zig-sqlite
This package is a thin wrapper around sqlite's C API.
Maintainer note: I'm currently on a break working with Zig and don't intend to work on new features for zig-sqlite. I will keep it updated for the latest Zig versions because that doesn't take too much of my time.
Status
While the core functionality works right now, the API is still subject to changes.
If you use this library, expect to have to make changes when you update the code.
Zig release support
zig-sqlite only tracks Zig master (as can be found here). The plan is to support releases once Zig 1.0 is released but this can still change.
Unfortunately as of zig 0.11.0-dev.3312+ab37ab33c some things still don't compile properly, for example our tests don't compile. However this demo program compiles fine.
So your mileage may vary if you try to use zig-sqlite.
Table of contents
Requirements
Zig master is the only required dependency.
For sqlite, you have options depending on your target:
- On Windows the only supported way at the moment to build
zig-sqliteis with the bundled sqlite source code file. - On Linux we have two options:
- use the system and development package for sqlite (
libsqlite3-devfor Debian and derivatives,sqlite3-develfor Fedora) - use the bundled sqlite source code file.
- use the system and development package for sqlite (
Features
- Preparing, executing statements
- comptime checked bind parameters
- user defined SQL functions
Installation
There are three main ways to include zig-sqlite in your project:
- using zig's official package manager
- using the zigmod package manager
- using a git submodule
Official package manager
Add this as one of the .dependencies inside your build.zig.zon file:
.sqlite = .{
.url = "https://github.com/vrischmann/zig-sqlite/archive/COMMIT.tar.gz",
},Now in your build.zig you can access the module like this:
const sqlite = b.dependency("sqlite", .{
.target = target,
.optimize = optimize,
});
exe.addModule("sqlite", sqlite.module("sqlite"));
// links the bundled sqlite3, so leave this out if you link the system one
exe.linkLibrary(sqlite.artifact("sqlite"));zigmod
Add this to your zig.mod file:
dependencies:
- src: git https://github.com/vrischmann/zig-sqlite branch-master
Note that if you're building an executable and not a library you should use dev_dependencies instead.
Next run zigmod fetch; it should create a deps.zig file.
Now in your build.zig you can access the package like this:
const deps = @import("deps.zig");
...
deps.addAllTo(exe);This is the easiest way to add zig-sqlite because it uses the bundled source code, avoiding all sorts of linking problems.
Git submodule
If you don't want to use a package manager you can simply add this repository as a git submodule.
Then you need to chose if you want to use the system sqlite library or the bundled source code.
Using the system sqlite library
If you want to use the system sqlite library, add the following to your build.zig target(s):
exe.linkLibC();
exe.linkSystemLibrary("sqlite3");
exe.addPackage(.{ .name = "sqlite", .path = "third_party/zig-sqlite/sqlite.zig" });Using the bundled sqlite source code file
If you want to use the bundled sqlite source code file, first you need to add it as a static library in your build.zig file:
const sqlite = b.addStaticLibrary("sqlite", null);
sqlite.addCSourceFile("third_party/zig-sqlite/c/sqlite3.c", &[_][]const u8{"-std=c99"});
sqlite.linkLibC();If you need to define custom compile-time options for sqlite, modify the flags (second argument to addCSourceFile).
Now it's just a matter of linking your build.zig target(s) to this library instead of the system one:
exe.linkLibrary(sqlite);
exe.addPackagePath("sqlite", "third_party/zig-sqlite/sqlite.zig");
exe.addIncludeDir("third_party/zig-sqlite/c");If you're building with glibc you must make sure that the version used is at least 2.28.
You can do that in your build.zig file:
var target = b.standardTargetOptions(.{});
target.setGnuLibCVersion(2, 28, 0);
exe.setTarget(target);Or with -Dtarget:
$ zig build -Dtarget=native-linux-gnu.2.28
Usage
Import zig-sqlite like this:
const sqlite = @import("sqlite");Initialization
You must create and initialize an instance of sqlite.Db:
var db = try sqlite.Db.init(.{
.mode = sqlite.Db.Mode{ .File = "/home/vincent/mydata.db" },
.open_flags = .{
.write = true,
.create = true,
},
.threading_mode = .MultiThread,
});The init method takes a InitOptions struct which will be used to configure sqlite.
Only the mode field is mandatory, the other fields have sane default values.
Preparing a statement
Common use
sqlite works exclusively by using prepared statements. The wrapper type is sqlite.Statement. Here is how you get one:
const query =
\\SELECT id, name, age, salary FROM employees WHERE age > ? AND age < ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();The Db.prepare method takes a comptime query string.
