[llvm-commits] [llvm] r157298 - /llvm/trunk/lib/Transforms/Scalar/BoundsChecking.cpp

[Nuno Lopes]

Quoting Nick Lewycky <nicholas at mxc.ca>:

> Duncan Sands wrote:
>> Hi Nuno,
>>
>>> teach computeAllocSize about realloc, reallocf, and valloc
>>
>> ...
>>
>>> +      if ((FTy->getParamType(0)->isIntegerTy(32) ||
>>> +           FTy->getParamType(0)->isIntegerTy(64))&&
>>> +          (Callee->getName() == "malloc" ||
>>> +           Callee->getName() == "valloc" ||
>>> +           Callee->getName() == "_Znwj"  || // operator new(unsigned int)
>>> +           Callee->getName() == "_Znwm"  || // operator new(unsigned long)
>>> +           Callee->getName() == "_Znaj"  || // operator  
>>> new[](unsigned int)
>>> +           Callee->getName() == "_Znam")) { // operator  
>>> new[](unsigned long)
>>
>> shouldn't you query something like TargetLibraryInfo to check that  
>> these have
>> standard semantics.  Otherwise "malloc" might just be some  
>> arbitrary function
>> that has no relationship with C's standard malloc function.  See  
>> -fno-builtin.
>
> For that matter, this duplicates the logic in MemoryBuiltin.cpp's
> isMallocCall. Perhaps you should extend that.

Well, that's true, but that doesn't make the code correct :)
I agree we should be more careful here.

Nuno