docs(lsp): add capability inventory, gaps, and next steps (issue #3)

.gitea/workflows/release.yml

@@ -365,6 +365,12 @@ jobs:
         with:
           version: latest
 
+      - name: Set version from tag
+        working-directory: editors/vscode
+        run: |
+          TAG="${{ github.event.inputs.tag || github.ref_name }}"
+          npm pkg set version="${TAG#v}"
+
       - name: Install and package
         working-directory: editors/vscode
         run: |
@@ -402,6 +408,12 @@ jobs:
           distribution: temurin
           java-version: '21'
 
+      - name: Set version from tag
+        working-directory: editors/datagrip
+        run: |
+          TAG="${{ github.event.inputs.tag || github.ref_name }}"
+          sed -i "s/^pluginVersion=.*/pluginVersion=${TAG#v}/" gradle.properties
+
       - name: Build plugin
         working-directory: editors/datagrip
         run: ./gradlew buildPlugin

.github/workflows/release.yml

@@ -44,6 +44,10 @@ jobs:
           cache: npm
           cache-dependency-path: editors/vscode/package-lock.json
 
+      - name: Set version from tag
+        working-directory: editors/vscode
+        run: npm pkg set version="${GITHUB_REF_NAME#v}"
+
       - name: Install and package
         working-directory: editors/vscode
         run: |
@@ -73,6 +77,10 @@ jobs:
       - name: Setup Gradle
         uses: gradle/actions/setup-gradle@v4
 
+      - name: Set version from tag
+        working-directory: editors/datagrip
+        run: sed -i "s/^pluginVersion=.*/pluginVersion=${GITHUB_REF_NAME#v}/" gradle.properties
+
       - name: Build plugin
         working-directory: editors/datagrip
         run: ./gradlew buildPlugin

AGENTS.md

@@ -36,7 +36,7 @@ pkg/pgast/         — go-pgquery wrapper: SQL → real PG AST (lint, v2)
 pkg/lint/          — rule engine + rule packs (v2)
 pkg/lsp/           — LSP server (v3)
 editors/vscode/    — VSCode extension (v3)
-editors/datagrip/  — LSP4IJ integration (v4)
+editors/datagrip/  — native JetBrains plugin, shells out to the CLI (v4)
 testdata/corpus/   — real-world .pgsql procedures used as the safety/idempotence harness
 ```
 
@@ -59,6 +59,14 @@ testdata/corpus/   — real-world .pgsql procedures used as the safety/idempoten
 3. **Idempotence**: `fmt(fmt(x)) == fmt(x)`.
 4. **Graceful degradation**: any span the parser cannot handle is passed through verbatim
    rather than corrupted.
+5. **Runtime safety gate**: every frontend (CLI `fmt`, LSP `textDocument/formatting` and
+   `rangeFormatting`) calls `format.SemanticallyEqual(src, out)` before writing or returning
+   formatted output. It re-lexes both sides and compares non-trivia token streams
+   (case-insensitive for identifiers/keywords, exact otherwise, recursing into dollar-quoted
+   bodies). If it ever returns false, the formatter has a bug — the caller must refuse to
+   write/emit the result and keep the original source, never guess or best-effort it. This is
+   not just a test assertion (`pkg/format/format_test.go`); it is enforced at runtime so a
+   formatter bug can never silently drop or alter code.
 
 ## Commands
 
@@ -111,8 +119,9 @@ go fmt ./...             # Format Go code
   Per-platform VSIX (`win32/linux/darwin × x64/arm64`) built in CI matrix.
 
 ### V4 — DataGrip
-- `editors/datagrip`: integrate via LSP4IJ (free, works across JetBrains editions incl.
-  DataGrip). No core changes expected.
+- `editors/datagrip`: native JetBrains plugin (no LSP4IJ dependency) that shells out to the
+  `pgtidy` binary directly. Formatting and version info run `pgtidy fmt`/`version`; lint
+  diagnostics and quick-fixes run via a native `ExternalAnnotator` calling `pgtidy lint --json`.
