanged)
build.gradle.kts copyServerJar updated to match with a glob + rename so
it is version-independent. CI and release workflows updated to reference
the new names; the post-build rename step in the release workflow is
removed.
…maven-central verifyPlugin downloads IDE distributions and is slow + disk-heavy; removed it from build-intellij-plugin. For both cimcheck and cimxml releases, the GitHub Packages, GitHub Release, and JetBrains Marketplace jobs now use `if: always() && ...` so they proceed even when publish-maven-central fails.
spah-soptim
force-pushed
the
cimcheck-intellij
branch
from
baa57d0 to
7c761cf
Compare
The build-vsix and build-intellij-plugin jobs were building with the placeholder version (0.0.0), causing CI artifacts to be named cimcheck-vscode-0.0.0.vsix and cimcheck-intellij-0.0.0.zip. Add fetch-tags: true and an "Apply snapshot versions" step to both jobs so compute-version.sh can resolve the last release tag and set-versions.sh propagates the snapshot version into gradle.properties and package.json before the builds run.
- analysis: drop duplicate ClassRefKey/PropertyRefKey records; dedup sets now hold ClassReference/PropertyReference directly - core: extract shared IriFormat util (shortIri/appendIris), removing the copy duplicated in SparqlQueryValidator and ShaclShapeAnalyzer - SparqlQueryValidator: collapse validateReferences' 10 params into an AnalysisRefs record; merge formatClass/PropertyMessage into one formatMissingTermMessage; use the imported PrefixMapping - SparqlValidationApi: add getUpdateProfileDependencies(String, Map) and getGraphDependencies(String, Collection) overloads for API symmetry - lsp: switch volatile LanguageClient client to AtomicReference in SchemaManager and SparqlTextDocumentService - lsp: delete unused DiagnosticConverter and fix dangling javadoc link - tests: cover the previously-untested dependency overloads
…ndpoint Validate SPARQL cells inside VSCode SPARQL Notebooks, not just .rq/.sparql files, and honour the notebook "# [endpoint=...]" directive as a schema source. - vscode client: add the vscode-notebook-cell scheme to the document selector and an onNotebook:sparql-notebook activation event so cells reach the server. - server: augment file-extension detection with languageId-based classification (captured at didOpen, since didChange carries no languageId) so extension-less notebook cell URIs are recognised; IntelliJ/file behaviour is unchanged. - server: resolve the per-cell schema from the endpoint directive — a local .ttl/.rdf/.owl file is loaded via the existing CgmesSchemaLoader and cached by resolved path; absent directive falls back to .cgmes/validation.json. Remote SPARQL endpoints are recognised but not yet loaded.
A notebook "# [endpoint=https://…]" directive now loads the schema from the endpoint itself, where the CGMES profiles live in per-profile named graphs. - core: CgmesSchemaLoader.indexFromGraphs(Iterable<Graph>) wraps each graph as a CimProfile and builds an RdfsSchemaIndex, skipping non-profile/duplicate graphs. Re-asserts the cim namespace prefix (detected from the graph's IRIs) before wrapping, since profile detection is prefix-based and SPARQL CONSTRUCT results arrive without prefixes. - lsp: EndpointGraphFetcher enumerates schema named graphs (rdfs:Class/owl:Ontology, excluding instance data) and CONSTRUCTs each via Jena QueryExecutionHTTP with a per-query timeout. - lsp: SchemaManager loads remote endpoints asynchronously on the schema-loader thread (never blocking the validator), caches by endpoint, negative-caches failures, and revalidates open documents once the schema lands.
