build: remove DataFusion from default builds

.cargo/config.toml

@@ -3,9 +3,7 @@ pr-check = "run -q -p xtask -- pr-check"
 licenses = "run -q -p xtask -- licenses"
 xtask = "run -q -p xtask --"
 
-# The app and CLI default to the reference table executor, so plain
-# `cargo check -p plotx` and `cargo run -p plotx` already are the fast path and
-# need no alias. Shipping builds must opt into DataFusion, and the release
-# workflow calls this alias instead of spelling the feature out, so the flag
-# lives in one version-controlled place.
-release-build = "build --locked --release -p plotx --features datafusion"
+# Development and shipping builds use the reference table executor. DataFusion
+# remains available through a separate, explicit maintenance build.
+release-build = "build --locked --release -p plotx"
+datafusion-build = "build --locked --release -p plotx --features datafusion"

.github/workflows/ci.yml

@@ -118,12 +118,6 @@ jobs:
     steps:
       - uses: actions/checkout@v7
 
-      # `--all-features` builds plotx-substrait, which generates code with protoc.
-      - name: Install protoc
-        uses: arduino/setup-protoc@v3
-        with:
-          repo-token: ${{ secrets.GITHUB_TOKEN }}
-
       - name: Update Rust toolchain
         run: |
           rustup update --no-self-update stable
@@ -154,11 +148,6 @@ jobs:
     steps:
       - uses: actions/checkout@v7
 
-      - name: Install protoc
-        uses: arduino/setup-protoc@v3
-        with:
-          repo-token: ${{ secrets.GITHUB_TOKEN }}
-
       - name: Update Rust toolchain
         run: |
           rustup update --no-self-update stable
@@ -169,7 +158,7 @@ jobs:
           save-if: ${{ github.ref == 'refs/heads/main' }}
           cache-on-failure: true
 
-      - name: Run the test suite in both backend configurations
+      - name: Run the reference-backend test suite
         run: cargo xtask pr-check test
 
   # Validates that the user manual builds. Deployment is separate: Cloudflare

AGENTS.md

@@ -58,16 +58,17 @@ These instructions apply to the entire repository.
 - `plotx-data` stays engine-free: never add datafusion, datafusion-substrait or
   sqlparser to it. Backend adapters reach its internals through the
   `#[doc(hidden)] pub` seams; keep that set minimal instead of widening the API.
-- The app and CLI default to the reference table executor. DataFusion is an
-  opt-in `datafusion` feature, and releases are built with `cargo release-build`
-  so the flag lives in one place. Keep both configurations compiling.
+- The app, CLI, and shipping builds use the reference table executor. DataFusion
+  remains available through the opt-in `datafusion` feature and the
+  `cargo datafusion-build` maintenance alias, but default PR checks and CI do
+  not compile or validate DataFusion or Substrait.
 - Keep Rust source files below the repository's 800-line limit; prefer cohesive
   modules over large files. Extract tests to a sibling `*_tests.rs` referenced
   with `#[path]` when a file approaches it.
 - Run `cargo pr-check` before completing code changes. It checks formatting,
   source sizes, dependency licenses and advisories, a default-configuration
-  build of both frontends, Clippy with warnings denied, and the test suite in
-  both backend configurations. It requires `cargo-deny`.
+  build of both frontends, Clippy with warnings denied, and the reference-backend
+  test suite. It requires `cargo-deny`.
 
 ## Documentation
 

CONTRIBUTING.md

@@ -20,23 +20,25 @@ long wait — use the narrowest one that answers your question:
 | Day-to-day development: build and run the app | `cargo run -p plotx` |
 | Test the crate you changed | `cargo test -p plotx-core` |
 | Optimized build for performance work | `cargo build --release -p plotx` |
-| Shipping configuration, adds the DataFusion table backend | `cargo release-build` |
+| Shipping configuration, uses the reference table executor | `cargo release-build` |
+| Explicit DataFusion maintenance build | `cargo datafusion-build` |
 | The same checks as CI, before a pull request | `cargo pr-check` |
 
 Two things that are easy to trip over:
 
 - Development builds keep the scientific processing crates optimized so they
   stay usable under a debugger, but the UI and renderer are not. Judge frame
   rate or rendering performance only with an optimized build.
-- The app and CLI default to an in-memory reference table executor.
-  Shipping builds enable the DataFusion backend via `cargo release-build`;
-  the About window shows which engine is active.
+- The app, CLI, and shipping build use the in-memory reference table executor.
+  DataFusion and Substrait remain in the workspace for explicit maintenance,
+  but `cargo pr-check` and CI do not build or validate them.
 
-`cargo pr-check` requires `protoc` and `cargo-deny`
-(`cargo install --locked cargo-deny`); nothing else does. CI runs exactly the
+`cargo pr-check` requires `cargo-deny`
+(`cargo install --locked cargo-deny`). CI runs exactly the
 same steps, split into parallel jobs (`cargo xtask pr-check quick|lint|test`).
 `cargo licenses` (requires `cargo-about`) regenerates
-`dist/THIRD-PARTY-LICENSES.html`, the report of bundled third-party licenses.
+`dist/THIRD-PARTY-LICENSES.html` from the reference-only Windows shipping app's
+dependency graph.
 
