Johan Engelen

> > > I think not. initCount is shared, but not __gshared, i.e. it is a variable per-thread. > > > > > > `shared` means shared between threads. `__gshared`...

OK, but probably the runtime code is not meant to be restarted like that anyway, i.e. more bugs besides the monotime init. Is this a solution? ```d pragma(crt_constructor) extern(C) void...

And `rt_init/rt_term` should be modified such that calling `rt_init` asserts an error when it is called after `rt_term` has really terminated the runtime (`_initCount == 0`).

Is there a good way to disable all `gdb`-related tests? (the gdb version on Alpine is too new and there are no alpine packages for older gdb versions)

The tester on Alpine Linux with musl is green now :))

Green again!

Any objections before merging? With this in place, in follow up I intend to work towards statically linked build and uploading a release to GH. I'll upstream the druntime fix....

@kinke Is it worthwhile letting the CI for cross-compiling druntime/phobos (and ldc-build-runtime probably too?) know about LDC's folder with cmake modules..... Or should I just move the `CHECK_MUSL` macro inside...

> If the compiler implicitly added `-lunwind` as 'platform-lib' for musl targets, the druntime integration tests would require less adaptations (then only required when linking via C(++) compiler). LDC already...

I'll check whether it is needed with newer compilers (weka is currently still at 1.30)