Fix flake.

.github/workflows/libc.yml

@@ -0,0 +1,23 @@
+name: Benchmark libc
+
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    strategy:
+      matrix:
+        os: [ubuntu-24.04, ubuntu-24.04-arm, macos-13, macos-15]
+    runs-on: ${{ matrix.os }}
+
+    steps:
+    - uses: actions/checkout@v4
+    - uses: actions/setup-go@v5
+      with: { go-version: stable }
+
+    - name: Benchmark
+      shell: bash
+      run: sqlite3/libc/benchmark.sh

.github/workflows/repro.sh

@@ -1,28 +1,9 @@
 #!/usr/bin/env bash
 set -euo pipefail
 
-if [[ "$OSTYPE" == "linux"* ]]; then
-  WASI_SDK="https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-x86_64-linux.tar.gz"
-  BINARYEN="https://github.com/WebAssembly/binaryen/releases/download/version_123/binaryen-version_123-x86_64-linux.tar.gz"
-elif [[ "$OSTYPE" == "darwin"* ]]; then
-  WASI_SDK="https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-arm64-macos.tar.gz"
-  BINARYEN="https://github.com/WebAssembly/binaryen/releases/download/version_123/binaryen-version_123-arm64-macos.tar.gz"
-elif [[ "$OSTYPE" == "msys" || "$OSTYPE" == "cygwin" ]]; then
-  WASI_SDK="https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-x86_64-windows.tar.gz"
-  BINARYEN="https://github.com/WebAssembly/binaryen/releases/download/version_123/binaryen-version_123-x86_64-windows.tar.gz"
-fi
-
-# Download tools
-mkdir -p tools/
-[ -d "tools/wasi-sdk" ] || curl -#L "$WASI_SDK" | tar xzC tools &
-[ -d "tools/binaryen" ] || curl -#L "$BINARYEN" | tar xzC tools &
-wait
-
-[ -d "tools/wasi-sdk" ] || mv "tools/wasi-sdk"* "tools/wasi-sdk"
-[ -d "tools/binaryen" ] || mv "tools/binaryen"* "tools/binaryen"
-
 # Download and build SQLite
 sqlite3/download.sh
+sqlite3/tools.sh
 embed/build.sh
 embed/bcw2/build.sh
 

.github/workflows/test.yml

@@ -17,12 +17,17 @@ on:
       - '**.yml'
   workflow_dispatch:
 
+permissions:
+  contents: read
+
 jobs:
   test:
     strategy:
       matrix:
         os: [macos-latest, ubuntu-latest, windows-latest]
     runs-on: ${{ matrix.os }}
+    permissions:
+      contents: write
 
     steps:
     - uses: actions/checkout@v4

driver/driver.go

@@ -241,8 +241,9 @@ func (n *connector) Connect(ctx context.Context) (ret driver.Conn, err error) {
 		}
 	}()
 
-	old := c.Conn.SetInterrupt(ctx)
-	defer c.Conn.SetInterrupt(old)
+	if old := c.Conn.SetInterrupt(ctx); old != ctx {
+		defer c.Conn.SetInterrupt(old)
+	}
 
 	if !n.pragmas {
 		err = c.Conn.BusyTimeout(time.Minute)
@@ -362,8 +363,9 @@ func (c *conn) BeginTx(ctx context.Context, opts driver.TxOptions) (driver.Tx, e
 		c.txReset = `; PRAGMA query_only=` + string(c.readOnly)
 	}
 
-	old := c.Conn.SetInterrupt(ctx)
-	defer c.Conn.SetInterrupt(old)
+	if old := c.Conn.SetInterrupt(ctx); old != ctx {
+		defer c.Conn.SetInterrupt(old)
+	}
 
 	err := c.Conn.Exec(txBegin)
 	if err != nil {
@@ -382,8 +384,10 @@ func (c *conn) Commit() error {
 
 func (c *conn) Rollback() error {
 	// ROLLBACK even if interrupted.
-	old := c.Conn.SetInterrupt(context.Background())
-	defer c.Conn.SetInterrupt(old)
+	ctx := context.Background()
+	if old := c.Conn.SetInterrupt(ctx); old != ctx {
+		defer c.Conn.SetInterrupt(old)
+	}
 	return c.Conn.Exec(`ROLLBACK` + c.txReset)
 }
 
@@ -393,8 +397,9 @@ func (c *conn) Prepare(query string) (driver.Stmt, error) {
 }
 
 func (c *conn) PrepareContext(ctx context.Context, query string) (driver.Stmt, error) {
-	old := c.Conn.SetInterrupt(ctx)
-	defer c.Conn.SetInterrupt(old)
+	if old := c.Conn.SetInterrupt(ctx); old != ctx {
+		defer c.Conn.SetInterrupt(old)
+	}
 
 	s, tail, err := c.Conn.Prepare(query)
 	if err != nil {
@@ -419,8 +424,9 @@ func (c *conn) ExecContext(ctx context.Context, query string, args []driver.Name
 		return resultRowsAffected(0), nil
 	}
 
-	old := c.Conn.SetInterrupt(ctx)
-	defer c.Conn.SetInterrupt(old)
+	if old := c.Conn.SetInterrupt(ctx); old != ctx {
+		defer c.Conn.SetInterrupt(old)
+	}
 
 	err := c.Conn.Exec(query)
 	if err != nil {
@@ -483,8 +489,10 @@ func (s *stmt) ExecContext(ctx context.Context, args []driver.NamedValue) (drive
 		return nil, err
 	}
 
-	old := s.Stmt.Conn().SetInterrupt(ctx)
-	defer s.Stmt.Conn().SetInterrupt(old)
+	c := s.Stmt.Conn()
+	if old := c.SetInterrupt(ctx); old != ctx {
+		defer c.SetInterrupt(old)
+	}
 
 	err = errors.Join(
 		s.Stmt.Exec(),
@@ -493,7 +501,7 @@ func (s *stmt) ExecContext(ctx context.Context, args []driver.NamedValue) (drive
 		return nil, err
 	}
 
-	return newResult(s.Stmt.Conn()), nil
+	return newResult(c), nil
 }
 
 func (s *stmt) QueryContext(ctx context.Context, args []driver.NamedValue) (driver.Rows, error) {
@@ -678,13 +686,14 @@ func (r *rows) scanType(index int) scantype {
 
 func (r *rows) loadColumnMetadata() {
 	if r.nulls == nil {
+		c := r.Stmt.Conn()
 		count := r.Stmt.ColumnCount()
 		nulls := make([]bool, count)
 		types := make([]string, count)
 		scans := make([]scantype, count)
 		for i := range nulls {
 			if col := r.Stmt.ColumnOriginName(i); col != "" {
-				types[i], _, nulls[i], _, _, _ = r.Stmt.Conn().TableColumnMetadata(
+				types[i], _, nulls[i], _, _, _ = c.TableColumnMetadata(
 					r.Stmt.ColumnDatabaseName(i),
 					r.Stmt.ColumnTableName(i),
 					col)
@@ -762,8 +771,10 @@ func (r *rows) ColumnTypeScanType(index int) (typ reflect.Type) {
 }
 
 func (r *rows) Next(dest []driver.Value) error {
-	old := r.Stmt.Conn().SetInterrupt(r.ctx)
-	defer r.Stmt.Conn().SetInterrupt(old)
+	c := r.Stmt.Conn()
+	if old := c.SetInterrupt(r.ctx); old != r.ctx {
+		defer c.SetInterrupt(old)
+	}
 
 	if !r.Stmt.Step() {
 		if err := r.Stmt.Err(); err != nil {

driver/driver_test.go

@@ -248,8 +248,10 @@ func Test_nested_context(t *testing.T) {
 	want(inner, 0)
 	cancel()
 
-	if inner.Next() || !errors.Is(inner.Err(), sqlite3.INTERRUPT) {
-		t.Fatal(inner.Err())
+	var terr interface{ Temporary() bool }
+	if inner.Next() || !errors.Is(inner.Err(), context.Canceled) &&
+		(!errors.As(inner.Err(), &terr) || !terr.Temporary()) {
+		t.Fatalf("got %v, want cancellation", inner.Err())
 	}
 
 	want(outer, 1)

embed/bcw2/build.sh

@@ -13,11 +13,10 @@ mkdir -p build/ext/
 cp "$ROOT"/sqlite3/*.[ch] build/
 cp "$ROOT"/sqlite3/*.patch build/
 