 
 ## Verification
 - **Formatter:** `go test ./...` runs golden-file tests + corpus harness asserting idempotence

cmd/pgtidy/fmt.go

@@ -63,6 +63,10 @@ func cmdFmt(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 			return 2
 		}
 		out := format.File(parser.Parse(string(src)), st)
+		if !format.SemanticallyEqual(string(src), out) {
+			_, _ = fmt.Fprintln(stderr, "pgtidy: refusing to format stdin: formatter safety check failed (output would change code content)")
+			return 2
+		}
 		switch {
 		case check:
 			if out != string(src) {
@@ -86,6 +90,11 @@ func cmdFmt(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 			continue
 		}
 		out := format.File(parser.Parse(string(src)), st)
+		if !format.SemanticallyEqual(string(src), out) {
+			_, _ = fmt.Fprintf(stderr, "pgtidy: refusing to format %s: formatter safety check failed (output would change code content)\n", path)
+			exit = 2
+			continue
+		}
 		changed := out != string(src)
 		if changed {
 			anyDiff = true

cmd/pgtidy/lint.go

@@ -1,6 +1,7 @@
 package main
 
 import (
+	"encoding/json"
 	"fmt"
 	"io"
 	"os"
@@ -18,9 +19,10 @@ import (
 //	file:line:col: [RULEID] severity: message
 func cmdLint(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 	var (
-		only  []string // --only=ID,ID rule filter
-		fix   bool
-		files []string
+		only       []string // --only=ID,ID rule filter
+		fix        bool
+		jsonOutput bool
+		files      []string
 	)
 	for _, a := range args {
 		switch {
@@ -29,6 +31,8 @@ func cmdLint(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 			return 0
 		case a == "--fix":
 			fix = true
+		case a == "--json":
+			jsonOutput = true
 		case strings.HasPrefix(a, "--only="):
 			ids := strings.Split(strings.TrimPrefix(a, "--only="), ",")
 			for _, id := range ids {
@@ -66,7 +70,13 @@ func cmdLint(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 		return diags, nil
 	}
 
-	printDiags := func(diags []diagnostics.Diagnostic) {
+	var jsonDiags []diagnostics.Diagnostic
+
+	reportDiags := func(diags []diagnostics.Diagnostic) {
+		if jsonOutput {
+			jsonDiags = append(jsonDiags, diags...)
+			return
+		}
 		for _, d := range diags {
 			loc := d.File
 			if d.Line > 0 {
@@ -92,14 +102,14 @@ func cmdLint(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 			_, _ = fmt.Fprintf(stderr, "pgtidy: %v\n", err)
 			return 2
 		}
-		if fix {
+		if fix && !jsonOutput {
 			fixed := lint.ApplyFixes(string(src), diags)
 			if fixed != string(src) {
 				_, _ = io.WriteString(stdout, fixed)
 				return 0
 			}
 		}
-		printDiags(diags)
+		reportDiags(diags)
 		if len(diags) > 0 {
 			found = true
 		}
@@ -115,7 +125,7 @@ func cmdLint(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 				_, _ = fmt.Fprintf(stderr, "pgtidy: %v\n", err)
 				return 2
 			}
-			if fix {
+			if fix && !jsonOutput {
 				fixed := lint.ApplyFixes(string(src), diags)
 				if fixed != string(src) {
 					if err := os.WriteFile(path, []byte(fixed), 0o644); err != nil {
@@ -130,13 +140,24 @@ func cmdLint(args []string, stdin io.Reader, stdout, stderr io.Writer) int {
 					}
 				}
 			}
-			printDiags(diags)
+			reportDiags(diags)
 			if len(diags) > 0 {
 				found = true
 			}
 		}
 	}
 
+	if jsonOutput {
+		if jsonDiags == nil {
+			jsonDiags = []diagnostics.Diagnostic{}
+		}
+		enc := json.NewEncoder(stdout)
+		if err := enc.Encode(jsonDiags); err != nil {
+			_, _ = fmt.Fprintf(stderr, "pgtidy: encoding json: %v\n", err)
+			return 2
+		}
+	}
+
 	if found {
 		return 1
 	}
@@ -150,6 +171,7 @@ Read SQL from files (or stdin) and report lint findings.
 
 Flags:
   --fix           Apply autofixes for fixable rules (MIG001, MIG003) and rewrite files
+  --json          Emit findings as a JSON array instead of text (ignores --fix)
   --only=ID,...   comma-separated rule IDs to enable (default: all rules)
   -h, --help      show this help
 

docs/lsp-status.md

@@ -0,0 +1,116 @@
+# PgTidy LSP Status & Roadmap
+
+This document describes the current state of the PgTidy LSP server (`pkg/lsp`,
+`cmd/pgtidy/lsp.go`), what it actually provides today, and concrete next steps.
+It was written as part of issue #3 ("See what we can provide for LSP").
+
+> Note: The V3 milestone is already implemented and shipping (`docs/todo.md` marks
+> V3 done), so this is an inventory + gap analysis, not a greenfield proposal.