Diagnostics
If you want failure diagnostics you can use prepareWithDiags like this:
var diags = sqlite.Diagnostics{};
var stmt = db.prepareWithDiags(query, .{ .diags = &diags }) catch |err| {
std.log.err("unable to prepare statement, got error {}. diagnostics: {s}", .{ err, diags });
return err;
};
defer stmt.deinit();Executing a statement
For queries which do not return data (INSERT, UPDATE) you can use the exec method:
const query =
\\UPDATE foo SET salary = ? WHERE id = ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();
try stmt.exec(.{
.salary = 20000,
.id = 40,
});See the section "Bind parameters and resultset rows" for more information on the types mapping rules.
Reuse a statement
You can reuse a statement by resetting it like this:
const query =
\\UPDATE foo SET salary = ? WHERE id = ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();
var id: usize = 0;
while (id < 20) : (id += 1) {
stmt.reset();
try stmt.exec(.{
.salary = 2000,
.id = id,
});
}Reading data
For queries which return data you have multiple options:
Statement.allwhich takes an allocator and can allocate memory.Statement.onewhich does not take an allocator and cannot allocate memory (aside from what sqlite allocates itself).Statement.oneAllocwhich takes an allocator and can allocate memory.
Type parameter
All these methods take a type as first parameter.
The type represents a "row", it can be:
- a struct where each field maps to the corresponding column in the resultset (so field 0 must map to column 1 and so on).
- a single type, in that case the resultset must only return one column.
The type can be a pointer but only when using the methods taking an allocator.
Not all types are allowed, see the section "Bind parameters and resultset rows" for more information on the types mapping rules.
Non allocating
Using one:
const query =
\\SELECT name, age FROM employees WHERE id = ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();
const row = try stmt.one(
struct {
name: [128:0]u8,
age: usize,
},
.{},
.{ .id = 20 },
);
if (row) |row| {
std.log.debug("name: {}, age: {}", .{std.mem.spanZ(&row.name), row.age});
}Notice that to read text we need to use a 0-terminated array; if the name column is bigger than 127 bytes the call to one will fail.
If the length of the data is variable then the sentinel is mandatory: without one there would be no way to know where the data ends in the array.
However if the length is fixed, you can read into a non 0-terminated array, for example:
const query =
\\SELECT id FROM employees WHERE name = ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();
const row = try stmt.one(
[16]u8,
.{},
.{ .name = "Vincent" },
);
if (row) |id| {
std.log.debug("id: {s}", .{std.fmt.fmtSliceHexLower(&id)});
}If the column data doesn't have the correct length a error.ArraySizeMismatch will be returned.
The convenience function sqlite.Db.one works exactly the same way:
const query =
\\SELECT age FROM employees WHERE id = ?
;
const row = try db.one(usize, query, .{}, .{ .id = 20 });
if (row) |age| {
std.log.debug("age: {}", .{age});
}Allocating
Using all:
const query =
\\SELECT name FROM employees WHERE age > ? AND age < ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();
const names = try stmt.all([]const u8, allocator, .{}, .{
.age1 = 20,
.age2 = 40,
});
for (names) |name| {
std.log.debug("name: {s}", .{ name });
}Using oneAlloc:
const query =
\\SELECT name FROM employees WHERE id = ?
;
var stmt = try db.prepare(query);
defer stmt.deinit();
const row = try stmt.oneAlloc([]const u8, allocator, .{}, .{
.id = 200,
});
if (row) |name| {
std.log.debug("name: {}", .{name});
}Iterating
Another way to get the data returned by a query is to use the sqlite.Iterator type.
You can only get one by calling the iterator method on a statement.
The iterator method takes a type which is the same as with all, one or oneAlloc: every row retrieved by calling next or nextAlloc will have this type.
Iterating is done by calling the next or nextAlloc method on an iterator. Just like before, next cannot allocate memory while nextAlloc can allocate memory.
next or nextAlloc will either return an optional value or an error; you should keep iterating until null is returned.
Non allocating
var stmt = try db.prepare("SELECT age FROM user WHERE age < ?");
defer stmt.deinit();
var iter = try stmt.iterator(usize, .{
.age = 20,
});
while (try iter.next(.{})) |age| {
std.debug.print("age: {}\n", .{age});
}Allocating
var stmt = try db.prepare("SELECT name FROM user WHERE age < ?");
defer stmt.deinit();
var iter = try stmt.iterator([]const u8, .{
.age = 20,
});
while (true) {
var arena = std.heap.ArenaAllocator.init(allocator);
defer arena.deinit();
const name = (try iter.nextAlloc(arena.allocator(), .{})) orelse break;
std.debug.print("name: {}\n", .{name});
}Bind parameters and resultset rows
Since sqlite doesn't have many types only a small number of Zig types are allowed in binding parameters and in resultset mapping types.