-# https://sqlite.org/src/info/c09656c62155a6e8
-curl -# https://sqlite.org/src/tarball/sqlite.tar.gz?r=c09656c6 | tar xz
+# https://sqlite.org/src/info/3215186aa9204149
+curl -# https://sqlite.org/src/tarball/sqlite.tar.gz?r=3215186a | tar xz
 
 cd sqlite
-cat ../repro.patch | patch -p0 --no-backup-if-mismatch
 if [[ "$OSTYPE" == "msys" || "$OSTYPE" == "cygwin" ]]; then
 	MSYS_NO_PATHCONV=1 nmake /f makefile.msc sqlite3.c "OPTS=-DSQLITE_ENABLE_UPDATE_DELETE_LIMIT -DSQLITE_ENABLE_ORDERED_SET_AGGREGATES"
 else
@@ -61,7 +60,7 @@ cd ~-
 
 "$BINARYEN/wasm-ctor-eval" -g -c _initialize bcw2.wasm -o bcw2.tmp
 "$BINARYEN/wasm-opt" -g --strip --strip-producers -c -O3 \
-	bcw2.tmp -o bcw2.wasm \
+	bcw2.tmp -o bcw2.wasm --low-memory-unused \
 	--enable-simd --enable-mutable-globals --enable-multivalue \
 	--enable-bulk-memory --enable-reference-types \
 	--enable-nontrapping-float-to-int --enable-sign-ext

embed/bcw2/go.mod

@@ -4,11 +4,11 @@ go 1.23.0
 
 toolchain go1.24.0
 
-require github.com/ncruces/go-sqlite3 v0.24.0
+require github.com/ncruces/go-sqlite3 v0.25.0
 
 require (
 	github.com/ncruces/julianday v1.0.0 // indirect
 	github.com/ncruces/sort v0.1.5 // indirect
 	github.com/tetratelabs/wazero v1.9.0 // indirect
-	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/sys v0.32.0 // indirect
 )

embed/bcw2/go.sum

@@ -1,12 +1,12 @@
-github.com/ncruces/go-sqlite3 v0.24.0 h1:Z4jfmzu2NCd4SmyFwLT2OmF3EnTZbqwATvdiuNHNhLA=
-github.com/ncruces/go-sqlite3 v0.24.0/go.mod h1:/Vs8ACZHjJ1SA6E9RZUn3EyB1OP3nDQ4z/ar+0fplTQ=
+github.com/ncruces/go-sqlite3 v0.25.0 h1:trugKUs98Zwy9KwRr/EUxZHL92LYt7UqcKqAfpGpK+I=
+github.com/ncruces/go-sqlite3 v0.25.0/go.mod h1:n6Z7036yFilJx04yV0mi5JWaF66rUmXn1It9Ux8dx68=
 github.com/ncruces/julianday v1.0.0 h1:fH0OKwa7NWvniGQtxdJRxAgkBMolni2BjDHaWTxqt7M=
 github.com/ncruces/julianday v1.0.0/go.mod h1:Dusn2KvZrrovOMJuOt0TNXL6tB7U2E8kvza5fFc9G7g=
 github.com/ncruces/sort v0.1.5 h1:fiFWXXAqKI8QckPf/6hu/bGFwcEPrirIOFaJqWujs4k=
 github.com/ncruces/sort v0.1.5/go.mod h1:obJToO4rYr6VWP0Uw5FYymgYGt3Br4RXcs/JdKaXAPk=
 github.com/tetratelabs/wazero v1.9.0 h1:IcZ56OuxrtaEz8UYNRHBrUa9bYeX9oVY93KspZZBf/I=
 github.com/tetratelabs/wazero v1.9.0/go.mod h1:TSbcXCfFP0L2FGkRPxHphadXPjo1T6W+CseNNY7EkjM=
-golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
-golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/text v0.22.0 h1:bofq7m3/HAFvbF51jz3Q9wLg3jkvSPuiZu/pD1XwgtM=
-golang.org/x/text v0.22.0/go.mod h1:YRoo4H8PVmsu+E3Ou7cqLVH8oXWIHVoX0jqUWALQhfY=
+golang.org/x/sys v0.32.0 h1:s77OFDvIQeibCmezSnk/q6iAfkdiQaJi4VzroCFrN20=
+golang.org/x/sys v0.32.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY=
+golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4=

embed/bcw2/repro.patch

@@ -1,23 +0,0 @@
-# https://sqlite.org/src/vpatch?from=67809715977a5bad&to=3f57584710d61174
---- tool/mkpragmatab.tcl
-+++ tool/mkpragmatab.tcl
-@@ -526,14 +526,17 @@
-   puts $fd [format {#define PragFlg_%-10s 0x%02x /* %s */} \
-              $f $fv $flagMeaning($f)]
-   set fv [expr {$fv*2}]
- }
- 
--# Sort the column lists so that longer column lists occur first
-+# Sort the column lists so that longer column lists occur first.
-+# In the event of a tie, sort column lists lexicographically.
- #
- proc colscmp {a b} {
--  return [expr {[llength $b] - [llength $a]}]
-+  set rc [expr {[llength $b] - [llength $a]}]
-+  if {$rc} {return $rc}
-+  return [string compare $a $b]
- }
- set cols_list [lsort -command colscmp $cols_list]
- 
- # Generate the array of column names used by pragmas that act like
- # queries.

embed/build.sh

@@ -12,7 +12,7 @@ trap 'rm -f sqlite3.tmp' EXIT
 "$WASI_SDK/clang" --target=wasm32-wasi -std=c23 -g0 -O2 \
 	-Wall -Wextra -Wno-unused-parameter -Wno-unused-function \
 	-o sqlite3.wasm "$ROOT/sqlite3/main.c" \
-	-I"$ROOT/sqlite3" \
+	-I"$ROOT/sqlite3/libc" -I"$ROOT/sqlite3" \
 	-mexec-model=reactor \
 	-msimd128 -mmutable-globals -mmultivalue \
 	-mbulk-memory -mreference-types \
@@ -27,7 +27,7 @@ trap 'rm -f sqlite3.tmp' EXIT
 
 "$BINARYEN/wasm-ctor-eval" -g -c _initialize sqlite3.wasm -o sqlite3.tmp
 "$BINARYEN/wasm-opt" -g --strip --strip-producers -c -O3 \
-	sqlite3.tmp -o sqlite3.wasm \
+	sqlite3.tmp -o sqlite3.wasm --low-memory-unused \
 	--enable-simd --enable-mutable-globals --enable-multivalue \
 	--enable-bulk-memory --enable-reference-types \
 	--enable-nontrapping-float-to-int --enable-sign-ext

embed/exports.txt

@@ -66,6 +66,7 @@ sqlite3_errmsg
 sqlite3_error_offset
 sqlite3_errstr
 sqlite3_exec
+sqlite3_exec_go
 sqlite3_expanded_sql
 sqlite3_file_control
 sqlite3_filename_database

error.go

@@ -75,7 +75,7 @@ func (e *Error) As(err any) bool {
 
 // Temporary returns true for [BUSY] errors.
 func (e *Error) Temporary() bool {
-	return e.Code() == BUSY
+	return e.Code() == BUSY || e.Code() == INTERRUPT
 }
 
 // Timeout returns true for [BUSY_TIMEOUT] errors.

ext/blobio/blob_test.go

@@ -64,7 +64,7 @@ func Example() {
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(blobio.Register)
 	sqlite3.AutoExtension(array.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func Test_readblob(t *testing.T) {

ext/bloom/bloom_test.go

@@ -14,7 +14,7 @@ import (
 
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(bloom.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func TestRegister(t *testing.T) {

ext/closure/closure_test.go

@@ -4,6 +4,7 @@ import (
 	_ "embed"
 	"fmt"
 	"log"
+	"os"
 	"testing"
 
 	"github.com/ncruces/go-sqlite3"
@@ -14,7 +15,7 @@ import (
 
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(closure.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func Example() {

ext/csv/csv_test.go

@@ -3,6 +3,7 @@ package csv_test
 import (
 	"fmt"
 	"log"
+	"os"
 	"testing"
 
 	"github.com/ncruces/go-sqlite3"
@@ -56,7 +57,7 @@ func Example() {
 
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(csv.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func TestRegister(t *testing.T) {

ext/pivot/pivot_test.go

@@ -3,6 +3,7 @@ package pivot_test
 import (
 	"fmt"
 	"log"
+	"os"
 	"strings"
 	"testing"
 
@@ -85,7 +86,7 @@ func Example() {
 
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(pivot.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func TestRegister(t *testing.T) {

ext/statement/stmt_test.go

@@ -3,6 +3,7 @@ package statement_test
 import (
 	"fmt"
 	"log"
+	"os"
 	"testing"
 
 	"github.com/ncruces/go-sqlite3"
@@ -50,7 +51,7 @@ func Example() {
 
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(statement.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func TestRegister(t *testing.T) {

ext/stats/stats_test.go

@@ -2,6 +2,7 @@ package stats_test
 
 import (
 	"math"
+	"os"
 	"testing"
 