+
+## What the server provides today
+
+Verified by hand against the built binary (`pgtidy lsp`, JSON-RPC 2.0 over stdio,
+`Content-Length` framing) — no external LSP library, all wire types hand-rolled.
+
+### Advertised capabilities (`initialize` → `capabilities`)
+
+| Capability | Value | Where |
+|---|---|---|
+| `textDocumentSync` | `1` (full sync) | `serverCaps` |
+| `documentFormattingProvider` | `true` | `handle("initialize")` |
+| `documentRangeFormattingProvider` | `true` | `handle("initialize")` |
+| `codeActionProvider` | `true` | `handle("initialize")` |
+
+### Supported methods
+
+| Method | Direction | Behavior |
+|---|---|---|
+| `initialize` / `shutdown` / `exit` / `initialized` | req/resp + notif | Lifecycle. `exit`/`shutdown` acknowledged with `null` result. |
+| `textDocument/didOpen` | notif | Stores document text; triggers `publishDiagnostics`. |
+| `textDocument/didChange` | notif | Stores latest content version; triggers `publishDiagnostics`. |
+| `textDocument/didClose` | notif | Drops text + fix cache; clears diagnostics with empty list. |
+| `textDocument/formatting` | req/resp | Full-doc format via `pkg/format`; returns one `fullReplace` `TextEdit`. |
+| `textDocument/rangeFormatting` | req/resp | Formats doc, returns minimal edit over the selected line range. |
+| `textDocument/codeAction` | req/resp | Returns quick-fix `WorkspaceEdit`s for fixable diagnostics overlapping the range. |
+| `textDocument/publishDiagnostics` | notif | Sent on every open/change; diagnostic code = `RuleID`, source = `pgtidy`. |
+| `$/cancelRequest` | req | Ignored (per LSP, no response). |
+| unknown | req/resp | `-32601 method not found` (when the request has an `id`). |
+
+### Verified at runtime (e2e smoke test)
+
+- `initialize` returns the capability block above.
+- `didOpen` on `select * from t;` → `publishDiagnostics` with `COR001`
+  ("SELECT * is fragile…", severity 4 = hint, code `COR001`).
+- `textDocument/formatting` on that input → edit replacing with
+  `SELECT *\nFROM t;\n` (keyword casing + clause-per-line applied).
+- `textDocument/hover` → `-32601 method not found` (not implemented — correct).
+
+## What the server does NOT provide (gaps)
+
+These are the most useful, well-scoped gaps to fill next. None are blockers for the
+current shipping state.
+
+1. **No `hover`.** `textDocument/hover` is unimplemented and returns `-32601`.
+   A natural first add: return the `RuleID` + a short explanation for diagnostics
+   on the hovered range, or a keyword/type doc for `hover` on SQL identifiers.
+2. **No `documentSymbol` / `documentLink`.** No outline/symbol tree. For a formatter
+   that already parses `CREATE FUNCTION`/`PROCEDURE` headers into a CST, a symbol
+   provider listing functions/procedures would be low-cost and high-value in large
+   schema files.
+3. **`hover`-style diagnostics shape.** Diagnostics currently use a `Range` whose
+   `end.character` is `start.character + 1` (a 1-char caret), not the actual
+   offending span. A real highlight range would improve editor UX.
+4. **`textDocument/willSave` / `willSaveWaitUntil` / `didSave`.** No save hooks —
+   "format-on-save" must currently be driven by the client binding
+   `textDocument/formatting` to the editor's save event. A `willSaveWaitUntil`
+   handler would let the server own format-on-save.
+5. **No `completion`.** `textDocument/completion` is not implemented. Not urgent for
+   a formatter/linter, but relevant if PL/pgSQL autocompletion (keywords, types) is
+   ever in scope.
+6. **No diagnostics debounce/coalescing beyond full-sync.** Every `didChange`
+   re-runs the full lint engine. Fine for now; a debounce + incremental re-check
+   becomes relevant on large files.
+7. **`initializationOptions` / workspace config.** `initialize` params are parsed
+   nowhere — no way to pass style overrides or a config path over the protocol.
+8. **No `textDocument/prepareRename`, `rename`, `references`, `foldingRange`.**
+   Low priority; would be natural extensions once symbol info exists.
+
+## Conventions to keep consistent
+
+- **One core, many frontends.** The LSP reuses `pkg/diagnostics.Diagnostic` and
+  `pkg/lint` directly — no parallel diagnostic model. New LSP features should reuse
+  these, not fork them.
+- **Safety gate is non-negotiable.** Both `formatting` and `rangeFormatting` call
+  `format.SemanticallyEqual(src, out)` before returning edits; on failure they return
+  an empty edit (keep original). Any new code path that formats must honor this
+  invariant (invariant #5 in `AGENTS.md`).
