feat(native): make native discovery the default compiler engine (#1003)

* feat(native): reproduce each supported Python's text semantics natively

Native discovery deferred to Python whenever the running Python's Unicode
data differed from the one table it carried (CPython 3.12), so 3.13 and
3.14 never ran natively. Generate a `str.isalnum()` table for the Unicode
version of each supported CPython (15.0.0, 15.1.0, 16.0.0), select it by
`unicodedata.unidata_version` per call, and reproduce `inspect.cleandoc`
per Python minor version (3.13 strips spaces only). Unknown Python or
Unicode versions still defer to Python.

* feat(native): make native discovery the default compiler engine

With SQLBUILD_COMPILER_ENGINE unset, compiles now discover project inputs
natively. The differential harness is byte-identical on the per-PR corpus
(on Python 3.12, 3.13 and 3.14), the dense corpus and a private real
project, and cold, edit and no-change compiles are no slower.
SQLBUILD_COMPILER_ENGINE=python keeps the Python discovery as the oracle,
and CI keeps comparing both engines.

Compiler and Rules stores stay namespaced by engine, so the first compile
after upgrading is cold once; semantic fingerprints do not change.
Tests that hardcoded the Python engine's cache paths now resolve the
active engine's directories, and an E2E test proves the unset default
runs native discovery with output identical to the Python engine.

* fix(native): name engine-namespaced cache paths and test native discovery on Python 3.13 and 3.14

Docs now name the native engine's compile reuse and Rules cache
directories, the unused Python-engine reuse path constant is removed, and
a CI job runs the native text oracles and a bounded engine differential
on Python 3.13 and 3.14.

* ci: generate the base benchmark project with the base checkout's own environment

* test(lineage): read the structural cache from the active engine's directory

Author: kvlonge

.github/workflows/verify.yml

@@ -296,6 +296,50 @@ jobs:
           COMPILER_DIFFERENTIAL_ARGS="--stage-captures --require-discovery-coverage
           --require-render-coverage"
 
+  compiler_differential_python_versions:
+    name: Native discovery on Python ${{ matrix.python }}
+    needs: [changes, lockfile]
+    if: needs.changes.outputs.compiler_differential == 'true'
+    runs-on: ubuntu-24.04
+    timeout-minutes: 20
+    strategy:
+      fail-fast: false
+      matrix:
+        python: ["3.13", "3.14"]
+    env:
+      UV_PYTHON: ${{ matrix.python }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v5
+        with:
+          version: "0.12.3"
+          enable-cache: true
+          cache-dependency-glob: |
+            uv.lock
+            Cargo.lock
+      - uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: stable
+      - name: Install Python
+        run: uv python install ${{ matrix.python }}
+      - name: Install dependencies
+        run: uv sync --locked --all-groups --all-extras --python ${{ matrix.python }}
+      - name: Require native discovery to support this Python
+        run: >-
+          uv run --no-sync python -c "import sys, unicodedata, sqlbuild._native as n;
+          assert n.native_text_supported(sys.version_info[:2], unicodedata.unidata_version)"
+      - name: Compare native text helpers with this Python
+        run: >-
+          uv run --no-sync pytest
+          tests/integration/src/sqlbuild/compiler/test_native_python_compat_oracle.py
+          -n auto --dist loadfile
+      - name: Compare engines on fixtures, examples, the failure corpus and a few seeds
+        env:
+          PYTHONUNBUFFERED: "1"
+        run: >-
+          uv run --no-sync python -m scripts.run_compiler_differential
+          --corpus fixtures examples failures seeds --seeds 4 --jobs 4
+
   format_performance:
     name: Formatter performance guards
     needs: [changes, lockfile]
@@ -488,7 +532,7 @@ jobs:
         run: uv sync --locked --all-groups --all-extras
       - name: Install base dependencies
         working-directory: base
-        run: uv sync --locked --all-extras
+        run: uv sync --locked --all-groups --all-extras
       - name: Compare base and head cold, warm and one-model edit compiles
         env:
           KIND: ${{ matrix.kind }}

crates/sqlbuild-core/src/text/_helpers/alnum_tables.rs

@@ -0,0 +1,8 @@
+//! The generated `str.isalnum()` tables, one per Unicode version of a supported CPython.
+
+/// `(unicodedata.unidata_version, ranges)` for every table native text helpers can select.
+pub(crate) const ALNUM_TABLES: [(&str, &[(u32, u32)]); 3] = [
+    ("15.0.0", include!("../alnum_ranges/unicode_15_0_0.in")),
+    ("15.1.0", include!("../alnum_ranges/unicode_15_1_0.in")),
+    ("16.0.0", include!("../alnum_ranges/unicode_16_0_0.in")),
+];

crates/sqlbuild-core/src/text/_helpers/mod.rs

@@ -1,3 +1,3 @@
-pub(crate) mod alnum_ranges;
+pub(crate) mod alnum_tables;
 pub(crate) mod cleandoc;
 pub(crate) mod sequence_matcher;

…d-core/src/text/_helpers/alnum_ranges.rs …/src/text/alnum_ranges/unicode_15_0_0.incrates/sqlbuild-core/src/text/_helpers/alnum_ranges.rs renamed to crates/sqlbuild-core/src/text/alnum_ranges/unicode_15_0_0.in crates/sqlbuild-core/src/text/_helpers/alnum_ranges.rs renamed to crates/sqlbuild-core/src/text/alnum_ranges/unicode_15_0_0.in

@@ -1,10 +1,6 @@
-//! Generated by `scripts/generate_python_char_table.py`; do not edit.
-
-/// The Unicode version of the Python that generated the table.
-pub(crate) const PYTHON_ALNUM_UNICODE_VERSION: &str = "15.0.0";
-
-/// Inclusive code point ranges where Python's `str.isalnum()` is true, in order.
-pub(crate) const PYTHON_ALNUM_RANGES: &[(u32, u32)] = &[
+// Generated by `scripts/generate_python_char_table.py`; do not edit.
+// Inclusive code point ranges where `str.isalnum()` is true in Unicode 15.0.0, in order.
+&[
     (0x0030, 0x0039),
     (0x0041, 0x005A),
     (0x0061, 0x007A),
@@ -752,4 +748,4 @@ pub(crate) const PYTHON_ALNUM_RANGES: &[(u32, u32)] = &[
     (0x2F800, 0x2FA1D),
     (0x30000, 0x3134A),
     (0x31350, 0x323AF),
-];
+]