 	"github.com/ncruces/go-sqlite3"
@@ -12,7 +13,7 @@ import (
 
 func TestMain(m *testing.M) {
 	sqlite3.AutoExtension(stats.Register)
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func TestRegister_variance(t *testing.T) {

go.mod

@@ -8,16 +8,16 @@ require (
 	github.com/ncruces/julianday v1.0.0
 	github.com/ncruces/sort v0.1.5
 	github.com/tetratelabs/wazero v1.9.0
-	golang.org/x/crypto v0.36.0
-	golang.org/x/sys v0.31.0
+	golang.org/x/crypto v0.37.0
+	golang.org/x/sys v0.32.0
 )
 
 require (
 	github.com/dchest/siphash v1.2.3 // ext/bloom
 	github.com/google/uuid v1.6.0 // ext/uuid
 	github.com/psanford/httpreadat v0.1.0 // example
-	golang.org/x/sync v0.12.0 // test
-	golang.org/x/text v0.23.0 // ext/unicode
+	golang.org/x/sync v0.13.0 // test
+	golang.org/x/text v0.24.0 // ext/unicode
 	lukechampine.com/adiantum v1.1.1 // vfs/adiantum
 )
 

go.sum

@@ -10,13 +10,13 @@ github.com/psanford/httpreadat v0.1.0 h1:VleW1HS2zO7/4c7c7zNl33fO6oYACSagjJIyMIw
 github.com/psanford/httpreadat v0.1.0/go.mod h1:Zg7P+TlBm3bYbyHTKv/EdtSJZn3qwbPwpfZ/I9GKCRE=
 github.com/tetratelabs/wazero v1.9.0 h1:IcZ56OuxrtaEz8UYNRHBrUa9bYeX9oVY93KspZZBf/I=
 github.com/tetratelabs/wazero v1.9.0/go.mod h1:TSbcXCfFP0L2FGkRPxHphadXPjo1T6W+CseNNY7EkjM=
-golang.org/x/crypto v0.36.0 h1:AnAEvhDddvBdpY+uR+MyHmuZzzNqXSe/GvuDeob5L34=
-golang.org/x/crypto v0.36.0/go.mod h1:Y4J0ReaxCR1IMaabaSMugxJES1EpwhBHhv2bDHklZvc=
-golang.org/x/sync v0.12.0 h1:MHc5BpPuC30uJk597Ri8TV3CNZcTLu6B6z4lJy+g6Jw=
-golang.org/x/sync v0.12.0/go.mod h1:1dzgHSNfp02xaA81J2MS99Qcpr2w7fw1gpm99rleRqA=
-golang.org/x/sys v0.31.0 h1:ioabZlmFYtWhL+TRYpcnNlLwhyxaM9kWTDEmfnprqik=
-golang.org/x/sys v0.31.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
-golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY=
-golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4=
+golang.org/x/crypto v0.37.0 h1:kJNSjF/Xp7kU0iB2Z+9viTPMW4EqqsrywMXLJOOsXSE=
+golang.org/x/crypto v0.37.0/go.mod h1:vg+k43peMZ0pUMhYmVAWysMK35e6ioLh3wB8ZCAfbVc=
+golang.org/x/sync v0.13.0 h1:AauUjRAJ9OSnvULf/ARrrVywoJDy0YS2AwQ98I37610=
+golang.org/x/sync v0.13.0/go.mod h1:1dzgHSNfp02xaA81J2MS99Qcpr2w7fw1gpm99rleRqA=
+golang.org/x/sys v0.32.0 h1:s77OFDvIQeibCmezSnk/q6iAfkdiQaJi4VzroCFrN20=
+golang.org/x/sys v0.32.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/text v0.24.0 h1:dd5Bzh4yt5KYA8f9CJHCP4FB4D51c2c6JvN37xJJkJ0=
+golang.org/x/text v0.24.0/go.mod h1:L8rBsPeo2pSS+xqN0d5u2ikmjtmoJbDBT1b7nHvFCdU=
 lukechampine.com/adiantum v1.1.1 h1:4fp6gTxWCqpEbLy40ExiYDDED3oUNWx5cTqBCtPdZqA=
 lukechampine.com/adiantum v1.1.1/go.mod h1:LrAYVnTYLnUtE/yMp5bQr0HstAf060YUF8nM0B6+rUw=

gormlite/go.mod

@@ -5,7 +5,7 @@ go 1.23.0
 toolchain go1.24.0
 
 require (
-	github.com/ncruces/go-sqlite3 v0.24.0
+	github.com/ncruces/go-sqlite3 v0.25.0
 	gorm.io/gorm v1.25.12
 )
 
@@ -14,6 +14,6 @@ require (
 	github.com/jinzhu/now v1.1.5 // indirect
 	github.com/ncruces/julianday v1.0.0 // indirect
 	github.com/tetratelabs/wazero v1.9.0 // indirect
-	golang.org/x/sys v0.30.0 // indirect
-	golang.org/x/text v0.22.0 // indirect
+	golang.org/x/sys v0.32.0 // indirect
+	golang.org/x/text v0.24.0 // indirect
 )

gormlite/go.sum

@@ -2,15 +2,15 @@ github.com/jinzhu/inflection v1.0.0 h1:K317FqzuhWc8YvSVlFMCCUb36O/S9MCKRDI7QkRKD
 github.com/jinzhu/inflection v1.0.0/go.mod h1:h+uFLlag+Qp1Va5pdKtLDYj+kHp5pxUVkryuEj+Srlc=
 github.com/jinzhu/now v1.1.5 h1:/o9tlHleP7gOFmsnYNz3RGnqzefHA47wQpKrrdTIwXQ=
 github.com/jinzhu/now v1.1.5/go.mod h1:d3SSVoowX0Lcu0IBviAWJpolVfI5UJVZZ7cO71lE/z8=
-github.com/ncruces/go-sqlite3 v0.24.0 h1:Z4jfmzu2NCd4SmyFwLT2OmF3EnTZbqwATvdiuNHNhLA=
-github.com/ncruces/go-sqlite3 v0.24.0/go.mod h1:/Vs8ACZHjJ1SA6E9RZUn3EyB1OP3nDQ4z/ar+0fplTQ=
+github.com/ncruces/go-sqlite3 v0.25.0 h1:trugKUs98Zwy9KwRr/EUxZHL92LYt7UqcKqAfpGpK+I=
+github.com/ncruces/go-sqlite3 v0.25.0/go.mod h1:n6Z7036yFilJx04yV0mi5JWaF66rUmXn1It9Ux8dx68=
 github.com/ncruces/julianday v1.0.0 h1:fH0OKwa7NWvniGQtxdJRxAgkBMolni2BjDHaWTxqt7M=
 github.com/ncruces/julianday v1.0.0/go.mod h1:Dusn2KvZrrovOMJuOt0TNXL6tB7U2E8kvza5fFc9G7g=
 github.com/tetratelabs/wazero v1.9.0 h1:IcZ56OuxrtaEz8UYNRHBrUa9bYeX9oVY93KspZZBf/I=
 github.com/tetratelabs/wazero v1.9.0/go.mod h1:TSbcXCfFP0L2FGkRPxHphadXPjo1T6W+CseNNY7EkjM=
-golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
-golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/text v0.22.0 h1:bofq7m3/HAFvbF51jz3Q9wLg3jkvSPuiZu/pD1XwgtM=
-golang.org/x/text v0.22.0/go.mod h1:YRoo4H8PVmsu+E3Ou7cqLVH8oXWIHVoX0jqUWALQhfY=
+golang.org/x/sys v0.32.0 h1:s77OFDvIQeibCmezSnk/q6iAfkdiQaJi4VzroCFrN20=
+golang.org/x/sys v0.32.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/text v0.24.0 h1:dd5Bzh4yt5KYA8f9CJHCP4FB4D51c2c6JvN37xJJkJ0=
+golang.org/x/text v0.24.0/go.mod h1:L8rBsPeo2pSS+xqN0d5u2ikmjtmoJbDBT1b7nHvFCdU=
 gorm.io/gorm v1.25.12 h1:I0u8i2hWQItBq1WfE0o2+WuL9+8L21K9e2HHSTE/0f8=
 gorm.io/gorm v1.25.12/go.mod h1:xh7N7RHfYlNc5EmcI/El95gXusucDrQnHXe0+CgWcLQ=

sqlite3/libc/benchmark.sh

@@ -0,0 +1,12 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+cd -P -- "$(dirname -- "$0")"
+
+touch empty.S
+./build.sh empty.S
+go test -bench=.
+rm -f empty.S
+
+./build.sh
+go test -bench=.

