Thread safe RC Array and Ptr. Plus C++ headers for code integration.

https://github.com/libmir/mir-algorithm
Docs:
http://mir-algorithm.libmir.org
http://mir-algorithm.libmir.org/mir_rc_array.html
http://mir-algorithm.libmir.org/mir_rc_ptr.html

Also, Ndslice and Series RC wrappers are available:
http://mir-algorithm.libmir.org/mir_ndslice_allocation.html#.rcslice
http://mir-algorithm.libmir.org/mir_ndslice_allocation.html#.mininitRcslice
http://mir-algorithm.libmir.org/mir_ndslice_allocation.html#.bitRcslice
http://mir-algorithm.libmir.org/mir_series.html#.rcseries
http://mir-algorithm.libmir.org/mir_series.html#.rcTroykaSeries
http://mir-algorithm.libmir.org/mir_series.html#.rcUnionSeries

mir.qualifier from mir-core library contains useful utilities to work with RC and/or const data.

1. lightScope - returns a rc-free view of a Slice/Series. For example, for Slice!(RCI!double) it returns Slice!(double*)

2. lightConst - returns a mutable Slice/Series view to constant data. In terms of Phobos: it converts a constant range structure to a mutable range structure with constant elements.

3. lightImmutable  - the same for immutable qualifier

DMD related regression:
https://issues.dlang.org/show_bug.cgi?id=19774

and a new bug:
https://issues.dlang.org/show_bug.cgi?id=19793

This work has been sponsored by Symmetry Investments and Kaleidic Associates.

http://symmetryinvestments.com/
https://github.com/kaleidicassociates/

Best,
Ilya

On Wednesday, 24 April 2019 at 01:34:58 UTC, 9il wrote:
>
> Thread safe RC Array and Ptr. Plus C++ headers for code integration.
>
> [snip]

Cool.

Does this make any use of DIP1000? How is the run-time/memory performance vs. the GC versions?

On Wednesday, 24 April 2019 at 10:52:14 UTC, jmh530 wrote:
> On Wednesday, 24 April 2019 at 01:34:58 UTC, 9il wrote:
>>
>> Thread safe RC Array and Ptr. Plus C++ headers for code integration.
>>
>> [snip]
>
> Cool.
>
> Does this make any use of DIP1000? How is the run-time/memory performance vs. the GC versions?

Can't answer for mir, but automem uses DIP1000:

https://github.com/atilaneves/automem

I haven't profiled it against "the GC version", but it would highly depend on the allocators used and the application.

On Wednesday, 24 April 2019 at 10:52:14 UTC, jmh530 wrote:
> On Wednesday, 24 April 2019 at 01:34:58 UTC, 9il wrote:
>>
>> Thread safe RC Array and Ptr. Plus C++ headers for code integration.
>>
>> [snip]
>
> Cool.
>
> Does this make any use of DIP1000? How is the run-time/memory performance vs. the GC versions?

RC types are created to be used with DIP1000. Plus, Mir Algorithm used in production with this DIP. See configuration "dips" [1]

Well, the allocator support is not ready yet. But the mir_rc_context already contains `void* allocator` that will be replaced in the future with nothrow @nogc allocator interface (maybe this or next year). For now, Mir RC types are like a C++ shared_ptr without allocators and with the same performance.

1. https://github.com/libmir/mir-algorithm/blob/master/dub.sdl#L26

On Wednesday, 24 April 2019 at 14:05:28 UTC, 9il wrote:
> [snip]
>
> RC types are created to be used with DIP1000. Plus, Mir Algorithm used in production with this DIP. See configuration "dips" [1]
>
> Well, the allocator support is not ready yet. But the mir_rc_context already contains `void* allocator` that will be replaced in the future with nothrow @nogc allocator interface (maybe this or next year). For now, Mir RC types are like a C++ shared_ptr without allocators and with the same performance.
>
> 1. https://github.com/libmir/mir-algorithm/blob/master/dub.sdl#L26

Thanks.