+- **No new external deps.** The wire layer is intentionally dependency-free. New
+  protocol types should be added as local structs, not pulled in from an LSP library.
+
+## Concrete next steps (recommended, smallest-first)
+
+Ranked by effort/value for the smallest useful delta:
+
+1. **Add a real diagnostic highlight range** (swap the 1-char caret for the actual
+   offending span) — ~1 file, no new method, immediate UX win. Reuses existing
+   `RuleID`/severity data.
+2. **Add `textDocument/hover`** returning the rule explanation for the hovered
+   diagnostic, or a keyword/type glossary. Reuses `pkg/lint` rule metadata.
+3. **Add `documentSymbol`** listing `CREATE FUNCTION`/`PROCEDURE` signatures.
+   Reuses the existing CST header parse in `pkg/format`.
+4. **Add `willSaveWaitUntil`** to own format-on-save instead of relying on client
+   binding.
+
+> Do NOT: expand the LSP surface into a broad design (workspace features,
+   incremental parsing, custom `textDocument/*` extensions) as part of this issue.
+   Keep any change scoped to the above and evidence-backed by a `pkg/lsp` test
+   (see `server_test.go` for the framed-request/response harness).
+
+## Verification
+
+- `go build ./cmd/pgtidy` succeeds.
+- `go test ./...` passes (LSP unit tests in `pkg/lsp/server_test.go` exercise
+  `initialize`, formatting, range formatting, `didClose` diagnostics clearing).
+- Runtime e2e smoke test (framed JSON-RPC over stdio) confirmed `initialize`
+  capabilities, `COR001` diagnostics, and a formatting edit; `hover` correctly
+  returns `-32601`.

docs/todo.md

@@ -119,6 +119,7 @@ Legend: ✅ done · 🚧 in progress · ⬜ not started
 - `cmd/pgtidy/lsp.go`: `pgtidy lsp` subcommand; config discovered from cwd.
 - `editors/vscode/`: TS extension using `vscode-languageclient`; launches `pgtidy lsp` via stdio; `.pgsql` mapped to `sql` language; `pgtidy.path` / `pgtidy.enable` settings.
 - _Range formatting: future._
+- Full capability inventory, runtime-verified gaps, and ranked next steps in `docs/lsp-status.md` (issue #3).
 
 ## ✅ V4 — DataGrip
 - `editors/datagrip/`: Gradle-based JetBrains plugin targeting DataGrip 2024.3+ via LSP4IJ.

editors/datagrip/README.md

@@ -1,37 +1,27 @@
 # PgTidy — DataGrip / JetBrains Plugin
 
-Formats and lints SQL files in DataGrip (and any JetBrains IDE) via the LSP4IJ plugin.
+Formats SQL files in DataGrip (and any JetBrains IDE) by shelling out to the `pgtidy` binary.
 
 ## Requirements
 
 - `pgtidy` binary on `PATH` — download from [releases](https://git.warky.dev/wdevs/PgTidy/releases) or build with `go install git.warky.dev/wdevs/pgtidy/cmd/pgtidy@latest`
-- [LSP4IJ](https://plugins.jetbrains.com/plugin/23257-lsp4ij) plugin installed (free, by Red Hat)
 - DataGrip 2024.3+ (or any JetBrains IDE 2024.3+)
 
 ## Install the plugin
 
-### Option A — Install from disk (`.zip`)
-
 1. Build: `./gradlew buildPlugin` (output in `build/distributions/`)
 2. In DataGrip: **Settings → Plugins → ⚙ → Install Plugin from Disk…** → select the `.zip`
 3. Restart the IDE
 
-### Option B — Install LSP4IJ and configure manually (no plugin build needed)
-
-1. Install **LSP4IJ** from the marketplace (**Settings → Plugins → Marketplace → search "LSP4IJ"**)
-2. Go to **Settings → Language Servers → + (Add)**
-3. Fill in:
-   - **Name:** `PgTidy`
-   - **Command:** `pgtidy lsp`
-4. Under **Mappings**, add file patterns: `*.sql`, `*.pgsql`
-5. Click **OK** and restart the IDE
-
 ## Features
 
-- **Formatting** — `Code → Reformat Code` (`Ctrl+Alt+L`) formats the current SQL file
-- **Diagnostics** — lint findings shown as inspections (MIG001–MIG003, COR001–003, NAM001–003)
+- **Formatting** — `Format with PgTidy` (`Ctrl+Alt+Shift+P` or right-click in the editor) formats the current SQL file via `pgtidy fmt`
+- **Version check** — `Tools → Show PgTidy Version` and the status bar widget show the installed `pgtidy` version
+- **Diagnostics** — lint findings (MIG001–MIG003, COR001–003, NAM001–003) are shown inline as editor annotations, via a native `ExternalAnnotator` that runs `pgtidy lint --json`
 - **Quick fixes** — intention actions for MIG001 (add `CONCURRENTLY`) and MIG003 (add `NOT VALID`)
 
 ## Troubleshooting
 
-**Server not starting:** Check **View → Tool Windows → LSP4IJ Consoles** for stderr output. Most common cause: `pgtidy` not found on `PATH` — update the command to the full path, e.g. `/usr/local/bin/pgtidy lsp`.
