On Tuesday, December 2, 2014 1:37:48 AM UTC+10, Max Suster wrote:
>
>
> *Keno *- I got a little ahead of myself with my last post. While
> exception handling works fine in several examples, after adding the lines
> to bootstrap.cpp, I now get a nasty error from Clang when using OpenCV
> functions that require casting with RTTI (i.e., dyne_cast) . Such OpenCV
> functions worked before adding the exception handling flags to
> bootstrap.cpp.
>
> e.g.,
> julia> img = imread(filename) # => here it must convert a const char*
> to const String&
> Assertion failed: ((Flags & RF_IgnoreMissingEntries) && "Referenced value
> not in value map!"), function RemapInstruction, file
> /Users/maximilianosuster/julia-v0.4.0/deps/llvm-svn/lib/Transforms/Utils/ValueMapper.cpp,
>
> line 194.
>
> The error can be eliminated by commenting out the following line (and not
> the others):
> clang_compiler->getLangOpts().Exceptions = 1; // exception handling
>
> This is the function that the error refers to in
> Transforms/Utils/ValueMapper.cpp
>
> void llvm::RemapInstruction(Instruction *I, ValueToValueMapTy &VMap,
> RemapFlags Flags) { // Remap operands. for
> (User::op_iterator op = I->op_begin(), E = I->op_end(); op != E; ++op) {
> Value *V = MapValue(*op, VMap, Flags); // If we aren't ignoring missing
> entries, assert that something happened. if (V != 0) *op = V; else
> assert((Flags & RF_IgnoreMissingEntries) && "Referenced
> value not in value map!"); }
>
>
> I tried to understand what is going on, but AFAIK Clang has its own
> built-in RTTI and its not entirely clear to me why switching on exceptions
> in bootstrap.cpp compromises the casting of const char* to String&
> reference. I tried enabling RTTI =1 and RTTIData = 1 in bootstrap.cpp and
> removed the -fno-rtti flag from the BuildBootsrap.Makefile. This
> generated a fatal ERROR upon Pkg.build("Cxx").
>
> Any thoughts on what is going on?
>
>
Without knowing the context where the map is created and modified, is it
possible that a previous caught exception has left the map in a
questionable state or failed to add an entry when it terminated some
function that wasn't expecting exceptions?
> *Lex *- thanks for the tip on the use of exceptions. The immediate aim
> is to simply avoid crashing the julia REPL every time Clang does not like
> something in the C++ function arguments. I would prefer eventually to
> transfer most error checking/exception handling to Julia, but there are
> cases where some algorithms are best executed in C++.
>
Sure, using C++ is necessary for now, just don't let the exception escape
the C++.
