CMake: Add option to build with WAVM

[chfast]

This adds -DHERA_WAVM option what for now will build WAVM and link hera with it. Because Hera does not uses WAVM it will not do anything useful, but we can check if the build works.

Also the HERA_WAVM=1 macro is predefined for build with WAVM.

This depends on LLVM 6.0 so you have to install libllvm-6.0-dev. See http://apt.llvm.org.

The WAVM produces a number of libraries. In this PR only the libruntime is used, probably more is going to be needed when we start using WAVM.

Part of #158.

[chfast]

I think I should test the build with WAVM on any CI servers.

[jakelang]

I think there is an existing option in the abstraction layer PR called -DWAVM_SUPPORTED. I will adjust to use this one instead.

[jakelang]

In addition to the Runtime lib (and the dependencies pulled in implicitly) we also need the WASM library to convert the bytecode into WAVM's internal IR.

[chfast]

Ready to merge?

[jakelang]

If the WAVM include root is added to Hera's include search list then yeah, I think it is ready.

[axic]

Added a sample code to pull in all the required bits for loading & running (in dc50760) and it seems to compile.

[axic]

My bad, used the wrong ifdef, however now it does complain:

In file included from /home/builder/project/src/hera.cpp:41:
In file included from deps/src/wavm/Include/IR/Module.h:3:
In file included from deps/src/wavm/Include/Inline/Assert.h:3:
deps/src/wavm/Include/Platform/Platform.h:89:2: error: unknown type name 'DLL_IMPORT'
        PLATFORM_API Uptr getPageSizeLog2();
        ^
deps/src/wavm/Include/Platform/Platform.h:31:23: note: expanded from macro 'PLATFORM_API'
        #define PLATFORM_API DLL_IMPORT
                             ^
deps/src/wavm/Include/Platform/Platform.h:89:19: error: expected ';' after top level declarator
        PLATFORM_API Uptr getPageSizeLog2();
                         ^

Resolution: manually added #define DLL_IMPORT before the includes, but not sure if there's a cmake solution for this.

[axic]

@chfast So it seems the sample code compiles and links, but there is some kind of runtime linking error:

testeth 1.4.0.dev0
terminate called after throwing an instance of 'boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::system::system_error> >'
  what():  boost::dll::shared_library::load() failed (dlerror system message: /home/builder/build/src/libhera.so: undefined symbol: _ZN4llvm17JITSymbolResolver6anchorEv): Bad file descriptor
Received 'aborted' signal

which resolves to llvm::JITSymbolResolver::anchor()

[axic]

Lastly it fails to build if not built as shared library.

[axic]

This seems to be working on dynamic loading now (tests fail because the code will throw intentionally).

Static linking however doesn't work because the flag -Werror=maybe-uninitialized is passed down from aleth.

[chfast]

It should build with GCC 8 now.

[axic]

The new sed doesn't work on mac:

[ 10%] Performing patch step for 'wavm'
sed: 1: "CMakeLists.txt": invalid command code C

[axic]

It still fails on mac:

[ 10%] Performing patch step for 'wavm'
sed: 1: "CMakeLists.txt": invalid command code C
sed: 1: "CMakeLists.txt": invalid command code C

[axic]

Moving them back to -iE fixes it.

[chfast]

sed is still failing now?

[axic]

Yes, changing your last commit to sed -iE makes everything to work.

[chfast]

But I removed regex from sed.

[axic]

sed on Mac is kind of weird

[axic]

On Mac, though -iE workarounds that by somehow making it to ignore the extension parameter:

     -i extension
             Edit files in-place, saving backups with the specified extension.  If a zero-length extension is given, no backup will be saved.  It is not recommended to give a zero-
             length extension when in-place editing files, as you risk corruption or partial content in situations where disk space is exhausted, etc.

[chfast]

I will replace it with CMake script not to have problems with it on Windows.

[axic]

I will replace it with CMake script not to have problems with it on Windows.

I think I will merge this broken sed -iE version so we can get going and you can fix it in a new PR.