feat(BREAKING): replace sqlformat with our own SQL formatter

.github/workflows/ci.generated.yml

@@ -45,6 +45,15 @@ jobs:
       - name: Test debug
         if: matrix.config.kind == 'test_debug'
         run: cargo test
+      - name: Test tracing
+        if: matrix.config.kind == 'test_debug'
+        run: cargo test --features tracing
+      - name: Clippy
+        if: matrix.config.kind == 'test_debug'
+        run: cargo clippy --all-targets -- -D warnings
+      - name: Check formatting
+        if: matrix.config.kind == 'test_debug'
+        run: cargo fmt --check
       - name: Test release
         if: matrix.config.kind == 'test_release'
         run: cargo test --release
@@ -90,4 +99,3 @@ jobs:
           ./.github/workflows/ci.ts --lint
           ./.github/workflows/publish_npm.ts --lint
           ./.github/workflows/release.ts --lint
-          ./.github/workflows/check_updates.ts --lint

.github/workflows/ci.ts

@@ -63,6 +63,9 @@ const buildJob = job("build", {
     },
 
     { name: "Test debug", if: isDebug, run: "cargo test" },
+    { name: "Test tracing", if: isDebug, run: "cargo test --features tracing" },
+    { name: "Clippy", if: isDebug, run: "cargo clippy --all-targets -- -D warnings" },
+    { name: "Check formatting", if: isDebug, run: "cargo fmt --check" },
     { name: "Test release", if: isRelease, run: "cargo test --release" },
 
     getTagVersion,
@@ -122,7 +125,6 @@ const buildJob = job("build", {
         "./.github/workflows/ci.ts --lint",
         "./.github/workflows/publish_npm.ts --lint",
         "./.github/workflows/release.ts --lint",
-        "./.github/workflows/check_updates.ts --lint",
       ],
     },
   ],

CLAUDE.md

@@ -1,32 +1,45 @@
 # dprint-plugin-sql
 
-Wrapper around [sqlformat-rs](https://github.com/shssoichiro/sqlformat-rs) for use as a dprint formatting plugin.
+SQL formatting plugin for dprint, with its own lexer, parser and formatter that build dprint-core's IR.
 
 ## Commands
 
 - `cargo test` — run all tests
+- `cargo test --test specs` — run only the spec tests
+- `cargo clippy --all-targets --all-features -- -D warnings` — lint
 - `cargo build --target wasm32-unknown-unknown --features wasm --release` — build the wasm plugin
 - `cd deployment/npm && node setup.js && npm test` — test the npm package (requires wasm build first)
 
 ## Architecture
 
-- `src/format_text.rs` — core formatting logic, delegates to `sqlformat::format()`
+The pipeline is lex → parse → generate IR → print → verify.
+
+- `src/lexer.rs` — splits the text into tokens, losslessly. This is where the dialects differ (`#`, `[..]`, backslash escapes, dollar quotes, `GO`, `DELIMITER`, client commands)
+- `src/parser.rs` — builds statements, procedural blocks (`BEGIN .. END`, `IF`, loops, `CASE` statements, declaration sections), brackets and `CASE` expressions. Also splits T-SQL statements written without semicolons
+- `src/ast.rs` — the tree the parser builds
+- `src/generation/generate.rs` — the layout: splits statements into clauses, lists and conditions, and generates the IR
+- `src/keywords.rs` — sorted tables of keywords
+- `src/format_text.rs` — formatting a file, and the verification that formatting changed nothing but whitespace (keyword casing aside), which errors rather than return changed code
+- `src/format_range.rs` — formatting the top level statements a range touches
 - `src/wasm_plugin.rs` — dprint wasm plugin interface (compiled only for wasm32 target with `wasm` feature)
 - `src/configuration/` — config types, resolution, and builder
 - `deployment/npm/` — npm package that ships `plugin.wasm`
 - `deployment/schema.json` — JSON schema for plugin configuration
-- `scripts/update.ts` — Deno script that checks crates.io for new sqlformat versions and auto-publishes
 - `scripts/generateReleaseNotes.ts` — generates changelog for GitHub releases
 
-## Adding a new config option
+## Principles
 
-When sqlformat-rs adds new options:
+- Formatting only ever changes the whitespace between tokens. Anything the parser doesn't understand is written as a run of tokens rather than reported, since the dialect it's in may use those words differently
+- When the dialects disagree on what some text is, the configured dialect decides. The generic dialect must pick what's least likely to be wrong, since a wrong guess there is the most likely kind of bug
+- Formatting must be stable: formatting the output again gives the same output. Every spec is formatted twice to check this, and `tests/test.rs` also formats every spec's input in every dialect at two widths and checks it's stable and changes nothing but whitespace
+
+## Adding a new config option
 
 1. Add the field to `Configuration` in `src/configuration/configuration.rs` (add enums with `generate_str_to_from!` if needed)
 2. Add resolution in `src/configuration/resolve_config.rs` using `get_value()`
 3. Add a builder method in `src/configuration/builder.rs` and update the `check_all_values_set` test
-4. Wire it through in `src/format_text.rs` when constructing `FormatOptions`
-5. Add the property to `deployment/schema.json`
+4. Use it in `src/generation/generate.rs`
+5. Add the property to `deployment/schema.json` and the table in `README.md`
 6. Add a spec test file in `tests/specs/Config/`
 
 ## Test specs
@@ -36,10 +49,11 @@ Spec files live in `tests/specs/` and use the dprint-development format:
 - `== message ==` starts a test case
 - `[expect]` separates input from expected output
 - `~~ key: value, key: value ~~` at the top of a file sets config for all specs in that file
-- Each config variation needs its own file since config is file-level
+- Each config variation needs its own file since config is file-level, which is why each dialect has its own directory
+- `[|` and `|]` mark the range to format in a spec's text
+- A spec file can't start with a SQL comment, since a first line starting with `--` names the spec's file
 
 ## Releases
 
 - Manual: trigger the `release` workflow with patch/minor
-- Automatic: `check_updates` workflow runs daily, detects new sqlformat versions, and auto-publishes
 - On tag push: CI creates a GitHub release with `plugin.wasm`, and `publish_npm` publishes to npm

Cargo.toml

@@ -7,7 +7,7 @@ homepage = "https://github.com/dprint/dprint-plugin-sql"
 keywords = ["formatting", "formatter", "sql"]
 license = "MIT"
 repository = "https://github.com/dprint/dprint-plugin-sql"
-description = "SQL formatter for dprint via sqlformat-rs."
+description = "SQL formatter for dprint."
 
 [lib]
 crate-type = ["lib", "cdylib"]
@@ -22,13 +22,19 @@ panic = "abort"
 
 [features]
 wasm = ["serde_json", "dprint-core/wasm"]
+tracing = ["dprint-core/tracing"]
+
+[[test]]
+name = "specs"
+path = "tests/spec_test.rs"
+harness = false
 
 [dependencies]
-dprint-core = { version = "0.69.1", default-features = false }
+dprint-core = { version = "0.71.0", features = ["formatting"], default-features = false }
 serde = { version = "1.0.88", features = ["derive"] }
 serde_json = { version = "1.0", optional = true }
-sqlformat = "0.5.0"
+thiserror = "2"
 
 [dev-dependencies]
-dprint-development = "0.11.0"
+dprint-development = "0.12.1"
 serde_json = { version = "1.0" }