spah-soptim
force-pushed
the
cimcheck-intellij
branch
from
6b51bb3 to
726431c
Compare
…es to improve responsiveness
…nd integrate into IDE
…/Turtle When a SHACL/Turtle document's schema cannot be resolved (e.g. an unreachable # [endpoint=...]), the LSP previously published no diagnostics, silently hiding broken shapes. Mirror the existing SPARQL fallback: report Turtle parse errors and syntax-check embedded SPARQL fragments, skipping schema-dependent checks. - core: add static SparqlValidationApi.checkShaclSyntaxOnly(Graph), the SHACL counterpart of checkSyntaxOnly(String) - lsp: extract parseTurtle helper; route the unresolved-schema + endpoint branch of validateShacl through new publishShaclSyntaxOnly - test: ShaclSyntaxOnlyValidationTest covering valid/broken fragments, sh:prefixes, and the no-schema/no-fragment cases
…ocumentation, bundle cgmes 3.0
… detection Move SPARQL-endpoint schema loading out of the LSP and into the core library so it is usable headlessly (CLI/pipeline), and add automatic detection of which instance named graph holds which CGMES profile. Core (de.soptim.opencgmes.cimcheck.core.schema): - SparqlGraphSource abstraction with HttpSparqlGraphSource (moved from the LSP EndpointGraphFetcher, keeping the Fuseki update→query 405 sibling fallback) and DatasetSparqlGraphSource for in-process/test use. - NamedGraphProfileResolver: classifies each instance graph to its dominant profile by discriminating-term coverage (terms declared by exactly one profile), sampling predicates + rdf:type objects only — it never validates instance data. - EndpointSchema + EndpointSchemaLoader: one call fetches the schema graphs, builds the index, and auto-maps instance graphs to profiles. An endpoint that exposes no CIM schema yields EndpointSchema.noSchema() rather than throwing. LSP: SchemaManager.loadRemoteEndpoint now uses EndpointSchemaLoader, stores the auto-detected per-graph scope in ResolvedSchema, and shows visible warnings when an endpoint has no schema (falls back to syntax-only) or when a graph could not be classified. Old LSP EndpointGraphFetcher removed. CLI: new --endpoint/-e (auto schema + graph mapping) and --strict-endpoint (no-schema → exit 2 instead of the default warn + syntax-only). Verified end-to-end against a live Fuseki: graphs auto-mapped, per-graph scoped validation flags a class only valid in another profile, and both no-schema branches behave as designed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…OT_CONFIGURED)
CGMES queries routinely navigate the RDFS schema directly, e.g.
GRAPH <cgmes:schema:EQ> { ?t rdfs:subClassOf* cim:Switch } .
Endpoint auto-detection excluded schema graphs from the per-graph scope, so such
queries were flagged GRAPH_NOT_CONFIGURED even though they are valid — they read
the meta-schema, not instance data.
EndpointSchemaLoader now maps every detected schema graph to all profiles, so CIM
terms used inside a schema graph validate permissively (cim:Switch resolves; a
typo is still caught) instead of producing a spurious warning. EndpointSchema
gains instanceGraphsMapped() so the LSP/CLI status messages still report the
instance-graph count separately from the schema graphs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ects
The implied-type INFO hint ("?x has no explicit rdf:type but its use of <P>
implies <C>") only counts a concrete rdf:type as "typed", so it fired on the
canonical CGMES idiom
GRAPH <schema> { ?type rdfs:subClassOf* cim:Switch }
GRAPH <model> { ?sw a ?type ; cim:IdentifiedObject.name ?name }
where ?sw is clearly typed — just dynamically. The hint there is pure noise.
SubjectTypeInference.subjectsWithVariableType() collects subjects that carry a
variable type assertion (?s a ?var); SemanticChecks now skips the implied-type
hint for those, while still emitting it for genuinely untyped variables.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…x-only fallback A misspelt SHACL term (e.g. sh:taaargetClass, sh:dataype) was reported by the full SHACL validation path but NOT when the schema could not be resolved — e.g. when a # [endpoint=...] points at an unreachable / non-query endpoint, which makes the editor fall back to checkShaclSyntaxOnly(). That fallback only did a Turtle parse + embedded-SPARQL syntax check and returned no shape annotations, so the typo went silently unreported in VS Code. The vocabulary-typo check is schema-independent (it consults only the bundled W3C sh/rdf/rdfs/owl vocabularies), so it can and should run without a schema: - ShaclShapeAnalyzer.checkVocabularyOnly(graph): new schema-free entry point; the underlying vocab check is now static. - SparqlValidationApi.checkShaclSyntaxOnly: includes those vocab findings in shapeAnnotations. - LSP publishShaclSyntaxOnly and CLI validateSyntaxOnly: render the shape annotations (they previously dropped everything but embedded-SPARQL results). