Hi,

This week I ran a benchmark for the 3 possible approaches to _d_arrayctor. They are: the old hook on one hand and two implementations of the new template. One of these uses a third pointer parameter for _d_arractor, in order to convert its strong purity to weak purity, while the other only takes the source array as a parameter and returns the created array. The latter doesn't use NRVO, thought. I gave more details about these 3 approaches and abount my benchmark in [this post].

After finishing the benchmark, I resumed work on _d_arrayappendcTX. I'm currently debugging an error where the template is called 3 times instead of one and I'm trying to figure out what code in the compiler is re-run so that this ends up happening.

If I get stuck with _d_arrayappendcTX, I'll start looking at _d_newitem{T,iT,U}. These hooks are likely to be simpler to implement because they only call the GC.

Thanks,
Teodor

On Friday, 26 November 2021 at 23:16:49 UTC, Teodor Dutu wrote:

Forgot to add the link: https://forum.dlang.org/post/hajlsppmugslhinluzos@forum.dlang.org

On Friday, 26 November 2021 at 23:16:49 UTC, Teodor Dutu wrote:

👍