Change Log: 2.083.0

Prior to this release, the following code did not build in -betterC, but will with this release.

// (Prior to this release) Error: TypeInfo cannot be used with -betterC
struct Sint
{
    int x;
    this(int v) { x = v;}
}

extern(C) void main()
{
    Sint[6] a1 = [Sint(1), Sint(2), Sint(3), Sint(1), Sint(2), Sint(3)];
    assert(a1[0..3] == a1[3..$]);
}

// (Prior to this release) Linker Error: undefined reference to `core.internal.string.dstrcmp`
extern(C) void main()
{
    auto s = "abc";
    switch(s)
    {
        case "abc":
            break;
        default:
            break;
    }
}

// (Prior to this release) Linker Error: undefined reference to `core.internal.string.dstrcmp`
extern(C) void main()
{
    char[6] a = [1,2,3,1,2,3];
    assert(a[0..3] >= a[3..$]);  // failed for any byte-sized type (e.g. `char`, `ubyte`, etc...)
}

D_ModuleInfo, D_Exceptions, and D_TypeInfo version identifiers were added to allow better precision in druntime and other library implementations.

See Predefined Versions

When a conditional expression is the left operand of an assign expression, they would be parsed as an assignment to the result of the conditional. This has now been deprecated. To preserve the existing behaviour, add parentheses for the conditional without including the assignment:

bool test;
int a, b, c;
...
test ? a = b : c = 2;   // Deprecated
(test ? a = b : c) = 2; // Equivalent

This makes the intent clearer, because the first statement can easily be misread as the following code:

test ? a = b : (c = 2);

When writing debug code, one isn't interested in the type safety, but a pleasant debugging experience. The type checker already allowed to escape pure and @nogc within debug statement. With this release, @system code can be called from debug statements too:

void main()
{
    int[] arr = [1, 2];
    debug unsafeCode(arr);
}

@system void unsafeCode(T)(T[] arr)
{
    import core.stdc.stdio;
    printf("arr.ptr: %p", arr.ptr);
}

opDot was the D1 analog to alias this. However, alias this covers all use cases of opDot, but ensures safety.

struct S
{
    int a, b;
}
struct T
{
    S s;

    S* opDot()
    {
        return &s;
    }
}

void main()
{
    T t;
    t.a = 4;
    assert(t.a == 4);
    t.b = 5;
}

With alias this:

struct S
{
    int a, b;
}
struct T
{
    S s;

    alias s this;
}

void main() @safe
{
    T t;
    t.a = 4;
    assert(t.a == 4);
    t.b = 5;
}

Example

template AliasSeq(TList...)
{
    alias AliasSeq = TList;
}

enum MyEnum
{
    @("uda0") value0,
    @disable value1,
    deprecated value2  // Deprecation: enum member `main.MyEnum.value2` is deprecated
}

static assert(__traits(getAttributes, MyEnum.value0) == AliasSeq!("uda0"));

void main()
{
    auto v1 = MyEnum.value1;  // Error: enum member `main.MyEnum.value1` cannot be used because it is annotated with `@disable`
}

The deprecation period for C-style array declarations has expired. Beginning with this release, declaring arrays with C-style syntax will result in an error.

int a[2];  // Error: instead of C-style syntax, use D-style `int a[2]`
int[2] b;  // OK

It used to ignore the const and was missing the $$C escape prefix.

extern(C++) struct foo(T) {}

extern(C++) void test(foo!(const(char)) a) {}
// New
static assert(test.mangleof == "?test@@YAXU?$foo@$$CBD@@@Z");
// Old
//static assert(test.mangleof == "?test@@YAXU?$foo@D@@@Z");

To solve the fragile base class problem [1] in Objective-C, fields have a dynamic offset instead of a static offset. The compiler outputs a statically known offset which later the dynamic loader can update, if necessary, when the application is loaded. Due to this behavior it doesn't make sense to be able to get the offset of a field at compile time, because this offset might not actually be the same at runtime.

To get the offset or value of a field, that is correct at runtime, functionality from the Objective-C runtime can be used instead [2].