sqlite3/libc/build.sh

@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+cd -P -- "$(dirname -- "$0")"
+
+ROOT=../../
+BINARYEN="$ROOT/tools/binaryen/bin"
+WASI_SDK="$ROOT/tools/wasi-sdk/bin"
+SRCS="${1:-libc.c}"
+"../tools.sh"
+
+trap 'rm -f libc.c libc.tmp' EXIT
+cat << EOF > libc.c
+#include <string.h>
+#include <stdlib.h>
+EOF
+
+"$WASI_SDK/clang" --target=wasm32-wasi -std=c23 -g0 -O2 \
+	-Wall -Wextra -Wno-unused-parameter -Wno-unused-function \
+	-o libc.wasm -I. "$SRCS" \
+	-mexec-model=reactor \
+	-msimd128 -mmutable-globals -mmultivalue \
+	-mbulk-memory -mreference-types \
+	-mnontrapping-fptoint -msign-ext \
+	-fno-stack-protector -fno-stack-clash-protection \
+	-Wl,-z,stack-size=1024 \
+	-Wl,--stack-first \
+	-Wl,--import-undefined \
+	-Wl,--initial-memory=16777216 \
+	-Wl,--export=memchr \
+	-Wl,--export=memcmp \
+	-Wl,--export=memcpy \
+	-Wl,--export=memset \
+	-Wl,--export=strchr \
+	-Wl,--export=strchrnul \
+	-Wl,--export=strcmp \
+	-Wl,--export=strcspn \
+	-Wl,--export=strlen \
+	-Wl,--export=strncmp \
+	-Wl,--export=strspn \
+	-Wl,--export=qsort
+
+"$BINARYEN/wasm-ctor-eval" -g -c _initialize libc.wasm -o libc.tmp
+"$BINARYEN/wasm-opt" -g --strip --strip-producers -c -O3 \
+	libc.tmp -o libc.wasm \
+	--enable-simd --enable-mutable-globals --enable-multivalue \
+	--enable-bulk-memory --enable-reference-types \
+	--enable-nontrapping-float-to-int --enable-sign-ext
+
+"$BINARYEN/wasm-dis" -o libc.wat libc.wasm

sqlite3/libc/libc_test.go

@@ -0,0 +1,429 @@
+package libc
+
+import (
+	"context"
+	_ "embed"
+	"os"
+	"testing"
+
+	"github.com/tetratelabs/wazero"
+	"github.com/tetratelabs/wazero/api"
+)
+
+//go:embed libc.wasm
+var binary []byte
+
+const (
+	page = 64 * 1024
+	size = 1024 * 1024 * 4
+	ptr1 = 1024 * 1024
+	ptr2 = ptr1 + size
+)
+
+var (
+	memory  []byte
+	module  api.Module
+	memset  api.Function
+	memcpy  api.Function
+	memchr  api.Function
+	memcmp  api.Function
+	strlen  api.Function
+	strchr  api.Function
+	strcmp  api.Function
+	strspn  api.Function
+	strncmp api.Function
+	strcspn api.Function
+	stack   [8]uint64
+)
+
+func call(fn api.Function, arg ...uint64) uint64 {
+	copy(stack[:], arg)
+	err := fn.CallWithStack(context.Background(), stack[:])
+	if err != nil {
+		panic(err)
+	}
+	return stack[0]
+}
+
+func TestMain(m *testing.M) {
+	ctx := context.Background()
+
+	runtime := wazero.NewRuntime(ctx)
+	mod, err := runtime.Instantiate(ctx, binary)
+	if err != nil {
+		panic(err)
+	}
+
+	module = mod
+	memset = mod.ExportedFunction("memset")
+	memcpy = mod.ExportedFunction("memcpy")
+	memchr = mod.ExportedFunction("memchr")
+	memcmp = mod.ExportedFunction("memcmp")
+	strlen = mod.ExportedFunction("strlen")
+	strchr = mod.ExportedFunction("strchr")
+	strcmp = mod.ExportedFunction("strcmp")
+	strspn = mod.ExportedFunction("strspn")
+	strncmp = mod.ExportedFunction("strncmp")
+	strcspn = mod.ExportedFunction("strcspn")
+	memory, _ = mod.Memory().Read(0, mod.Memory().Size())
+
+	os.Exit(m.Run())
+}
+
+func Benchmark_memset(b *testing.B) {
+	clear(memory)
+
+	b.SetBytes(size)
+	b.ResetTimer()
+	for range b.N {
+		call(memset, ptr1, 3, size)
+	}
+}
+
+func Benchmark_memcpy(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr2:ptr2+size], 5)
+
+	b.SetBytes(size)
+	b.ResetTimer()
+	for range b.N {
+		call(memcpy, ptr1, ptr2, size)
+	}
+}
+
+func Benchmark_memchr(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size/2], 7)
+	fill(memory[ptr1+size/2:ptr1+size], 5)
+
+	b.SetBytes(size/2 + 1)
+	b.ResetTimer()
+	for range b.N {
+		call(memchr, ptr1, 5, size)
+	}
+}
+
+func Benchmark_memcmp(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size], 7)
+	fill(memory[ptr2:ptr2+size/2], 7)
+	fill(memory[ptr2+size/2:ptr2+size], 5)
+
+	b.SetBytes(size/2 + 1)
+	b.ResetTimer()
+	for range b.N {
+		call(memcmp, ptr1, ptr2, size)
+	}
+}
+
+func Benchmark_strlen(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size-1], 5)
+
+	b.SetBytes(size)
+	b.ResetTimer()
+	for range b.N {
+		call(strlen, ptr1)
+	}
+}
+
+func Benchmark_strchr(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size/2], 7)
+	fill(memory[ptr1+size/2:ptr1+size-1], 5)
+
+	b.SetBytes(size/2 + 1)
+	b.ResetTimer()
+	for range b.N {
+		call(strchr, ptr1, 5)
+	}
+}
+
+func Benchmark_strcmp(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size-1], 7)
+	fill(memory[ptr2:ptr2+size/2], 7)
+	fill(memory[ptr2+size/2:ptr2+size-1], 5)
+
+	b.SetBytes(size/2 + 1)
+	b.ResetTimer()
+	for range b.N {
+		call(strcmp, ptr1, ptr2, size)
+	}
+}
+
+func Benchmark_strncmp(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size-1], 7)
+	fill(memory[ptr2:ptr2+size/2], 7)
+	fill(memory[ptr2+size/2:ptr2+size-1], 5)
+
+	b.SetBytes(size/2 + 1)
+	b.ResetTimer()
+	for range b.N {
+		call(strncmp, ptr1, ptr2, size-1)
+	}
+}
+
+func Benchmark_strspn(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size/2], 7)
+	fill(memory[ptr1+size/2:ptr1+size-1], 5)
+	memory[ptr2+0] = 3
+	memory[ptr2+1] = 5
+	memory[ptr2+2] = 7
+	memory[ptr2+3] = 9
+
+	b.SetBytes(size)
+	b.ResetTimer()
+	for range b.N {
+		call(strspn, ptr1, ptr2)
+	}
+}
+
+func Benchmark_strcspn(b *testing.B) {
+	clear(memory)
+	fill(memory[ptr1:ptr1+size/2], 7)
+	fill(memory[ptr1+size/2:ptr1+size-1], 5)
+	memory[ptr2+0] = 3
+	memory[ptr2+1] = 9
+
+	b.SetBytes(size)
+	b.ResetTimer()
+	for range b.N {
+		call(strcspn, ptr1, ptr2)
+	}
+}
+
+func Test_memchr(t *testing.T) {
+	for length := range 64 {
+		for pos := range length + 2 {
+			for alignment := range 24 {
+				clear(memory[:2*page])
+
+				ptr := (page - 8) + alignment
+				fill(memory[ptr:ptr+max(pos, length)], 5)
+				memory[ptr+pos] = 7
+
+				want := 0
+				if pos < length {
+					want = ptr + pos
+				}
+
+				got := call(memchr, uint64(ptr), 7, uint64(length))
+				if uint32(got) != uint32(want) {
+					t.Errorf("memchr(%d, %d, %d) = %d, want %d",
+						ptr, 7, uint64(length), uint32(got), uint32(want))
+				}
+			}
+		}
+
+		clear(memory)
+		ptr := len(memory) - length
+		fill(memory[ptr:ptr+length], 5)
+		memory[len(memory)-1] = 7
+
+		want := len(memory) - 1
+		if length == 0 {
+			want = 0
+		}
+
+		got := call(memchr, uint64(ptr), 7, uint64(length))
+		if uint32(got) != uint32(want) {
+			t.Errorf("memchr(%d, %d, %d) = %d, want %d",
+				ptr, 7, uint64(length), uint32(got), uint32(want))
+		}
+	}
+}
+
+func Test_strlen(t *testing.T) {
+	for length := range 64 {
+		for alignment := range 24 {
+			clear(memory[:2*page])
+
+			ptr := (page - 8) + alignment
+			fill(memory[ptr:ptr+length], 5)
+
+			got := call(strlen, uint64(ptr))
+			if uint32(got) != uint32(length) {
+				t.Errorf("strlen(%d) = %d, want %d",
+					ptr, uint32(got), uint32(length))
+			}
+
+			memory[ptr-1] = 5
+			got = call(strlen, uint64(ptr))
+			if uint32(got) != uint32(length) {
+				t.Errorf("strlen(%d) = %d, want %d",
+					ptr, uint32(got), uint32(length))
+			}
+		}
+
+		clear(memory)
+		ptr := len(memory) - length - 1
+		fill(memory[ptr:ptr+length], 5)
+
+		got := call(strlen, uint64(ptr))
+		if uint32(got) != uint32(length) {
+			t.Errorf("strlen(%d) = %d, want %d",
+				ptr, uint32(got), uint32(length))
+		}
+	}
+}
+
+func Test_strchr(t *testing.T) {
+	for length := range 64 {
+		for pos := range length + 2 {
+			for alignment := range 24 {
+				clear(memory[:2*page])
+
+				ptr := (page - 8) + alignment
+				fill(memory[ptr:ptr+max(pos, length)], 5)
+				memory[ptr+pos] = 7
+				memory[ptr+length] = 0
+
+				want := 0
+				if pos < length {
+					want = ptr + pos
+				}
+
+				got := call(strchr, uint64(ptr), 7)
+				if uint32(got) != uint32(want) {
+					t.Errorf("strchr(%d, %d) = %d, want %d",
+						ptr, 7, uint32(got), uint32(want))
+				}
+			}
+		}
+
+		clear(memory)
+		ptr := len(memory) - length
+		fill(memory[ptr:ptr+length], 5)
+		memory[len(memory)-1] = 7
+
+		want := len(memory) - 1
+		if length == 0 {
+			continue
+		}
+
+		got := call(strchr, uint64(ptr), 7)
+		if uint32(got) != uint32(want) {
+			t.Errorf("strchr(%d, %d) = %d, want %d",
+				ptr, 7, uint32(got), uint32(want))
+		}
+	}
+}
+
+func Test_strspn(t *testing.T) {
+	for length := range 64 {
+		for pos := range length + 2 {
+			for alignment := range 24 {
+				clear(memory[:2*page])
+
+				ptr := (page - 8) + alignment
+				fill(memory[ptr:ptr+max(pos, length)], 5)
+				memory[ptr+pos] = 7
+				memory[ptr+length] = 0
+				memory[128] = 3
+				memory[129] = 5
+
+				want := min(pos, length)
+
+				got := call(strspn, uint64(ptr), 129)
+				if uint32(got) != uint32(want) {
+					t.Errorf("strspn(%d, %d) = %d, want %d",
+						ptr, 129, uint32(got), uint32(want))
+				}
+
+				got = call(strspn, uint64(ptr), 128)
+				if uint32(got) != uint32(want) {
+					t.Errorf("strspn(%d, %d) = %d, want %d",
+						ptr, 128, uint32(got), uint32(want))
+				}
+			}
+		}
+
+		clear(memory)
+		ptr := len(memory) - length
+		fill(memory[ptr:ptr+length], 5)
+		memory[len(memory)-1] = 7
+		memory[128] = 3
+		memory[129] = 5
+
+		want := length - 1
+		if length == 0 {
+			continue
+		}
+
+		got := call(strspn, uint64(ptr), 129)
+		if uint32(got) != uint32(want) {
+			t.Errorf("strspn(%d, %d) = %d, want %d",
+				ptr, 129, uint32(got), uint32(want))
+		}
+
+		got = call(strspn, uint64(ptr), 128)
+		if uint32(got) != uint32(want) {
+			t.Errorf("strspn(%d, %d) = %d, want %d",
+				ptr, 128, uint32(got), uint32(want))
+		}
+	}
+}
+
+func Test_strcspn(t *testing.T) {
+	for length := range 64 {
+		for pos := range length + 2 {
+			for alignment := range 24 {
+				clear(memory[:2*page])
+
+				ptr := (page - 8) + alignment
+				fill(memory[ptr:ptr+max(pos, length)], 5)
+				memory[ptr+pos] = 7
+				memory[ptr+length] = 0
+				memory[128] = 3
+				memory[129] = 7
+
+				want := min(pos, length)
+
+				got := call(strcspn, uint64(ptr), 129)
+				if uint32(got) != uint32(want) {
+					t.Errorf("strcspn(%d, %d) = %d, want %d",
+						ptr, 129, uint32(got), uint32(want))
+				}
+
+				got = call(strcspn, uint64(ptr), 128)
+				if uint32(got) != uint32(want) {
+					t.Errorf("strcspn(%d, %d) = %d, want %d",
+						ptr, 128, uint32(got), uint32(want))
+				}
+			}
+		}
+
+		clear(memory)
+		ptr := len(memory) - length
+		fill(memory[ptr:ptr+length], 5)
+		memory[len(memory)-1] = 7
+		memory[128] = 3
+		memory[129] = 7
+
+		want := length - 1
+		if length == 0 {
+			continue
+		}
+
+		got := call(strcspn, uint64(ptr), 129)
+		if uint32(got) != uint32(want) {
+			t.Errorf("strcspn(%d, %d) = %d, want %d",
+				ptr, 129, uint32(got), uint32(want))
+		}
+
+		got = call(strcspn, uint64(ptr), 128)
+		if uint32(got) != uint32(want) {
+			t.Errorf("strcspn(%d, %d) = %d, want %d",
+				ptr, 128, uint32(got), uint32(want))
+		}
+	}
+}
+
+func fill(s []byte, v byte) {
+	for i := range s {
+		s[i] = v
+	}
+}