+**"Cannot start pgtidy" error:** the `pgtidy` binary isn't on `PATH` for the IDE process. Install it or restart the IDE after adding it to `PATH`.
+
+**No diagnostics showing:** diagnostics only run on files with a `.sql` or `.pgsql` extension; check `pgtidy lint --json` runs cleanly on the file from a terminal with the same `PATH` as the IDE.

editors/datagrip/build.gradle.kts

@@ -23,7 +23,6 @@ dependencies {
         val platformType    = providers.gradleProperty("platformType")
         create(platformType, platformVersion)
         bundledPlugin("com.intellij.database")
-        plugin("com.redhat.devtools.lsp4ij:${providers.gradleProperty("lsp4ijVersion").get()}")
     }
 }
 

editors/datagrip/gradle.properties

@@ -5,4 +5,3 @@ pluginSinceBuild=243
 # DataGrip 2024.3
 platformVersion=2024.3
 platformType=DB
-lsp4ijVersion=0.20.1

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyDiagnostic.kt

@@ -0,0 +1,18 @@
+package com.pgtidy.datagrip
+
+data class PgTidyFix(
+    val offset: Int,
+    val end: Int,
+    val new: String,
+    val title: String,
+)
+
+data class PgTidyDiagnostic(
+    val ruleId: String,
+    val severity: String,
+    val message: String,
+    val file: String,
+    val line: Int,
+    val col: Int,
+    val fix: PgTidyFix?,
+)

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyExternalAnnotator.kt

@@ -0,0 +1,50 @@
+package com.pgtidy.datagrip
+
+import com.intellij.lang.annotation.AnnotationHolder
+import com.intellij.lang.annotation.ExternalAnnotator
+import com.intellij.lang.annotation.HighlightSeverity
+import com.intellij.openapi.editor.Document
+import com.intellij.psi.PsiFile
+
+class PgTidyExternalAnnotator : ExternalAnnotator<String, List<PgTidyDiagnostic>>() {
+
+    override fun collectInformation(file: PsiFile): String? {
+        val ext = file.virtualFile?.extension?.lowercase()
+        if (ext != "sql" && ext != "pgsql") return null
+        return file.viewProvider.document?.text
+    }
+
+    override fun doAnnotate(source: String): List<PgTidyDiagnostic> = PgTidyLintRunner.run(source)
+
+    override fun apply(file: PsiFile, diagnostics: List<PgTidyDiagnostic>, holder: AnnotationHolder) {
+        val document = file.viewProvider.document ?: return
+        for (d in diagnostics) {
+            if (d.line <= 0 || d.line > document.lineCount) continue
+            val range = lineTailRange(document, d.line, d.col) ?: continue
+
+            val severity = when (d.severity) {
+                "error" -> HighlightSeverity.ERROR
+                "warning" -> HighlightSeverity.WARNING
+                else -> HighlightSeverity.WEAK_WARNING
+            }
+
+            val builder = holder.newAnnotation(severity, "[${d.ruleId}] ${d.message}").range(range)
+            if (d.fix != null) {
+                builder.withFix(PgTidyQuickFix(d.fix))
+            }
+            builder.create()
+        }
+    }
+
+    /** Byte column [col] (1-based) to end-of-line, converted to a document TextRange. */
+    private fun lineTailRange(document: Document, line1Based: Int, col1Based: Int): com.intellij.openapi.util.TextRange? {
+        val lineIdx = line1Based - 1
+        val lineStart = document.getLineStartOffset(lineIdx)
+        val lineEnd = document.getLineEndOffset(lineIdx)
+        val lineText = document.getText(com.intellij.openapi.util.TextRange(lineStart, lineEnd))
+        val charCol = PgTidyOffsets.byteOffsetToCharIndex(lineText, col1Based - 1)
+        val start = lineStart + charCol
+        if (start >= lineEnd) return null
+        return com.intellij.openapi.util.TextRange(start, lineEnd)
+    }
+}

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyLintRunner.kt

@@ -0,0 +1,32 @@
+package com.pgtidy.datagrip
+
+import com.google.gson.Gson
+import com.google.gson.reflect.TypeToken
+
+object PgTidyLintRunner {
+    private val gson = Gson()
+    private val listType = TypeToken.getParameterized(List::class.java, PgTidyDiagnostic::class.java).type
+
+    /** Runs `pgtidy lint --json` over [source]. Returns an empty list if pgtidy is missing or output is unparseable. */