[1] https://en.wikipedia.org/wiki/Fragile_binary_interface_problem [2] https://developer.apple.com/documentation/objectivec/objective_c_runtime

Direct interop of D size_t and C++ size_t was working before 2.079, on all platforms except for 32-bit OS X. By mangling D long as C++ long long on 64-bit OS X starting with 2.079, size_t interop broke on a more relevant platform. With new/fixed aliases, e.g., core.stdc.stdint.int64_t, core.stdc.stdint.uint64_t and core.stdc.config.cpp_size_t, there are now proper tools for portable C++ interop wrt. integers. Reverting to the previous C++ mangling on 64-bit OS X (C++ long) may save mixed D/C++ code bases from the need of manual adaptations by skipping the 2.079-2.081 DMD versions.

User-defined attributes on function parameters behave analogous to existing UDAs:

void example(@(22) string param)
{
    @(11) string var;
    static assert([__traits(getAttributes, var)] == [11]);
    static assert([__traits(getAttributes, param)] == [22]);
}

Previously it was only possible to disable the trapping of exception by setting the global variable rt_trapExceptions to false. Now you can, for example, immediately open gdb at the uncaught exception:

> gdb -ex run --args  --DRT-trapExceptions=0
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib/libthread_db.so.1".
uncaught exception
object.Exception@src/rt_trap_exceptions_drt.d(4): foo
––––––––––
src/rt_trap_exceptions_drt.d:4 void rt_trap_exceptions_drt.test() [0x55591026]
src/rt_trap_exceptions_drt.d:9 _Dmain [0x55591058]

Program received signal SIGABRT, Aborted.
0x00007ffff6e7b86b in raise () from /usr/lib/libc.so.6
(gdb) bt full
#0  0x00007ffff6e7b86b in raise () from /usr/lib/libc.so.6
No symbol table info available.
#1  0x00007ffff6e6640e in abort () from /usr/lib/libc.so.6
No symbol table info available.
#2  0x00005555555918cc in _d_throwdwarf (o=0x7ffff7ea4000) at src/rt/dwarfeh.d:233
        eh = 0x7ffff7fa4740
        refcount = 0
        r = 5
#3  0x0000555555591027 in rt_trap_exceptions_drt.test() () at ../../src/object.d:2695
        innerLocal = 20
#4  0x0000555555591059 in D main (args=...) at src/rt_trap_exceptions_drt.d:9
        myLocal = "bar"

This Week in D for an in-depth explanation of rt_trapExceptions

The require function provides a means to construct a new value when the key is not present.

class C{}
C[string] aa;

auto a = aa.require("a", new C);    // lookup "a", construct if not present

The update function allows different operations to be performed depending on whether a value already exists or needs to be constructed.

class C{}
C[string] aa;

C older;
C newer;
aa.update("a",
{
    newer = new C;
    return newer;
},
(ref C c)
{
    older = c;
    newer = new C;
    return newer;
});

The functions avoid the need to perform multiple key lookups. Further details are available in the spec.

Now typeid(typeof(var)).getHash(&var) should be always equal hashOf(val). hashOf uses MurmurHash3 algorithm for bitwise hashing, including string hashing. Old implementation used polynomial hash for string hashing, which often provided lexicographical order of strings in associative array. However AA doesn't guarantee some specific order of elements, thus tests which rely on it should be changed.

The struct that uname returns - utsname - has the field version in C. The D definition of the struct cannot match that, because version is a keyword in D. So, it's been called update, which is not an obvious choice. It is more in line with current naming policies (and more in line with the name in C) for it to be called version_, since that's the closest that we can get to version in D. So, update has now been renamed to version_, and update is an alias of version_ which will be deprecated in a future release.

The float and double overloads of std.math.atan, std.math.atan2, std.math.tan, std.math.exp, std.math.expm1 and std.math.exp2 previously forwarded to the real implementation (inline assembly) and now got proper 'software' implementations in the corresponding precision.

While this may result in a slowdown in some cases for DMD (especially for exp() and exp2()), the overall speed-up factor for LDC is > 3, for both double and float.

