[cfe-commits] r84198 - in /cfe/trunk: include/clang-c/Index.h tools/CIndex/CIndex.cpp tools/CIndex/CIndex.exports

[Benjamin Kramer]

Am 18.10.2009 um 04:09 schrieb Daniel Dunbar:

> Hi Steve,
>
> I just noticed another problem here...
>
> On Thu, Oct 15, 2009 at 1:04 PM, Steve Naroff <snaroff at apple.com>  
> wrote:
>> Author: snaroff
>> Date: Thu Oct 15 15:04:39 2009
>> New Revision: 84198
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=84198&view=rev
>> Log:
>> Implement <rdar://problem/7303432> [Clang/Index] In-memory-style  
>> AST generation API (initial API implementation).
>>
>> Added clang_createTranslationUnitFromSourceFile().
>> Changed clang_createIndex() to lookup the location of clang (using  
>> dladdr).
>>
>> Modified:
>>    cfe/trunk/include/clang-c/Index.h
>>    cfe/trunk/tools/CIndex/CIndex.cpp
>>    cfe/trunk/tools/CIndex/CIndex.exports
>>
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> =====================================================================
>> --- cfe/trunk/tools/CIndex/CIndex.cpp (original)
>> +++ cfe/trunk/tools/CIndex/CIndex.cpp Thu Oct 15 15:04:39 2009
> ...
>> +static const char *clangPath;
>> +
>>  CXIndex clang_createIndex()
>>  {
>>   // FIXME: Program is leaked.
>> +
>> +  // Find the location where this library lives (libCIndex.dylib).
>> +  // We do the lookup here to avoid poking dladdr too many times.
>> +  // This silly cast below avoids a C++ warning.
>> +  Dl_info info;
>> +  if (dladdr((void *)(uintptr_t)clang_createTranslationUnit,  
>> &info) == 0)
>> +    assert(0 && "Call to dladdr() failed");
>> +
>> +  llvm::sys::Path CIndexPath(info.dli_fname);
>> +  std::string CIndexDir = CIndexPath.getDirname();
>> +
>> +  // We now have the CIndex directory, locate clang relative to it.
>> +  std::string ClangPath = CIndexDir + "/../bin/clang";
>> +
>> +  clangPath = ClangPath.c_str();
>
> This isn't safe, its storing a dangling pointer into clangPath. Also,
> it would be good not to add a global variable.

I moved the variable to CIndexer. I also noticed two other dangling  
pointers
that need fixing:

Line 455: return OMD->getSelector().getAsString().c_str();
Line 520: return OME->getSelector().getAsString().c_str();

> I think it would be better to have this live in CXIndex. While at it,
> it makes sense to lazily compute it, for clients that don't use
> clang_createTranslationUnitFromSourceFile.

This still needs to be done.