ldc-developers/ldc v0.16.0-alpha4

LDC 0.16.0 alpha4

on GitHub

(Changes from 0.16.0-alpha3 release are marked with (new) and gone.)

Big news:

• Frontend, druntime and Phobos are at version 2.067.1.
• LLVM 3.6.x is now supported. (#821, #855)
• LLVM 3.7.x is now supported.
• Coverage Analysis à la DMD added. Implemented by Johan Engelen! (#878)
• std.net.curl now loads libcurl at runtime if/when it is actually used. This is the same fix present in post-2.068 DMD, and avoids linking issues.

Platform support:

• Win64 runtime now supports VS2015.
• Varargs fix & ABI refactoring. (#768) Thanks to kinke for the great work!
• Provide version ARM_SoftFP when float-abi=softfp. (#857)

Bug fixes:

• Make GCC intrinsics usable with @nogc and const *. (#1090) (new) Thanks to Marco Leise for providing the patch!
• Treat atomic instructions as intrinsics. (#1089) (new)
• ldc2 does not respect -of on Windows. (#1084) (new)
• LLVM IR output: Do not repeat type in comments, print alloca sizes. (#1080) (new)
• Calling members of large (>64 kiB) classes through interface reference fails. (#1065) (new)
• Since 0.16.0, can't build any project with DUB whenever it has a dependency. (#1061)
• Interface dispatch is crashing. (#1058)
• merge-2.067: finally block executed twice. (#1054)
• Instruction does not dominate all uses. (#1042)
• Segfault while compiling hibernated. (#1024)
• 2.067 branch: Enormous compile time increase after dtor rework. (#1010)
• Do not emit bounds checking branches that are known to be taken (#1008)
• D variadic codegen bad for x86 32-bit and LLVM 3.6 (#1000)
• Building with -O2 -g leads to ICEs/crashes (inliner produces invalid debug info) (#998)
• Assigning struct initializer can create huge temoraries on stack. (#991)
• Segfault in cast_.d line 59 or 71 when casting class with struct containing union. (#989)
• MSVC build, lots and lots of warnings. (#988) (new)
• ICE with std.datetime.benchmark (#982)
• Now the right C++ mangler is used when targetting MSVC.
• Duplicate symbol when using -lib (works with dmd). (#978)
• Failed assertion because of codegening a destructor call on an unavailable var. (#953)
• Vector ops optimization: inline vector ops with -O2 (#938)
• Crash: "Error: size of type is not known". (#893)
• msgpack-d 0.9.3 unittesting fails. (#841)
• repeatable crash with vague error message. (#837)
• ARM float-abi=soft. (Phobos only - see #829)
• Wrong handling of -m32/-m64. (#824)
• [DUB][scid] test crashes. (#823)
• Can't link dub with i686 LDC 0.15.1. (#813)
• Segfault when casting a class with a 0 length array. (#812)
• Compiler silently dies while codegening. (#809)
• Vararg functions does not compile on x86_64. (#702)
• Make the x86_64 System V ABI code use Type::toArgTypes. (#288)
• ABI transformations are never applied for variadic arguments. (#172)
• stack trace missing on assert failure. (#115, fixed only on OS X, see #863)
• core.stdc.stdarg doesn't work on 64 bits. (#73)

Building LDC:

• CMake fails after git gc. (#874)
• Fix cmake bug when ldc is a submodule. (#860)
• druntime fails compilation with MinGW-x64. (#847)
• Now uses "REGEX REPLACE" for MSVC static libaries option.

Internals:

• The Travis-CI builds now use the docker environment.
• AppVeyor is used to check the Win64 build. Snapshots can be downloaded from https://ci.appveyor.com/project/kinke/ldc.
• Coverage Analysis for the LDC source code is now available at https://coveralls.io/r/ldc-developers/ldc. (#903, #380). Thanks a lot to Dmitri Makarov for implementing this!
• The CPP mangler calls now fatal() instead of assert(0) in many places. (#927, #928, #929)
• Updated to Doxygen v1.8.9.1 and changed path 'dmd' to 'dmd2'. (#846)
• Modernize bash completion setup. (#840)
• Updates to Travis-CI: gcc 4.9 and LLVM 3.7 are added. (#838)
• ldc2: valgrind: invalid read of size 4. (#825)

Known issues:

• LDC does not zero the padding area of a real variable. This may lead to wrong results if the padding area is also considered. See #770. Does not apply to real members inside structs etc.
• atomicLoad(), atomicStore() et al. with 128bit data requires an alignment of 16 bytes on x86-64. Add align(16) if you encounter a segfault. If you are using vibe.d then you need to use version ~master because of this.
• Phobos does not compile on MinGW platform.

Downloads