+    fun run(source: String): List<PgTidyDiagnostic> {
+        val proc = try {
+            ProcessBuilder("pgtidy", "lint", "--json")
+                .redirectErrorStream(false)
+                .start()
+        } catch (ex: Exception) {
+            return emptyList()
+        }
+
+        proc.outputStream.bufferedWriter().use { it.write(source) }
+        val output = proc.inputStream.bufferedReader().readText()
+        proc.errorStream.bufferedReader().readText()
+        proc.waitFor()
+
+        if (output.isBlank()) return emptyList()
+        return try {
+            gson.fromJson<List<PgTidyDiagnostic>>(output, listType) ?: emptyList()
+        } catch (ex: Exception) {
+            emptyList()
+        }
+    }
+}

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyOffsets.kt

@@ -0,0 +1,28 @@
+package com.pgtidy.datagrip
+
+/**
+ * pgtidy reports positions as UTF-8 byte offsets (Go strings are byte slices);
+ * IntelliJ documents index text as UTF-16 chars. Converts one to the other.
+ */
+object PgTidyOffsets {
+    fun byteOffsetToCharIndex(text: String, byteOffset: Int): Int {
+        var bytes = 0
+        var i = 0
+        while (i < text.length) {
+            val cp = text.codePointAt(i)
+            val charCount = Character.charCount(cp)
+            val byteLen = utf8Length(cp)
+            if (bytes + byteLen > byteOffset) return i
+            bytes += byteLen
+            i += charCount
+        }
+        return text.length
+    }
+
+    private fun utf8Length(codePoint: Int): Int = when {
+        codePoint <= 0x7F -> 1
+        codePoint <= 0x7FF -> 2
+        codePoint <= 0xFFFF -> 3
+        else -> 4
+    }
+}

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyQuickFix.kt

@@ -0,0 +1,25 @@
+package com.pgtidy.datagrip
+
+import com.intellij.codeInsight.intention.IntentionAction
+import com.intellij.openapi.editor.Editor
+import com.intellij.openapi.project.Project
+import com.intellij.psi.PsiFile
+import com.intellij.util.IncorrectOperationException
+
+class PgTidyQuickFix(private val fix: PgTidyFix) : IntentionAction {
+
+    override fun getText(): String = fix.title
+    override fun getFamilyName(): String = "PgTidy"
+    override fun startInWriteAction(): Boolean = true
+
+    override fun isAvailable(project: Project, editor: Editor?, file: PsiFile?): Boolean = editor != null
+
+    @Throws(IncorrectOperationException::class)
+    override fun invoke(project: Project, editor: Editor?, file: PsiFile?) {
+        val document = editor?.document ?: return
+        val text = document.text
+        val start = PgTidyOffsets.byteOffsetToCharIndex(text, fix.offset)
+        val end = PgTidyOffsets.byteOffsetToCharIndex(text, fix.end)
+        document.replaceString(start, end, fix.new)
+    }
+}

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyServerConnection.kt

@@ -1,9 +0,0 @@
-package com.pgtidy.datagrip
-
-import com.intellij.openapi.project.Project
-import com.redhat.devtools.lsp4ij.server.ProcessStreamConnectionProvider
-
-class PgTidyServerConnection(project: Project) : ProcessStreamConnectionProvider(
-    listOf("pgtidy", "lsp"),
-    project.basePath ?: System.getProperty("user.home"),
-)

editors/datagrip/src/main/kotlin/com/pgtidy/datagrip/PgTidyServerFactory.kt

@@ -1,10 +0,0 @@
-package com.pgtidy.datagrip
-
-import com.intellij.openapi.project.Project
-import com.redhat.devtools.lsp4ij.LanguageServerFactory
-import com.redhat.devtools.lsp4ij.server.StreamConnectionProvider
-
-class PgTidyServerFactory : LanguageServerFactory {
-    override fun createConnectionProvider(project: Project): StreamConnectionProvider =
-        PgTidyServerConnection(project)
-}

editors/datagrip/src/main/resources/META-INF/plugin.xml

@@ -5,13 +5,11 @@
     <vendor url="https://git.warky.dev/wdevs/PgTidy">Warky Devs</vendor>
     <description><![CDATA[
         PostgreSQL formatter and linter powered by <a href="https://github.com/hein/pgtidy">pgtidy</a>.<br/>
-        Requires the <code>pgtidy</code> binary on PATH and the
-        <a href="https://plugins.jetbrains.com/plugin/23257-lsp4ij">LSP4IJ</a> plugin.