sqlite3/libc/stdlib.h

@@ -0,0 +1,51 @@
+#ifndef _WASM_SIMD128_STDLIB_H
+#define _WASM_SIMD128_STDLIB_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include_next <stdlib.h>  // the system stdlib.h
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Shellsort with Gonnet & Baeza-Yates gap sequence.
+// Simple, no recursion, doesn't use the C stack.
+// Clang auto-vectorizes the inner loop.
+
+__attribute__((weak))
+void qsort(void *base, size_t nel, size_t width,
+           int (*comp)(const void *, const void *)) {
+  if (width == 0) return;
+
+  size_t wnel = width * nel;
+  size_t gap = nel;
+  while (gap > 1) {
+    gap = (5ull * gap - 1) / 11;
+    if (gap == 0) gap = 1;
+
+    size_t wgap = width * gap;
+    __builtin_assume(wgap < wnel);
+    for (size_t i = wgap; i < wnel; i += width) {
+      for (size_t j = i; !__builtin_sub_overflow(j, wgap, &j);) {
+        char *a = j + (char *)base;
+        char *b = a + wgap;
+        if (comp(a, b) <= 0) break;
+
+        size_t s = width;
+        do {
+          char tmp = *a;
+          *a++ = *b;
+          *b++ = tmp;
+        } while (--s);
+      }
+    }
+  }
+}
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // _WASM_SIMD128_STDLIB_H

