Hi all,

I think it is long overdue that we decide how to go on about doing the merge of 2.064. At this point, it is probably more "stable" (in the sense of bug-free) than master, simply because of all the fixed frontend bugs.

What about just merging it to master, releasing a beta, and looking whether any users actually complain about having to unnecessarily link libcurl on their machines? Even if the heisenbug strikes at your machine, the libcurl dependency isn't really a show-stopper in many cases. (I still didn't manage to reproduce the issue, by the way.)

The only thing we'd really need to fix is that the LLVM 3.4 release test suite build (which is run with -DMULTILIB="ON") currently exceeds the Travis time limit.

If we want a surefire fix for the libcurl issues, I suppose we could properly implement -defaultlib/-debuglib (such that they are overriding, and allow specifying multiple libraries to link to) and just split off the stuff depending on libcurl into a different library (and then link -lphobos2-ldc -ldruntime-ldc -lphobos2-curl-ldc -lphobos2-ldc -ldruntime-ldc or use section groups). We need to at least do the first part anyway as soon as we want to properly support using a shared druntime/Phobos.

I think at this point it is be pretty clear that any other issues are secondary to master being on a horribly outdated frontend version.

Best,
David