+        Requires the <code>pgtidy</code> binary on PATH.
     ]]></description>
 
     <depends>com.intellij.modules.platform</depends>
     <depends>com.intellij.database</depends>
-    <depends>com.redhat.devtools.lsp4ij</depends>
 
     <actions>
         <action id="com.pgtidy.FormatDocument"
@@ -33,17 +31,7 @@
         <statusBarWidgetFactory id="com.pgtidy.StatusBarWidget"
                                  implementation="com.pgtidy.datagrip.PgTidyStatusBarWidgetFactory"
                                  order="last"/>
+        <externalAnnotator language="" implementationClass="com.pgtidy.datagrip.PgTidyExternalAnnotator"/>
     </extensions>
 
-    <extensions defaultExtensionNs="com.redhat.devtools.lsp4ij">
-        <server id="com.pgtidy.lsp"
-                name="PgTidy"
-                factoryClass="com.pgtidy.datagrip.PgTidyServerFactory">
-            <description>PostgreSQL formatter and linter (pgtidy lsp)</description>
-        </server>
-
-        <fileNamePatternMapping patterns="*.sql;*.pgsql"
-                                serverId="com.pgtidy.lsp"
-                                languageId="SQL"/>
-    </extensions>
 </idea-plugin>

linux/arch/PKGBUILD

@@ -1,6 +1,6 @@
 # Maintainer: Hein (Warky Devs) <hein@warky.dev>
 pkgname=pgtidy-bin
-pkgver=0.0.5
+pkgver=0.0.7
 pkgrel=1
 pkgdesc="PostgreSQL SQL formatter and linter"
 arch=('x86_64' 'aarch64')

linux/centos/pgtidy.spec

@@ -1,5 +1,5 @@
 Name:           pgtidy
-Version:        0.0.5
+Version:        0.0.7
 Release:        1%{?dist}
 Summary:        PostgreSQL SQL formatter and linter
 

pkg/diagnostics/diagnostics.go

@@ -15,27 +15,27 @@ const (
 // TextFix is a byte-range replacement that can be applied to the source SQL.
 // Replace src[Offset:End] with New. An insertion has Offset == End.
 type TextFix struct {
-	Offset int    // byte offset in source (inclusive)
-	End    int    // byte offset in source (exclusive)
-	New    string // replacement text
-	Title  string // short description shown in editor UI
+	Offset int    `json:"offset"` // byte offset in source (inclusive)
+	End    int    `json:"end"`    // byte offset in source (exclusive)
+	New    string `json:"new"`    // replacement text
+	Title  string `json:"title"`  // short description shown in editor UI
 }
 
 // Diagnostic is a single lint finding.
 type Diagnostic struct {
 	// RuleID is the stable identifier for the rule that produced this finding
 	// (e.g. "MIG001").
-	RuleID string
+	RuleID string `json:"ruleId"`
 	// Severity is the urgency level.
-	Severity Severity
+	Severity Severity `json:"severity"`
 	// Message is a human-readable description of the finding.
-	Message string
+	Message string `json:"message"`
 	// File is the path to the source file, or "" for stdin.
-	File string
+	File string `json:"file"`
 	// Line is the 1-based line number of the finding.
-	Line int
+	Line int `json:"line"`
 	// Col is the 1-based column number of the finding.
-	Col int
+	Col int `json:"col"`
 	// Fix is non-nil when an autofix is available for this diagnostic.
-	Fix *TextFix
+	Fix *TextFix `json:"fix,omitempty"`
 }

pkg/format/format_test.go

@@ -7,7 +7,6 @@ import (
 	"testing"
 
 	"git.warky.dev/wdevs/pgtidy/pkg/config"
-	"git.warky.dev/wdevs/pgtidy/pkg/lexer"
 	"git.warky.dev/wdevs/pgtidy/pkg/parser"
 )
 
@@ -199,48 +198,7 @@ func TestCorpusIdempotentAndSafe(t *testing.T) {
 	t.Logf("formatted %d corpus files (idempotent + semantically equal)", seen)
 }
 
-// semanticallyEqual compares the non-trivia token streams of two sources,
-// treating unquoted identifiers/keywords case-insensitively and everything
-// else (strings, numbers, operators, punctuation) exactly. Dollar-quoted body
-// tokens are compared recursively so body whitespace normalization does not
-// trigger a false failure.