sqlite3/libc/string.h

@@ -0,0 +1,373 @@
+#ifndef _WASM_SIMD128_STRING_H
+#define _WASM_SIMD128_STRING_H
+
+#include <limits.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <wasm_simd128.h>
+#include <__macro_PAGESIZE.h>
+
+#include_next <string.h>  // the system string.h
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef __wasm_bulk_memory__
+
+// Use the builtins if compiled with bulk memory operations.
+// Clang will intrinsify using SIMD for small, constant N.
+// For everything else, this helps inlining.
+
+__attribute__((weak))
+void *memset(void *dest, int c, size_t n) {
+  return __builtin_memset(dest, c, n);
+}
+
+__attribute__((weak))
+void *memcpy(void *restrict dest, const void *restrict src, size_t n) {
+  return __builtin_memcpy(dest, src, n);
+}
+
+__attribute__((weak))
+void *memmove(void *dest, const void *src, size_t n) {
+  return __builtin_memmove(dest, src, n);
+}
+
+#endif  // __wasm_bulk_memory__
+
+#ifdef __wasm_simd128__
+
+// SIMD versions of some string.h functions.
+//
+// These assume aligned v128_t loads can't fail,
+// and so can't unaligned loads up to the last
+// aligned address less than memory size.
+//
+// These also assume unaligned access is not painfully slow,
+// but that bitmask extraction is really slow on AArch64.
+
+__attribute__((weak))
+int memcmp(const void *v1, const void *v2, size_t n) {
+  // memcmp can read up to n bytes from each object.
+  // Use unaligned loads to handle the case where
+  // the objects have mismatching alignments.
+  const v128_t *w1 = v1;
+  const v128_t *w2 = v2;
+  for (; n >= sizeof(v128_t); n -= sizeof(v128_t)) {
+    // Find any single bit difference.
+    if (wasm_v128_any_true(wasm_v128_load(w1) ^ wasm_v128_load(w2))) {
+      break;
+    }
+    w1++;
+    w2++;
+  }
+
+  // Continue byte-by-byte.
+  const unsigned char *u1 = (void *)w1;
+  const unsigned char *u2 = (void *)w2;
+  while (n--) {
+    if (*u1 != *u2) return *u1 - *u2;
+    u1++;
+    u2++;
+  }
+  return 0;
+}
+
+__attribute__((weak))
+void *memchr(const void *v, int c, size_t n) {
+  // When n is zero, a function that locates a character finds no occurrence.
+  // Otherwise, decrement n to ensure __builtin_sub_overflow "overflows"
+  // when n would go equal-to-or-below zero.
+  if (n-- == 0) {
+    return NULL;
+  }
+
+  // memchr must behave as if it reads characters sequentially
+  // and stops as soon as a match is found.
+  // Aligning ensures loads can't fail.
+  uintptr_t align = (uintptr_t)v % sizeof(v128_t);
+  const v128_t *w = (void *)(v - align);
+  const v128_t wc = wasm_i8x16_splat(c);
+
+  while (true) {
+    const v128_t cmp = wasm_i8x16_eq(*w, wc);
+    // Bitmask is slow on AArch64, any_true is much faster.
+    if (wasm_v128_any_true(cmp)) {
+      // Clear the bits corresponding to alignment
+      // so we can count trailing zeros.
+      int mask = wasm_i8x16_bitmask(cmp) >> align << align;
+      // At least one bit will be set, unless we cleared them.
+      // Knowing this helps the compiler. 
+      __builtin_assume(mask || align);
+      // If the mask is zero because of alignment,
+      // it's as if we didn't find anything.
+      if (mask) {
+        // We found a match, unless it is beyond the end of the object.
+        // Recall that we decremented n, so less-than-or-equal-to is correct.
+        size_t ctz = __builtin_ctz(mask);
+        return ctz <= n + align ? (void *)w + ctz : NULL;
+      }
+    }
+    // Decrement n; if it "overflows" we're done.
+    if (__builtin_sub_overflow(n, sizeof(v128_t) - align, &n)) {
+      return NULL;
+    }
+    align = 0;
+    w++;
+  }
+}
+
+__attribute__((weak))
+size_t strlen(const char *s) {
+  // strlen must stop as soon as it finds the terminator.
+  // Aligning ensures loads can't fail.
+  uintptr_t align = (uintptr_t)s % sizeof(v128_t);
+  const v128_t *w = (void *)(s - align);
+
+  while (true) {
+    // Bitmask is slow on AArch64, all_true is much faster.
+    if (!wasm_i8x16_all_true(*w)) {
+      const v128_t cmp = wasm_i8x16_eq(*w, (v128_t){});
+      // Clear the bits corresponding to alignment
+      // so we can count trailing zeros.
+      int mask = wasm_i8x16_bitmask(cmp) >> align << align;
+      // At least one bit will be set, unless we cleared them.
+      // Knowing this helps the compiler. 
+      __builtin_assume(mask || align);
+      if (mask) {
+        return (char *)w - s + __builtin_ctz(mask);
+      }
+    }
+    align = 0;
+    w++;
+  }
+}
+
+static int __strcmp(const char *s1, const char *s2) {
+  // Set limit to the largest possible valid v128_t pointer.
+  // Unsigned modular arithmetic gives the correct result
+  // unless memory size is zero, in which case all pointers are invalid.
+  const v128_t *const limit =
+      (v128_t *)(__builtin_wasm_memory_size(0) * PAGESIZE) - 1;
+
+  // Use unaligned loads to handle the case where
+  // the strings have mismatching alignments.
+  const v128_t *w1 = (void *)s1;
+  const v128_t *w2 = (void *)s2;
+  while (w1 <= limit && w2 <= limit) {
+    // Find any single bit difference.
+    if (wasm_v128_any_true(wasm_v128_load(w1) ^ wasm_v128_load(w2))) {
+      break;
+    }
+    // All bytes are equal.
+    // If any byte is zero (on both strings) the strings are equal.
+    if (!wasm_i8x16_all_true(wasm_v128_load(w1))) {
+      return 0;
+    }
+    w1++;
+    w2++;
+  }
+
+  // Continue byte-by-byte.
+  const unsigned char *u1 = (void *)w1;
+  const unsigned char *u2 = (void *)w2;
+  while (true) {
+    if (*u1 != *u2) return *u1 - *u2;
+    if (*u1 == 0) break;
+    u1++;
+    u2++;
+  }
+  return 0;
+}
+
+__attribute__((weak, always_inline))
+int strcmp(const char *s1, const char *s2) {
+  // Use strncmp when comparing against literal strings.
+  // If the literal is small, the vector search will be skipped.
+  if (__builtin_constant_p(strlen(s2))) {
+    return strncmp(s1, s2, strlen(s2));
+  }
+  return __strcmp(s1, s2);
+}
+
+__attribute__((weak))
+int strncmp(const char *s1, const char *s2, size_t n) {
+  // Set limit to the largest possible valid v128_t pointer.
+  // Unsigned modular arithmetic gives the correct result
+  // unless memory size is zero, in which case all pointers are invalid.
+  const v128_t *const limit =
+      (v128_t *)(__builtin_wasm_memory_size(0) * PAGESIZE) - 1;
+
+  // Use unaligned loads to handle the case where
+  // the strings have mismatching alignments.
+  const v128_t *w1 = (void *)s1;
+  const v128_t *w2 = (void *)s2;
+  for (; w1 <= limit && w2 <= limit && n >= sizeof(v128_t); n -= sizeof(v128_t)) {
+    // Find any single bit difference.
+    if (wasm_v128_any_true(wasm_v128_load(w1) ^ wasm_v128_load(w2))) {
+      break;
+    }
+    // All bytes are equal.
+    // If any byte is zero (on both strings) the strings are equal.
+    if (!wasm_i8x16_all_true(wasm_v128_load(w1))) {
+      return 0;
+    }
+    w1++;
+    w2++;
+  }
+
+  // Continue byte-by-byte.
+  const unsigned char *u1 = (void *)w1;
+  const unsigned char *u2 = (void *)w2;
+  while (n--) {
+    if (*u1 != *u2) return *u1 - *u2;
+    if (*u1 == 0) break;
+    u1++;
+    u2++;
+  }
+  return 0;
+}
+
+static char *__strchrnul(const char *s, int c) {
+  // strchrnul must stop as soon as a match is found.
+  // Aligning ensures loads can't fail.
+  uintptr_t align = (uintptr_t)s % sizeof(v128_t);
+  const v128_t *w = (void *)(s - align);
+  const v128_t wc = wasm_i8x16_splat(c);
+
+  while (true) {
+    const v128_t cmp = wasm_i8x16_eq(*w, (v128_t){}) | wasm_i8x16_eq(*w, wc);
+    // Bitmask is slow on AArch64, any_true is much faster.
+    if (wasm_v128_any_true(cmp)) {
+      // Clear the bits corresponding to alignment
+      // so we can count trailing zeros.
+      int mask = wasm_i8x16_bitmask(cmp) >> align << align;
+      // At least one bit will be set, unless we cleared them.
+      // Knowing this helps the compiler. 
+      __builtin_assume(mask || align);
+      if (mask) {
+        return (char *)w + __builtin_ctz(mask);
+      }
+    }
+    align = 0;
+    w++;
+  }
+}
+
+__attribute__((weak, always_inline))
+char *strchrnul(const char *s, int c) {
+  // For finding the terminator, strlen is faster.
+  if (__builtin_constant_p(c) && (char)c == 0) {
+    return (char *)s + strlen(s);
+  }
+  return __strchrnul(s, c);
+}
+
+__attribute__((weak, always_inline))
+char *strchr(const char *s, int c) {
+  // For finding the terminator, strlen is faster.
+  if (__builtin_constant_p(c) && (char)c == 0) {
+    return (char *)s + strlen(s);
+  }
+  char *r = __strchrnul(s, c);
+  return *(char *)r == (char)c ? r : NULL;
+}
+
+__attribute__((weak))
+size_t strspn(const char *s, const char *c) {
+#ifndef _REENTRANT
+  static // Avoid the stack for builds without threads.
+#endif
+  char byteset[UCHAR_MAX + 1];
+  const char *const a = s;
+
+  if (!c[0]) return 0;
+  if (!c[1]) {
+    // Set limit to the largest possible valid v128_t pointer.
+    // Unsigned modular arithmetic gives the correct result
+    // unless memory size is zero, in which case all pointers are invalid.
+    const v128_t *const limit =
+        (v128_t *)(__builtin_wasm_memory_size(0) * PAGESIZE) - 1;
+
+    const v128_t *w = (void *)s;
+    const v128_t wc = wasm_i8x16_splat(*c);
+    while (w <= limit) {
+      if (!wasm_i8x16_all_true(wasm_i8x16_eq(wasm_v128_load(w), wc))) {
+        break;
+      }
+      w++;
+    }
+
+    s = (void *)w;
+    while (*s == *c) s++;
+    return s - a;
+  }
+
+#if !__OPTIMIZE__ || __OPTIMIZE_SIZE__
+
+  // Unoptimized version.
+  memset(byteset, 0, sizeof(byteset));
+  while (*c && (byteset[*(unsigned char *)c] = 1)) c++;
+  while (byteset[*(unsigned char *)s]) s++;
+
+#else
+
+  // This is faster than memset.
+  volatile v128_t *w = (void *)byteset;
+  #pragma unroll
+  for (size_t i = sizeof(byteset) / sizeof(v128_t); i--;) w[i] = (v128_t){};
+  static_assert(sizeof(byteset) % sizeof(v128_t) == 0);
+
+  // Keeping byteset[0] = 0 avoids the other loop having to test for it.
+  while (*c && (byteset[*(unsigned char *)c] = 1)) c++;
+  #pragma unroll 4
+  while (byteset[*(unsigned char *)s]) s++;
+
+#endif
+
+  return s - a;
+}
+
+__attribute__((weak))
+size_t strcspn(const char *s, const char *c) {
+#ifndef _REENTRANT
+  static // Avoid the stack for builds without threads.
+#endif
+  char byteset[UCHAR_MAX + 1];
+  const char *const a = s;
+
+  if (!c[0] || !c[1]) return __strchrnul(s, *c) - s;
+
+#if !__OPTIMIZE__ || __OPTIMIZE_SIZE__
+
+  // Unoptimized version.
+  memset(byteset, 0, sizeof(byteset));
+  while ((byteset[*(unsigned char *)c] = 1) && *c) c++;
+  while (!byteset[*(unsigned char *)s]) s++;
+
+#else
+
+  // This is faster than memset.
+  volatile v128_t *w = (void *)byteset;
+  #pragma unroll
+  for (size_t i = sizeof(byteset) / sizeof(v128_t); i--;) w[i] = (v128_t){};
+  static_assert(sizeof(byteset) % sizeof(v128_t) == 0);
+
+  // Setting byteset[0] = 1 avoids the other loop having to test for it.
+  while ((byteset[*(unsigned char *)c] = 1) && *c) c++;
+  #pragma unroll 4
+  while (!byteset[*(unsigned char *)s]) s++;
+
+#endif
+
+  return s - a;
+}
+
+#endif  // __wasm_simd128__
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // _WASM_SIMD128_STRING_H