Here are the rules for bind parameters:
- any Zig
IntorComptimeIntis treated as aINTEGER. - any Zig
FloatorComptimeFloatis treated as aREAL. []const u8,[]u8is treated as aTEXT.- the custom
sqlite.Blobtype is treated as aBLOB. - the custom
sqlite.Texttype is treated as aTEXT. - the
nullvalue is treated as aNULL. - non-null optionals are treated like a regular value, null optionals are treated as a
NULL.
Here are the rules for resultset rows:
INTEGERcan be read into any ZigIntprovided the data fits.REALcan be read into any ZigFloatprovided the data fits.TEXTcan be read into a[]const u8or[]u8.TEXTcan be read into any array ofu8with a sentinel provided the data fits.BLOBfollows the same rules asTEXT.NULLcan be read into any optional.
Note that arrays must have a sentinel because we need a way to communicate where the data actually stops in the array, so for example use [200:0]u8 for a TEXT field.
Custom type binding and reading
Sometimes the default field binding or reading logic is not what you want, for example if you want to store an enum using its tag name instead of its integer value or if you want to store a byte slice as an hex string.
To accomplish this you must first define a wrapper struct for your type. For example if your type is a [4]u8 and you want to treat it as an integer:
pub const MyArray = struct {
data: [4]u8,
pub const BaseType = u32;
pub fn bindField(self: MyArray, _: std.mem.Allocator) !BaseType {
return std.mem.readIntNative(BaseType, &self.data);
}
pub fn readField(_: std.mem.Allocator, value: BaseType) !MyArray {
var arr: MyArray = undefined;
std.mem.writeIntNative(BaseType, &arr.data, value);
return arr;
}
};Now when you bind a value of type MyArray the value returned by bindField will be used for binding instead.
Same for reading, when you select into a MyArray row or field the value returned by readField will be used instead.
NOTE: when you do allocate in bindField or readField make sure to pass a std.heap.ArenaAllocator-based allocator.
The binding or reading code does not keep tracking of allocations made in custom types so it can't free the allocated data itself; it's therefore required to use an arena to prevent memory leaks.
Note about complex allocations
Depending on your queries and types there can be a lot of allocations required. Take the following example:
const User = struct {
id: usize,
first_name: []const u8,
last_name: []const u8,
data: []const u8,
};
fn fetchUsers(allocator: std.mem.Allocator, db: *sqlite.Db) ![]User {
var stmt = try db.prepare("SELECT id FROM user WHERE id > $id");
defer stmt.deinit();
return stmt.all(User, allocator, .{}, .{ .id = 20 });
}This will do multiple allocations:
- one for each id field in the
Usertype - one for the resulting slice
To facilitate memory handling, consider using an arena allocator like this:
var arena = std.heap.ArenaAllocator.init(allocator);
defer arena.deinit();
const users = try fetchUsers(arena.allocator(), db);
_ = users;This is especially recommended if you use custom types that allocate memory since, as noted above, it's necessary to prevent memory leaks.
Comptime checks
Prepared statements contain comptime metadata which is used to validate every call to exec, one and all at compile time.
Check the number of bind parameters.
The first check makes sure you provide the same number of bind parameters as there are bind markers in the query string.
Take the following code:
var stmt = try db.prepare("SELECT id FROM user WHERE age > ? AND age < ? AND weight > ?");
defer stmt.deinit();
const rows = try stmt.all(usize, .{}, .{
.age_1 = 10,
.age_2 = 20,
});
_ = rows;It fails with this compilation error:
/home/vincent/dev/perso/libs/zig-sqlite/sqlite.zig:738:17: error: number of bind markers not equal to number of fields
@compileError("number of bind markers not equal to number of fields");
^
/home/vincent/dev/perso/libs/zig-sqlite/sqlite.zig:817:22: note: called from here
self.bind(values);
^
/home/vincent/dev/perso/libs/zig-sqlite/sqlite.zig:905:41: note: called from here
var iter = try self.iterator(Type, values);
^
./src/main.zig:19:30: note: called from here
const rows = try stmt.all(usize, allocator, .{}, .{
^
./src/main.zig:5:29: note: called from here
pub fn main() anyerror!void {
Assign types to bind markers and check them.