+// semanticallyEqual is a test-local alias for the exported safety check.
 func semanticallyEqual(a, b string) bool {
-	ta := significant(a)
-	tb := significant(b)
-	if len(ta) != len(tb) {
-		return false
-	}
-	for i := range ta {
-		if ta[i].Kind != tb[i].Kind {
-			return false
-		}
-		switch ta[i].Kind {
-		case lexer.Ident:
-			if !strings.EqualFold(ta[i].Text, tb[i].Text) {
-				return false
-			}
-		case lexer.DollarString:
-			_, innerA, _, okA := splitDollarQuote(ta[i].Text)
-			_, innerB, _, okB := splitDollarQuote(tb[i].Text)
-			if okA != okB || (okA && !semanticallyEqual(innerA, innerB)) {
-				return false
-			}
-		default:
-			if ta[i].Text != tb[i].Text {
-				return false
-			}
-		}
-	}
-	return true
-}
-
-func significant(src string) []lexer.Token {
-	var out []lexer.Token
-	for _, t := range lexer.Lex(src) {
-		if t.Kind == lexer.EOF || t.IsTrivia() {
-			continue
-		}
-		out = append(out, t)
-	}
-	return out
+	return SemanticallyEqual(a, b)
 }

pkg/format/safety.go

@@ -0,0 +1,61 @@
+package format
+
+import (
+	"strings"
+
+	"git.warky.dev/wdevs/pgtidy/pkg/lexer"
+)
+
+// SemanticallyEqual reports whether a and b have the same non-trivia token
+// stream, i.e. formatting may only ever change whitespace/comment trivia and
+// layout — it must never add, remove, or alter a token of actual code.
+// Unquoted identifiers/keywords compare case-insensitively (casing is a
+// style choice); everything else (strings, numbers, operators, punctuation)
+// must match exactly. Dollar-quoted body tokens are compared recursively so
+// that independent body reformatting doesn't trigger a false failure.
+//
+// The CLI and LSP must call this before ever writing or emitting formatted
+// output: if it returns false, the formatter has a bug and the original
+// source must be kept, never the (corrupting) formatted output.
+func SemanticallyEqual(a, b string) bool {
+	ta := significantTokens(a)
+	tb := significantTokens(b)
+	if len(ta) != len(tb) {
+		return false
+	}
+	for i := range ta {
+		if ta[i].Kind != tb[i].Kind {
+			return false
+		}
+		switch ta[i].Kind {
+		case lexer.Ident:
+			if !strings.EqualFold(ta[i].Text, tb[i].Text) {
+				return false
+			}
+		case lexer.DollarString:
+			_, innerA, _, okA := splitDollarQuote(ta[i].Text)
+			_, innerB, _, okB := splitDollarQuote(tb[i].Text)
+			if okA != okB || (okA && !SemanticallyEqual(innerA, innerB)) {
+				return false
+			}
+		default:
+			if ta[i].Text != tb[i].Text {
+				return false
+			}
+		}
+	}
+	return true
+}
+
+// significantTokens lexes src and returns its tokens excluding EOF and trivia
+// (whitespace/comments).
+func significantTokens(src string) []lexer.Token {
+	var out []lexer.Token
+	for _, t := range lexer.Lex(src) {
+		if t.Kind == lexer.EOF || t.IsTrivia() {
+			continue
+		}
+		out = append(out, t)
+	}
+	return out
+}

pkg/lsp/server.go

@@ -124,6 +124,10 @@ func (s *server) handle(raw []byte) bool {
 			s.reply(req.ID, []textEdit{})
 			return false
 		}
+		if !format.SemanticallyEqual(text, formatted) {
+			s.reply(req.ID, []textEdit{})
+			return false
+		}
 		s.reply(req.ID, []textEdit{fullReplace(text, formatted)})
 	case "textDocument/rangeFormatting":
 		var p rangeFormattingParams
@@ -224,6 +228,9 @@ func (s *server) rangeFormat(text string, r lspRange) []textEdit {
 	if formatted == text {
 		return nil
 	}
+	if !format.SemanticallyEqual(text, formatted) {
+		return nil
+	}
 
 	// Split both versions into lines, keeping the trailing newline attached to
 	// each element so that joining them reconstructs the original string.