sqlite3/main.c

@@ -10,12 +10,11 @@
 #include "ext/spellfix.c"
 #include "ext/uint.c"
 // Bindings
-#include "bind.c"
-#include "column.c"
 #include "func.c"
 #include "hooks.c"
 #include "pointer.c"
-#include "result.c"
+#include "stmt.c"
+#include "text.c"
 #include "time.c"
 #include "vfs.c"
 #include "vtab.c"

sqlite3/result.c

@@ -1,14 +0,0 @@
-#include <math.h>
-#include <stdlib.h>
-
-#include "sqlite3.h"
-
-void sqlite3_result_text_go(sqlite3_context *ctx, const char *zData,
-                            sqlite3_uint64 nData) {
-  sqlite3_result_text64(ctx, zData, nData, &sqlite3_free, SQLITE_UTF8);
-}
-
-void sqlite3_result_blob_go(sqlite3_context *ctx, const void *zData,
-                            sqlite3_uint64 nData) {
-  sqlite3_result_blob64(ctx, zData, nData, &sqlite3_free);
-}

sqlite3/sqlite_opt.h

@@ -4,26 +4,24 @@
 #define SQLITE_THREADSAFE 0
 #define SQLITE_DEFAULT_WAL_SYNCHRONOUS 1
 #define SQLITE_LIKE_DOESNT_MATCH_BLOBS
-#define SQLITE_MAX_EXPR_DEPTH 0
 #define SQLITE_STRICT_SUBTYPE 1
-#define SQLITE_USE_ALLOCA
 #define SQLITE_OMIT_DEPRECATED
 #define SQLITE_OMIT_SHARED_CACHE
 #define SQLITE_OMIT_AUTOINIT
 
 // We need these:
 // #define SQLITE_DEFAULT_MEMSTATUS 0
+// #define SQLITE_MAX_EXPR_DEPTH 0
+// #define SQLITE_USE_ALLOCA
 // #define SQLITE_OMIT_DECLTYPE
 // #define SQLITE_OMIT_PROGRESS_CALLBACK
 
-// TODO add this:
-// #define SQLITE_ENABLE_API_ARMOR
-
 // Other Options
 
 #define SQLITE_ALLOW_URI_AUTHORITY
 #define SQLITE_TRUSTED_SCHEMA 0
 #define SQLITE_DEFAULT_FOREIGN_KEYS 1
+#define SQLITE_ENABLE_API_ARMOR
 #define SQLITE_ENABLE_ATOMIC_WRITE
 #define SQLITE_ENABLE_BATCH_ATOMIC_WRITE
 #define SQLITE_ENABLE_COLUMN_METADATA

sqlite3/stmt.c

@@ -2,6 +2,11 @@
 
 #include "sqlite3.h"
 
+int sqlite3_exec_go(sqlite3_stmt *stmt) {
+  while (sqlite3_step(stmt) == SQLITE_ROW);
+  return sqlite3_reset(stmt);
+}
+
 union sqlite3_data {
   sqlite3_int64 i;
   double d;
@@ -16,7 +21,7 @@ int sqlite3_columns_go(sqlite3_stmt *stmt, int nCol, char *aType,
   if (nCol != sqlite3_column_count(stmt)) {
     return SQLITE_MISUSE;
   }
-  int rc = SQLITE_OK;
+  bool check = false;
   for (int i = 0; i < nCol; ++i) {
     const void *ptr = NULL;
     switch (aType[i] = sqlite3_column_type(stmt, i)) {
@@ -36,16 +41,14 @@ int sqlite3_columns_go(sqlite3_stmt *stmt, int nCol, char *aType,
         ptr = sqlite3_column_blob(stmt, i);
         break;
     }
-    if (ptr == NULL && rc == SQLITE_OK) {
-      rc = sqlite3_errcode(sqlite3_db_handle(stmt));
-      if (rc == SQLITE_ROW || rc == SQLITE_DONE) {
-        rc = SQLITE_OK;
-      }
-    }
     aData[i].ptr = ptr;
     aData[i].len = sqlite3_column_bytes(stmt, i);
+    if (ptr == NULL) check = true;
   }
-  return rc;
+  if (check && SQLITE_NOMEM == sqlite3_errcode(sqlite3_db_handle(stmt))) {
+    return SQLITE_NOMEM;
+  }
+  return SQLITE_OK;
 }
 
 static_assert(offsetof(union sqlite3_data, i) == 0, "Unexpected offset");

sqlite3/text.c

@@ -10,4 +10,14 @@ int sqlite3_bind_text_go(sqlite3_stmt *stmt, int i, const char *zData,
 int sqlite3_bind_blob_go(sqlite3_stmt *stmt, int i, const char *zData,
                          sqlite3_uint64 nData) {
   return sqlite3_bind_blob64(stmt, i, zData, nData, &sqlite3_free);
+}
+
+void sqlite3_result_text_go(sqlite3_context *ctx, const char *zData,
+                            sqlite3_uint64 nData) {
+  sqlite3_result_text64(ctx, zData, nData, &sqlite3_free, SQLITE_UTF8);
+}
+
+void sqlite3_result_blob_go(sqlite3_context *ctx, const void *zData,
+                            sqlite3_uint64 nData) {
+  sqlite3_result_blob64(ctx, zData, nData, &sqlite3_free);
 }

