arteymix / numeric-glib

Numeric data types for GLib via GCC extensions

Numeric-GLib

Numeric data types for GLib via GCC extensions

Features

• 128 bit integers with int128 and uint128
• float80 and quad-precision float128
• decimal with decimal32, decimal64 and decimal128
• C99 complex, complex80 and complex128
• vectorized operations on supported types
• little and big endian variants with _le and _be suffix

All types are prefixed with numeric_, but can be used as-is by importing the Numeric namespace in Vala.

Usage

In Vala, one can simply add --pkg=numeric-glib-1.0 to include the type definitions and link against the shared library to initialize all the types and transformations.

vala --pkg=numeric-glib-1.0 main.vala

Or via Meson, the numeric_glib dependency can be included in any target.

project('Foo')

glib = dependency('glib-2.0')
gobject = dependency('gobject-2.0')
numeric_glib = dependency('numeric-glib-1.0', fallback: ['numeric-glib', 'numeric_glib_dep'])
numeric_glib_vala = dependency('numeric-glib-1.0', fallback: ['numeric-glib', 'numeric_glib_vala_dep'])

executable('foo', 'foo.c', dependencies: [glib, gobject, numeric_glib])
executable('foo', 'foo.vala', dependencies: [glib, gobject, numeric_glib_vala])

numeric-glib's Issues

Generate code for byte order conversion with G_DEFINE_BOXED_TYPE_WITH_CODE

Most of numeric-transform.c could be automated with macros given information about origin/destination byte order and registered automatically with G_DEFINE_BOXED_TYPE_WITH_CODE.

libquadmath printf hook does not work

Setup a minimal CI with github action

Add tests for GVariant serialization

Can we use + - operators?

Usage with printf

I'm trying to debug some code but I can't figure out how to print the value of a decimal128 to the screen. Can it be used in a printf statement?

Endianess

I think we can address endianess of types in Numeric-GLib by defining type with explicit endianess (e.g. Numeric.float_be. The type would be boxed and conversion would bring it to a native endianess for further processing.

Use G_DEFINE_BOXED_TYPE_WITH_CODE to register type transformations

Also apply to #17.

name

vala-gnumath vs numeric-glib.

Is decimal128 broken?

private static inline void test_foo () {
    Numeric.decimal128 a = 0.1;
    Numeric.decimal128 b = 0.1;
    Numeric.decimal128 c = 0.1;
    Numeric.decimal128 d = -0.3;

    Numeric.decimal128 expected = 0;

    assert_true (a + b + c + d == expected);
}

This test fails for me. It works if I replace decimal128 with decimal64 though. What's going on?

Cant build

../src/numeric-types.h:38:50: error: two or more data types in declaration specifiers
 typedef __float128                               _Float128;
                                                  ^~~~~~~~~

Can't use decimal128 for properties in a class inheriting from GLib.Object

It seems decimal128 can't be used for properties in classes inheriting from GLib.Object.

Failing Code

public class Foo : GLib.Object {
    public Numeric.decimal128 bar {
        get { return 0; }
    }
}

Error Output

numeric-glib-1.0.vapi:131.2-131.25: error: The type `Numeric.decimal128` doesn't declare a GValue set function
	public struct decimal128

Working Code

public class Foo {
    public Numeric.decimal128 bar {
        get { return 0; }
    }
}

Use GIR introspection to generate the VAPI

Type size information.

probably add the instance_size database here?