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ling A # [endpoint=...] pointing at a Fuseki operation endpoint other than query — e.g. the workshop notebook's SHACL cells use .../svedala/shacl?graph=cgmes:model:SSH (the server-side SHACL validator), the same URL the execution tooling needs — could not load a schema: cimcheck reads the schema via SPARQL, which /shacl does not serve. Now such endpoints transparently resolve to the dataset's /query service for schema loading, so one directive works for both execution and static validation. Changes to HttpSparqlGraphSource: - Strip any URL query string (?graph=...) up front — meaningless to the ASK/SELECT/CONSTRUCT schema queries and a source of routing confusion. - Generalize queryEndpointSibling: map the trailing Fuseki operation segment update/shacl/data/get/upload -> query (was update-only). query/sparql and a bare dataset URL are left as-is. - Probe the given URL first (so a dataset literally named shacl/update is honoured), then fall back to the /query sibling on failure. - Fix the caught exception type: Jena's QueryExecutionHTTP throws org.apache.jena.sparql.engine.http.QueryExceptionHTTP, NOT org.apache.jena.atlas.web.HttpException — the old catch never fired, so even the original /update fallback was a no-op live (only its string-helper unit test passed). The fallback now actually triggers. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…o syntax-only
When a # [endpoint=...] document's schema can't be resolved (endpoint failed, or
still loading) the editor already validated syntax-only and showed a bottom-right
notification, but the document itself gave no in-line signal that semantic checks
were skipped. Add a first-line WARNING diagnostic ("Schema could not be loaded
from the endpoint — only syntax is checked …") in both the SPARQL
(publishSyntaxOnly) and SHACL (publishShaclSyntaxOnly) fallbacks. It clears
automatically once the schema loads and the document is re-validated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
spah-soptim
force-pushed
the
cimcheck-intellij
branch
from
f2a4a5e to
ba3e974
Compare
… schema Previously only validation honoured a # [endpoint=...] directive; hover, completion and go-to-definition all read the workspace/bundled schema. So an endpoint cell validated against Fuseki but showed hover text, completions and "go to definition" from the bundled CGMES 3.0 — the wrong schema — or nothing when the term wasn't in the bundled schema at all. - Hover & completion now resolve the document's schema the same way validation does (endpoint schema when declared and resolved, else workspace). The endpoint-fetched index already carries rdfs:label/rdfs:comment, so descriptions are correct. When an endpoint is declared but not yet resolved, nothing is shown rather than wrong-schema info. - Go-to-definition: for an endpoint term there is no local file, so EndpointDefinitionPeek fetches the term's triples from the endpoint (HttpSparqlGraphSource.fetchResource), renders them as a cached read-only Turtle file, and returns a file:// Location at the term's line. Works for any SPARQL endpoint and uniformly across LSP clients (VS Code, IntelliJ). It no longer jumps to the misleading bundled file. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
spah-soptim
force-pushed
the
cimcheck-intellij
branch
from
ba3e974 to
c649552
Compare
The first-line "schema could not be loaded from the endpoint — only syntax is checked" diagnostic is now ERROR (was WARNING), so an unreachable/failed endpoint is clearly surfaced rather than blending in with advisory warnings.
…tribution
There is no longer a bundled default schema. Validation gets its schema from an
opencgmes.json ("schemas"/"schemasDirectory") or a "# [endpoint=...]" directive;
with neither, CIMcheck performs a syntax-only check.
Core:
- Delete BundledSchemas (+ test) and CgmesSchemaLoader.bundledDefault().
- Delete the vendored CGMES 3.0 RDFS resources under resources/cgmes/.
- ConfigTemplate: scaffold now documents pointing at your own schemas.
LSP:
- schema.SchemaLoader.loadWithSources returns Optional (empty = no schemas
configured); drop loadBundledWithSources and the bundled fallback.
- SchemaManager: no config / config-without-schemas → no workspace schema
(apiRef null) instead of the bundled default; BUNDLED sentinel removed.
- Document service: no resolved schema → syntax-only. The first-line "schema
could not be loaded" error notice is shown only for the endpoint case; the
plain no-config case is quietly syntax-only.
CLI:
- schema.SchemaLoader.load(config) returns Optional; drop loadBundled.
- ValidateCommand/ExplainCommand: no --schema/--config/--endpoint (or a config
without schemas) → syntax-only ("Info: no schema configured…"). InitCommand
messaging updated.
Docs / IntelliJ: core/vscode/intellij READMEs, plugin.xml and CreateConfigAction
reworded for the no-bundled-default behavior (server-JAR/JBR "bundled" wording
left intact). vscode README's stale "endpoint features = diagnostics only"
limitation corrected (hover/completion/go-to-def are endpoint-aware now).
W3C vocab attribution: the rdf/rdfs/owl/shacl vocabularies under resources/vocab
(used for standard-vocabulary term checking) are W3C works — added vocab/NOTICE.md
with source URLs, © W3C, and the W3C Software and Document License, and referenced
it from the READMEs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
spah-soptim
had a problem deploying
to
Maven Central Deployment
GitHub Actions
Failure
spah-soptim
had a problem deploying
to
JetBrains Marketplace Deployment
GitHub Actions
Failure
…nd add local search functionality
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.