This also implies less precise results, especially in single-precision, so if your code depended on more accurate results via 80-bit intermediate precision, you'll have to cast the argument(s) explicitly now.

std.algorithm.iteration.each is now capable of exiting early. When a No.each std.typecons.Flag is returned from the function that is called by each, the iteration will be aborted early. Analogously, returning Yes.each will continue the iteration. For example:

auto arr = [10, 20, 30];
arr.each!((n) { arr ~= n; return (n == 20) ? No.each : Yes.each; }); // aborts after the second iteration
assert(arr == [10, 20, 30, 10, 20]);

std.algorithm.iteration.joiner can now be requested to be a bidirectional range.

import std.algorithm.iteration : joiner;
import std.range : retro;
[[1, 2], [3, 4]].joiner.retro; // [4, 3, 2, 1]

A more complex example of inserting a format delimiter fully lazily:

import std.algorithm.comparison : equal;
import std.range : chain, cycle, iota, only, retro, take, zip;
import std.format : format;

static immutable number = "12345678";
static immutable delimiter = ",";
auto formatted = number.retro
    .zip(3.iota.cycle.take(number.length))
    .map!(z => chain(z[0].only, z[1] == 2 ? delimiter : null))
    .joiner
    .retro; // "12,345,678"

std.algorithm.skipOver which allows to move over a needle if it matches now accepts variadic arguments for needles to skip over:

auto s = "DLang.rocks";
assert(!s.skipOver("dlang", "DLF", "DLang "));
assert(s == "DLang.rocks");

assert(s.skipOver("dlang", "DLANG", "DLF", "D", "DL", "DLanp"));
assert(s == "ang.rocks");

The type of elements can be specified implicitly so that [1, 2].staticArray results in int[2], or explicitly, e.g. [1, 2].staticArray!float returns float[2]. When a is a range whose length is not known at compile time, the number of elements must be given as template argument (e.g. myrange.staticArray!2). Size and type can be combined, if the source range elements are implicitly convertible to the requested element type (eg: 2.iota.staticArray!(long[2])). When the range a is known at compile time, it can also be specified as a template argument to avoid having to specify the number of elements (e.g.: staticArray!(2.iota) or staticArray!(double, 2.iota)).

import std.range : iota;

auto input = 3.iota;
auto a = input.staticArray!2;
static assert(is(typeof(a) == int[2]));
assert(a == [0, 1]);
auto b = input.staticArray!(long[4]);
static assert(is(typeof(b) == long[4]));
assert(b == [0, 1, 2, 0]);

Due to the fact that Microsoft changes its timezone definitions too frequently to maintain the conversion between the IANA TZ Database names and the names that Microsoft uses in the standard library, several functions in std.datetime.timezone related to getting the time zone independently of the host OS were previously deprecated: TimeZone.getTimeZone, TimeZone.getInstalledTZNames, tzDatabaseNameToWindowsTZName, and windowsTZNameToTZDatabaseName. These functions have now been fully removed.

Instead of TimeZone.getTimeZone, std.datetime.timezone.PosixTimeZone.getTimeZone and std.datetime.timezone.WindowsTimeZone.getTimeZone should be used directly. Instead of Timezone.getInstalledTimeZones, std.datetime.timezone.PosixTimeZone.getInstalledTZNames and std.datetime.timezone.WindowsTimeZone.getInstalledTZNames should be used directly. And any programs looking to convert between the TZ Database names and the Windows names can use std.datetime.timezone.parseTZConversions with the current windowsZones.xml file to get the current conversions (see the documenation for std.datetime.timezone.parseTZConversions for details).

All released executables on Windows will now be code-signed by the following certificate.

    Issuer: C = US, O = DigiCert Inc, OU = www.digicert.com, CN = DigiCert SHA2 Assured ID Code Signing CA
    Serial Number: 04:e7:49:06:78:3a:e6:5f:54:37:fd:55:45:66:5f:e5
    Validity
        Not Before: Aug  7 00:00:00 2018 GMT
        Not After : Aug 11 12:00:00 2021 GMT
    Subject: C = US, ST = Washington, L = Kirkland, O = D Language Foundation, CN = D Language Foundation
    SHA1 Fingerprint: BD:E0:0F:CA:EF:6A:FA:37:15:DB:D4:AA:1A:43:2E:78:27:54:E6:60
    SHA256 Fingerprint: AA:B0:6F:7B:A8:26:52:51:C4:24:44:1B:34:8E:30:68:F4:34:DB:35:38:6A:17:A7:45:C3:27:77:11:24:C4:E5

Nightly builds and branch/feature previews will not be code-signed for now.