 The user manual is an Astro/Starlight site in `docs/` (`npm run build` to
 validate). UI screenshots come from the built-in harness: point `PLOTX_SHOT`

Cargo.toml

@@ -14,12 +14,10 @@ members = [
     "crates/app",
     "xtask",
 ]
-# Keep the Substrait exchange crate out of bare root builds: it is the only
-# member that requires protoc, prost, and pbjson-types. CI and `cargo pr-check`
-# use `--workspace`, so they still cover every crate.
+# DataFusion and Substrait remain workspace members for explicit maintenance,
+# but bare root builds stay on the lightweight reference executor.
 default-members = [
     "crates/data",
-    "crates/datafusion",
     "crates/io",
     "crates/analysis",
     "crates/processing",

README.md

@@ -45,17 +45,14 @@ cd plotx
 cargo run --release -p plotx
 ```
 
-That is the fast development build. It uses the built-in reference table
-executor, which keeps whole tables in memory and cannot spill, so large-table
-work needs the shipping engine instead:
+Development and shipping builds use the built-in reference table executor:
 
 ```sh
-cargo release-build          # the desktop app with the DataFusion backend
+cargo release-build          # the shipping desktop app
 ```
 
-Releases are produced with `cargo release-build`. The About window always
-names the active table engine, so you can tell which kind of build you are
-running.
+The dormant DataFusion backend remains available for explicit maintenance with
+`cargo datafusion-build`, but it is not part of default builds or releases.
 
 The development workflow and pre-submission checks are described in
 [CONTRIBUTING.md](CONTRIBUTING.md).

crates/app/Cargo.toml

@@ -11,8 +11,8 @@ name = "plotx"
 path = "src/main.rs"
 
 [features]
-# Default builds use the reference executor. Release builds enable DataFusion
-# through `cargo release-build`; the About window shows the active executor.
+# Development and shipping builds use the reference executor. This feature is
+# retained as an explicit maintenance switch for the dormant DataFusion backend.
 default = []
 datafusion = ["plotx-core/datafusion"]
 

crates/app/src/ui/canvas/mod.rs

@@ -414,13 +414,6 @@ fn welcome_page(app: &mut PlotxApp, ui: &mut Ui) {
                     egui::RichText::new("Scientific data analysis and figure preparation")
                         .color(ui.visuals().text_color()),
                 );
-                if !cfg!(feature = "datafusion") {
-                    ui.add_space(8.0);
-                    ui.label(
-                        egui::RichText::new("Development build: large tables may run more slowly.")
-                            .color(Color32::from_rgb(0xE0, 0x6C, 0x22)),
-                    );
-                }
                 ui.add_space(22.0);
                 welcome_start(app, ui);
             },

crates/cli/Cargo.toml

@@ -11,8 +11,8 @@ name = "plotx-cli"
 path = "src/main.rs"
 
 [features]
-# Default builds use the reference executor. Enable this feature for
-# spill-capable DataFusion execution on large tables.
+# Default builds use the reference executor. This feature is retained as an
+# explicit maintenance switch for the dormant DataFusion backend.
 default = []
 datafusion = ["plotx-core/datafusion"]
 

deny.toml

@@ -2,7 +2,7 @@
 # Keep the license allow-list aligned with xtask/about.toml.
 
 [graph]
-all-features = true
+all-features = false
 # Every platform CI builds on. Releases currently ship for 64-bit Windows
 # only, but the dependency policy must cover whatever CI compiles.
 targets = [

xtask/src/main.rs

@@ -82,7 +82,7 @@ fn pr_check(stage: Option<PrCheckStage>) -> Result<(), String> {
     use PrCheckStage::{Lint, Quick, Test};
     use PrCheckStep::{Cargo, DependencyPolicy, RustFileSizes};
 
-    let steps: [(PrCheckStage, PrCheckStep); 7] = [
+    let steps: [(PrCheckStage, PrCheckStep); 6] = [
         (
             Quick,
             Cargo {
@@ -117,8 +117,11 @@ fn pr_check(stage: Option<PrCheckStage>) -> Result<(), String> {
                 args: &[
                     "clippy",
                     "--workspace",
+                    "--exclude",
+                    "plotx-datafusion",
+                    "--exclude",
+                    "plotx-substrait",
                     "--all-targets",
-                    "--all-features",
                     "--locked",
                     "--quiet",
                     "--",
@@ -134,26 +137,10 @@ fn pr_check(stage: Option<PrCheckStage>) -> Result<(), String> {
                 args: &[
                     "test",
                     "--workspace",
-                    "--all-features",
-                    "--locked",
-                    "--profile",
-                    "pr-check",
-                    "--quiet",
-                ],
-            },
-        ),
-        // `--all-features` always enables plotx-core/datafusion, so the
-        // reference executor that default builds depend on is never exercised
-        // above.
-        (
-            Test,
-            Cargo {
-                name: "test (reference backend)",
-                args: &[
-                    "test",
-                    "-p",
-                    "plotx-core",
-                    "--no-default-features",
+                    "--exclude",
+                    "plotx-datafusion",
+                    "--exclude",
+                    "plotx-substrait",
                     "--locked",
                     "--profile",
                     "pr-check",
@@ -207,6 +194,12 @@ fn licenses() -> Result<(), String> {
     let about_args = [
         "about",
         "generate",
+        "--manifest-path",
+        "crates/app/Cargo.toml",
+        "--no-default-features",
+        "--target",
+        "x86_64-pc-windows-msvc",
+        "--locked",
         "-c",
         "xtask/about.toml",
         "xtask/about.hbs",
@@ -289,7 +282,16 @@ fn cargo_output(repo_root: &Path, args: &[&str]) -> Result<Output, String> {
 }
 
 fn run_cargo_deny_step(repo_root: &Path, index: usize, total: usize) -> Result<(), String> {
-    let args = ["deny", "--locked", "check"];
+    let args = [
+        "deny",
+        "--locked",
+        "--no-default-features",
+        "--exclude",
+        "plotx-datafusion",
+        "--exclude",
+        "plotx-substrait",
+        "check",
+    ];
     let started_at = Instant::now();
     print_step(index, total, "dependency policy")?;
     let output = cargo_output(repo_root, &args)?;