LDC 0.11.0: Release branch created, Windows/OS X issues

[David Nadlinger]

Hi all,

The most important piece of information first: I just created the 
branches for the upcoming 0.11.0 release. This means that furious 
hacking activity on the master branch can and should resume!

Even if we should be good to go on Linux, I didn't put together beta 
packages just yet, though. The reason for this is that we are somewhat 
stuck on both OS X and Windows/MinGW.

On OS X, we are hitting quite an ugly EH bug in LLVM 3.2 ([1], breaks 
std.file/std.parallelism optimized unit tests), which seems to be fixed 
in the upcoming 3.3 release. There are two open issue with regard to 
this, though:
  - The current branch LLVM 3.3 contains a regression [2] that breaks 
compilation of the D runtime libraries. While it is officially listed as 
a release blocker, there is no indication of the amount of work it is 
currently receiving, and we pretty much have to wait for the fix (or 
find a workaround) until we can properly start testing 3.3 on OS X.
  - The LLVM release is scheduled to take place on June 5th. Do we want 
to delay our own release until then, or should we put out the binaries 
based on the LLVM release_33 branch? I don't see a big problem with the 
latter approach, especially considering that the LLVM schedule has shown 
to be rather … flexible in the past.

Additionally, there is a frontend issue breaking initialization of 
structs containing 'real' fields in certain cases [3]. As the issue 
occurs only in 32 bit mode and OS X has pretty much transitioned to 64 
bit completely (especially because we require 10.7 anyway), I'm inclined 
to just leave this open as a known bug, waiting for the upstream fix. 
What do you think?

Now, for Windows/MinGW: I want to include a "preview" build with the 
release, and it would only be natural to use LLVM 3.3 there as well, as 
it contains my TLS-related patches. However, the current version suffers 
from another regression [4] that breaks compilation of all but the most 
trivial test cases when debug info is turned on.

To sum up, the main issue at this point are not so much the regressions 
itself, but how big a delay in our release cycle we want to accept, 
because we do not know how many problems with LLVM 3.3 there are still 
hiding behind the most apparent bugs (even if it is looking solid on 
Linux).

Thanks for your opinions!
  — David


[1] https://github.com/ldc-developers/ldc/issues/362
[2] http://llvm.org/bugs/show_bug.cgi?id=15972
[3] https://github.com/ldc-developers/ldc/issues/363
[4] http://llvm.org/bugs/show_bug.cgi?id=15408