remove thread.di header

[MartinNowak]

• Use extern(D) bindings for the core.thread<->rt.tlsgc
  interface to avoid importing a rt module in core.

[ibuclaw]

Wow.

[ibuclaw]

My first thoughts are, why the extern(D) hack? Why not just add an extern(C) interface?

[MartinNowak]

My first thoughts are, why the extern(D) hack? Why not just add an extern(C) interface?

Because extern(D) allows "typesafe" interfaces, i.e. when declaration and definition are not in sync your program will not link.
Also some people like to restrict extern(C) to real C, abusing extern(C) as simple mangling for D function gets in the way of that (see Issue 12959, Issue 7854), not to mention the different calling convention.

[dnadlinger]

+1, both for eliminating the .di file and using extern(D). I was actually working on something a few hours back, but also ran into the bugs Martin reported and didn't have the patience to work around them.

I spent the last couple of days staring at the core.thread source code and if I learnt one thing, it's that reading the source right now isn't enjoyable at all. After this is in, I'll post a second refactoring pull request to fix the nondescript internal names (slock/m_lock/…), misleading prefixes and a couple of outdated comments.

[dnadlinger]

Any objections to merging this?

[ibuclaw]

Not from me.

[alexrp]

LGTM, great progress towards sanity.

We may want to use extern (D) for the GC functions too, later down the road. The fact that those are extern (C) has bitten me in the rear on multiple occasions where I was unwittingly overwriting them, causing all sorts of fun crashes.

[dnadlinger]

Auto-merge toggled on