Just watched Don's DConf 2014 talk where he said D has to be ruthless about
memory inefficiency. Here's one thing that I think could help avoid unnecessary garbage: built-in syntax for this:

import core.stdc.stdlib : alloca;
ubyte[] buffer = (cast(ubyte*) alloca(bufsize)) [0 .. bufsize];

Often bufsize is not known at compile-time but it won't change after the buffer
allocation. So there's no reason to create garbage other than the *inconvenience*
of using alloca. Allocating in the stack seems ideal so I'd encourage that by a
clean syntax. I keep missing this feature.

On Thursday, 17 July 2014 at 16:28:00 UTC, Tero wrote:
> Just watched Don's DConf 2014 talk where he said D has to be ruthless about
> memory inefficiency. Here's one thing that I think could help avoid unnecessary garbage: built-in syntax for this:
>
> import core.stdc.stdlib : alloca;
> ubyte[] buffer = (cast(ubyte*) alloca(bufsize)) [0 .. bufsize];
>
> Often bufsize is not known at compile-time but it won't change after the buffer
> allocation. So there's no reason to create garbage other than the *inconvenience*
> of using alloca. Allocating in the stack seems ideal so I'd encourage that by a
> clean syntax. I keep missing this feature.

In theory there would probably be an allocator that uses alloca
when Andrei's std.allocator makes it in. Using alloca is actually
rather problematic in D right now though (with DMD at least)...
https://issues.dlang.org/show_bug.cgi?id=12820

Also, there used to be a built in syntax, 'scope foo = new Foo()' that would allocate on the stack, but that's deprecated now (and seemed to segfault when I tried it).

Tero:

> Allocating in the stack seems ideal so I'd encourage that by a clean syntax. I keep missing
> this feature.

See:
https://issues.dlang.org/show_bug.cgi?id=9832
http://en.cppreference.com/w/cpp/container/dynarray

Bye,
bearophile

Kapps:

> In theory there would probably be an allocator that uses alloca
> when Andrei's std.allocator makes it in.

How is it going to work?


> Also, there used to be a built in syntax, 'scope foo = new Foo()' that would allocate on the stack, but that's deprecated now

Perhaps once the "scope" (and memory lifetimes) is fleshed out, it will be un-deprecated :-)

Bye,
bearophile

On 2014-07-17 18:35, Kapps wrote:

> Also, there used to be a built in syntax, 'scope foo = new Foo()' that
> would allocate on the stack, but that's deprecated now (and seemed to
> segfault when I tried it).

It is still available and no sign of deprecation (except everyone saying that).

-- 
/Jacob Carlborg

On Thursday, 17 July 2014 at 19:41:57 UTC, Jacob Carlborg wrote:
> On 2014-07-17 18:35, Kapps wrote:
>
>> Also, there used to be a built in syntax, 'scope foo = new Foo()' that
>> would allocate on the stack, but that's deprecated now (and seemed to
>> segfault when I tried it).
>
> It is still available and no sign of deprecation (except everyone saying that).

"Discouraged and unmaintained" is probably a better description.

On 17/07/14 22:18, Dicebot wrote:

> "Discouraged and unmaintained" is probably a better description.

How does it need maintaining?

-- 
/Jacob Carlborg

On Thursday, 17 July 2014 at 16:28:00 UTC, Tero wrote:
> Just watched Don's DConf 2014 talk where he said D has to be ruthless about
> memory inefficiency. Here's one thing that I think could help avoid unnecessary garbage: built-in syntax for this:
>
> import core.stdc.stdlib : alloca;
> ubyte[] buffer = (cast(ubyte*) alloca(bufsize)) [0 .. bufsize];
>
> Often bufsize is not known at compile-time but it won't change after the buffer
> allocation. So there's no reason to create garbage other than the *inconvenience*
> of using alloca. Allocating in the stack seems ideal so I'd encourage that by a
> clean syntax. I keep missing this feature.

When talking about allocations:

The stack is just a fixed size chunk of pre-allocated memory with complicated rules and caveats for use. It's only advantage as a place for general purpose allocation is that it's "hot" memory, i.e. it's normally already in cache due to it's locality. The fact that it is 1) pre-allocated and 2) eagerly de-allocates on exiting a function can be easily implemented in any allocation scheme.

On Friday, 18 July 2014 at 06:46:16 UTC, Jacob Carlborg wrote:
> On 17/07/14 22:18, Dicebot wrote:
>
>> "Discouraged and unmaintained" is probably a better description.
>
> How does it need maintaining?

I mean that if you create bug report or enhancement request for scope classes it is quite unlikely anyone will pay attention to it.

On 07/17/2014 06:27 PM, Tero wrote:
> Just watched Don's DConf 2014 talk where he said D has to be ruthless about
> memory inefficiency. Here's one thing that I think could help avoid
> unnecessary garbage: built-in syntax for this:
>
> import core.stdc.stdlib : alloca;
> ubyte[] buffer = (cast(ubyte*) alloca(bufsize)) [0 .. bufsize];
>
> Often bufsize is not known at compile-time but it won't change after the
> buffer
> allocation. So there's no reason to create garbage other than the
> *inconvenience*
> of using alloca. Allocating in the stack seems ideal so I'd encourage
> that by a
> clean syntax. I keep missing this feature.

Well, there is always this hack:

import core.stdc.stdlib : alloca;
auto createBuffer(T,alias size)(T[] buf=(cast(T*)alloca(size))[0..size]){ return buf; }

void main(){
    auto bufsize=5;
    auto buffer=createBuffer!(ubyte,bufsize);
    import std.stdio;
    writeln(buffer);
}