The second (and more interesting) check makes sure you provide appropriately typed values as bind parameters.
This check is not automatic since with a standard SQL query we have no way to know the types of the bind parameters, to use it you must provide theses types in the SQL query with a custom syntax.
For example, take the same code as above but now we also bind the last parameter:
var stmt = try db.prepare("SELECT id FROM user WHERE age > ? AND age < ? AND weight > ?");
defer stmt.deinit();
const rows = try stmt.all(usize, .{ .allocator = allocator }, .{
.age_1 = 10,
.age_2 = 20,
.weight = false,
});
_ = rows;This compiles correctly even if the weight field in our user table is of the type INTEGER.
We can make sure the bind parameters have the right type if we rewrite the query like this:
var stmt = try db.prepare("SELECT id FROM user WHERE age > ? AND age < ? AND weight > ?{usize}");
defer stmt.deinit();
const rows = try stmt.all(usize, .{ .allocator = allocator }, .{
.age_1 = 10,
.age_2 = 20,
.weight = false,
});
_ = rows;Now this fails to compile:
/home/vincent/dev/perso/libs/zig-sqlite/sqlite.zig:745:25: error: value type bool is not the bind marker type usize
@compileError("value type " ++ @typeName(struct_field.field_type) ++ " is not the bind marker type " ++ @typeName(typ));
^
/home/vincent/dev/perso/libs/zig-sqlite/sqlite.zig:817:22: note: called from here
self.bind(values);
^
/home/vincent/dev/perso/libs/zig-sqlite/sqlite.zig:905:41: note: called from here
var iter = try self.iterator(Type, values);
^
./src/main.zig:19:30: note: called from here
const rows = try stmt.all(usize, allocator, .{}, .{
^
./src/main.zig:5:29: note: called from here
pub fn main() anyerror!void {
The syntax is straightforward: a bind marker ? followed by {, a Zig type name and finally }.
There are a limited number of types allowed currently:
- all integer types.
- all arbitrary bit-width integer types.
- all float types.
- bool.
- strings with
[]const u8or[]u8. - strings with
sqlite.Text. - blobs with
sqlite.Blob.
It's probably possible to support arbitrary types if they can be marshaled to a sqlite type. This is something to investigate.
NOTE: this is done at compile time and is quite CPU intensive, therefore it's possible you'll have to play with @setEvalBranchQuota to make it compile.
To finish our example, passing the proper type allows it compile:
var stmt = try db.prepare("SELECT id FROM user WHERE age > ? AND age < ? AND weight > ?{usize}");
defer stmt.deinit();
const rows = try stmt.all(usize, .{}, .{
.age_1 = 10,
.age_2 = 20,
.weight = @as(usize, 200),
});
_ = rows;User defined SQL functions
sqlite supports user-defined SQL functions which come in two types:
- scalar functions
- aggregate functions
In both cases the arguments are sqlite3_values and are converted to Zig values using the following rules:
TEXTvalues can be eithersqlite.Textor[]const u8BLOBvalues can be eithersqlite.Blobor[]const u8INTEGERvalues can be any Zig integerREALvalues can be any Zig float
Scalar functions
You can define a scalar function using db.createScalarFunction:
try db.createScalarFunction(
"blake3",
struct {
fn run(input: []const u8) [std.crypto.hash.Blake3.digest_length]u8 {
var hash: [std.crypto.hash.Blake3.digest_length]u8 = undefined;
std.crypto.hash.Blake3.hash(input, &hash, .{});
return hash;
}
}.run,
.{},
);
const hash = try db.one([std.crypto.hash.Blake3.digest_length]u8, "SELECT blake3('hello')", .{}, .{});Each input arguments in the function call in the statement is passed on to the registered run function.
Aggregate functions
You can define a scalar function using db.createAggregateFunction:
const MyContext = struct {
sum: u32,
};
var my_ctx = MyContext{ .sum = 0 };
try db.createAggregateFunction(
"mySum",
&my_ctx,
struct {
fn step(ctx: *MyContext, input: u32) void {
ctx.sum += input;
}
}.step,
struct {
fn finalize(ctx: *MyContext) u32 {
return ctx.sum;
}
}.finalize,
.{},
);
const result = try db.one(usize, "SELECT mySum(nb) FROM foobar", .{}, .{});Each input arguments in the function call in the statement is passed on to the registered step function.
The finalize function is called once at the end.
The context (2nd argument of createAggregateFunction) can be whatever you want; both step and finalize function must
have their first argument of the same type as the context.