sqlite3/tools.sh

@@ -0,0 +1,39 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+cd -P -- "$(dirname -- "$0")"
+
+ROOT=../
+
+if [[ "$OSTYPE" == "msys" || "$OSTYPE" == "cygwin" ]]; then
+  WASI_SDK="x86_64-windows"
+  BINARYEN="x86_64-windows"
+elif [[ "$OSTYPE" == "linux"* ]]; then
+  if [[ "$(uname -m)" == "x86_64" ]]; then
+    WASI_SDK="x86_64-linux"
+    BINARYEN="x86_64-linux"
+  else
+    WASI_SDK="arm64-linux"
+    BINARYEN="aarch64-linux"
+  fi
+elif [[ "$OSTYPE" == "darwin"* ]]; then
+  if [[ "$(uname -m)" == "x86_64" ]]; then
+    WASI_SDK="x86_64-macos"
+    BINARYEN="x86_64-macos"
+  else
+    WASI_SDK="arm64-macos"
+    BINARYEN="arm64-macos"
+  fi
+fi
+
+WASI_SDK="https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-25/wasi-sdk-25.0-$WASI_SDK.tar.gz"
+BINARYEN="https://github.com/WebAssembly/binaryen/releases/download/version_123/binaryen-version_123-$BINARYEN.tar.gz"
+
+# Download tools
+mkdir -p "$ROOT/tools"
+[ -d "$ROOT/tools/wasi-sdk" ] || curl -#L "$WASI_SDK" | tar xzC "$ROOT/tools" &
+[ -d "$ROOT/tools/binaryen" ] || curl -#L "$BINARYEN" | tar xzC "$ROOT/tools" &
+wait
+
+[ -d "$ROOT/tools/wasi-sdk" ] || mv "$ROOT/tools/wasi-sdk"* "$ROOT/tools/wasi-sdk"
+[ -d "$ROOT/tools/binaryen" ] || mv "$ROOT/tools/binaryen"* "$ROOT/tools/binaryen"

sqlite3/vfs.c

@@ -137,9 +137,10 @@ sqlite3_vfs *sqlite3_vfs_find(const char *zVfsName) {
 
   // Create a new C wrapper.
   sqlite3_vfs *head = go_vfs_list;
-  go_vfs_list = malloc(sizeof(sqlite3_vfs) + strlen(zVfsName) + 1);
+  size_t vfsNameLen = strlen(zVfsName);
+  go_vfs_list = malloc(sizeof(sqlite3_vfs) + vfsNameLen + 1);
   char *name = (char *)(go_vfs_list + 1);
-  strcpy(name, zVfsName);
+  memcpy(name, zVfsName, vfsNameLen + 1);
   *go_vfs_list = (sqlite3_vfs){
       .iVersion = 2,
       .szOsFile = sizeof(struct go_file),

stmt.go

@@ -110,10 +110,7 @@ func (s *Stmt) Step() bool {
 		s.err = INTERRUPT
 		return false
 	}
-	return s.step()
-}
 
-func (s *Stmt) step() bool {
 	rc := res_t(s.c.call("sqlite3_step", stk_t(s.handle)))
 	switch rc {
 	case _ROW:
@@ -141,10 +138,9 @@ func (s *Stmt) Exec() error {
 	if s.c.interrupt.Err() != nil {
 		return INTERRUPT
 	}
-	// TODO: implement this in C.
-	for s.step() {
-	}
-	return s.Reset()
+	rc := res_t(s.c.call("sqlite3_exec_go", stk_t(s.handle)))
+	s.err = nil
+	return s.c.error(rc)
 }
 
 // Status monitors the performance characteristics of prepared statements.
@@ -649,6 +645,7 @@ func (s *Stmt) ColumnValue(col int) Value {
 // [FLOAT] as float64, [NULL] as nil,
 // [TEXT] as string, and [BLOB] as []byte.
 func (s *Stmt) Columns(dest ...any) error {
+	defer s.c.arena.mark()()
 	types, ptr, err := s.columns(int64(len(dest)))
 	if err != nil {
 		return err
@@ -701,6 +698,7 @@ func (s *Stmt) Columns(dest ...any) error {
 // Any []byte are owned by SQLite and may be invalidated by
 // subsequent calls to [Stmt] methods.
 func (s *Stmt) ColumnsRaw(dest ...any) error {
+	defer s.c.arena.mark()()
 	types, ptr, err := s.columns(int64(len(dest)))
 	if err != nil {
 		return err
@@ -739,7 +737,6 @@ func (s *Stmt) ColumnsRaw(dest ...any) error {
 }
 
 func (s *Stmt) columns(count int64) ([]byte, ptr_t, error) {
-	defer s.c.arena.mark()()
 	typePtr := s.c.arena.new(count)
 	dataPtr := s.c.arena.new(count * 8)
 

tests/parallel/parallel_test.go

@@ -35,7 +35,7 @@ func TestMain(m *testing.M) {
 			log.Printf("%v (%d): %s", code, code, msg)
 		}
 	})
-	m.Run()
+	os.Exit(m.Run())
 }
 
 func Test_parallel(t *testing.T) {

tests/txn_test.go

@@ -287,8 +287,9 @@ func TestConn_Transaction_busy(t *testing.T) {
 	go cancel()
 
 	_, err = db2.BeginExclusive()
-	if !errors.Is(err, sqlite3.BUSY) && !errors.Is(err, sqlite3.INTERRUPT) {
-		t.Errorf("got %v, want sqlite3.BUSY or sqlite3.INTERRUPT", err)
+	var terr interface{ Temporary() bool }
+	if !errors.As(err, &terr) || !terr.Temporary() {
+		t.Errorf("got %v, want temporary", err)
 	}
 
 	err = nil

util/sql3util/wasm/build.sh

@@ -10,7 +10,9 @@ WASI_SDK="$ROOT/tools/wasi-sdk/bin"
 trap 'rm -f sql3parse_table.tmp' EXIT
 
 "$WASI_SDK/clang" --target=wasm32-wasi -std=c23 -g0 -Oz \
-	-Wall -Wextra -o sql3parse_table.wasm main.c \
+	-Wall -Wextra -Wno-unused-parameter -Wno-unused-function \
+	-o sql3parse_table.wasm main.c \
+	-I"$ROOT/sqlite3/libc" -I"$ROOT/sqlite3" \
 	-mexec-model=reactor \
 	-msimd128 -mmutable-globals -mmultivalue \
 	-mbulk-memory -mreference-types \
@@ -22,7 +24,7 @@ trap 'rm -f sql3parse_table.tmp' EXIT
 
 "$BINARYEN/wasm-ctor-eval" -c _initialize sql3parse_table.wasm -o sql3parse_table.tmp
 "$BINARYEN/wasm-opt" --strip --strip-debug --strip-producers -c -Oz \
-	sql3parse_table.tmp -o sql3parse_table.wasm \
+	sql3parse_table.tmp -o sql3parse_table.wasm --low-memory-unused \
 	--enable-simd --enable-mutable-globals --enable-multivalue \
 	--enable-bulk-memory --enable-reference-types \
 	--enable-nontrapping-float-to-int --enable-sign-ext

vfs/tests/mptest/wasm/build.sh

@@ -9,7 +9,7 @@ WASI_SDK="$ROOT/tools/wasi-sdk/bin"
 
 "$WASI_SDK/clang" --target=wasm32-wasi -std=c23 -g0 -O2 \
 	-o mptest.wasm main.c \
-	-I"$ROOT/sqlite3" \
+	-I"$ROOT/sqlite3/libc" -I"$ROOT/sqlite3" \
 	-msimd128 -mmutable-globals -mmultivalue \
 	-mbulk-memory -mreference-types \
 	-mnontrapping-fptoint -msign-ext \
@@ -26,7 +26,7 @@ WASI_SDK="$ROOT/tools/wasi-sdk/bin"
 	$(awk '{print "-Wl,--export="$0}' exports.txt)
 
 "$BINARYEN/wasm-opt" -g --strip --strip-producers -c -O3 \
-	mptest.wasm -o mptest.tmp \
+	mptest.wasm -o mptest.tmp --low-memory-unused \
 	--enable-simd --enable-mutable-globals --enable-multivalue \
 	--enable-bulk-memory --enable-reference-types \
 	--enable-nontrapping-float-to-int --enable-sign-ext

vfs/tests/speedtest1/wasm/build.sh

@@ -9,7 +9,7 @@ WASI_SDK="$ROOT/tools/wasi-sdk/bin"
 
 "$WASI_SDK/clang" --target=wasm32-wasi -std=c23 -g0 -O2 \
 	-o speedtest1.wasm main.c \
-	-I"$ROOT/sqlite3" \
+	-I"$ROOT/sqlite3/libc" -I"$ROOT/sqlite3" \
 	-msimd128 -mmutable-globals -mmultivalue \
 	-mbulk-memory -mreference-types \
 	-mnontrapping-fptoint -msign-ext \
@@ -21,7 +21,7 @@ WASI_SDK="$ROOT/tools/wasi-sdk/bin"
 	$(awk '{print "-Wl,--export="$0}' exports.txt)
 
 "$BINARYEN/wasm-opt" -g --strip --strip-producers -c -O3 \
-	speedtest1.wasm -o speedtest1.tmp \
+	speedtest1.wasm -o speedtest1.tmp --low-memory-unused \
 	--enable-simd --enable-mutable-globals --enable-multivalue \
 	--enable-bulk-memory --enable-reference-types \
 	--enable-nontrapping-float-to-int --enable-sign-ext