Hi all,
  I recently found this series of articles by one of the authors of the gold linker.

The series begins here: https://www.airs.com/blog/archives/38
Just ++ the number at the end to get to the next article in the series.

One particular interesting one is part 7 (44), where he explains an optimization performed by the linker around `call ___tls_get_addr`.

cheers,
  Johan

On Sunday, 12 July 2020 at 09:25:18 UTC, Johan wrote:
> Hi all,
>   I recently found this series of articles by one of the authors of the gold linker.
>
> The series begins here: https://www.airs.com/blog/archives/38
> Just ++ the number at the end to get to the next article in the series.
>
> One particular interesting one is part 7 (44), where he explains an optimization performed by the linker around `call ___tls_get_addr`.
>
> cheers,
>   Johan

It's a good read for D developers so that they understand what is going on underneath the TLS variables. My personal opinion about TLS is that it is a zoo. A lot of complexity for rather little extra functionality. Not only does TLS variables take more time to access, thread creation itself is more complex. The ELF TLS specification also doesn't cover language specific initialization just copy and zero initialization, then for D you have to call a function in order to initialize the TLS objects (double initialization).

My advise is, do not use TLS variables at all as far as you can. I wish we could rewrite history with this one.