[Lldb-commits] [PATCH] [DWARFExpression] Convert a result of DWARF pieces into a scalar value.

[Greg Clayton]

> On Apr 30, 2015, at 4:05 PM, Siva Chandra <sivachandra at google.com> wrote:
> 
> On Thu, Apr 30, 2015 at 3:41 PM, Greg Clayton <clayborg at gmail.com> wrote:
> 
>> This is an incorrect fix. The variable doesn't necessarily need to be a scalar.
> 
>> You might be describing where a uint128 or larger value is constructed from
> 
>> multiple registers. In your case this works, but this isn't the right fix. We
> 
>> should be able to describe a variable's value by saying "here is the buffer that
> 
>> contains it".
> 
> 
> This wording helps understand the code and its meaning better and I now agree that the change is in the wrong place.
> 
> I would like to use this as an opportunity to ask about few other pieces of code which don't seem to be in line [per my current understanding that is] with the above meaning.
> 
> Look at my comments inline with in code below from ValueObject.cpp:
> 
>  1804 addr_t
>  1805 ValueObject::GetAddressOf (bool scalar_is_load_address, AddressType *address_type)
>  1806 {
>  1807     if (!UpdateValueIfNeeded(false))
>  1808         return LLDB_INVALID_ADDRESS;
>  1809
>  1810     switch (m_value.GetValueType())
>  1811     {
>  1812     case Value::eValueTypeScalar:
>  1813     case Value::eValueTypeVector:
>  1814         if (scalar_is_load_address)
>  1815         {
>  1816             if(address_type)
>  1817                 *address_type = eAddressTypeLoad;
>  1818             return m_value.GetScalar().ULongLong(LLDB_INVALID_ADDRESS);
>  1819         }
>  1820         break;
>  1821
>  1822     case Value::eValueTypeLoadAddress:
>  1823     case Value::eValueTypeFileAddress:
>  1824     case Value::eValueTypeHostAddress:
> 
>      Can a value of type eValueTypeHostAddress have a valid address in the inferior process space? Even if it does, why is a scalar value being returned on line 1828.

File address means you have debug info that describes 0x1000, but it will get loaded elsewhere and it will need to be mapped to where the image was loaded (convert the file address to a valid load address, then read the variable).

Load address means a live address in the inferior, no translation needed.

Host address: it is in a malloc'ed buffer inside LLDB. This is used for freeze dried values (like all expression results). 

So you are correct in that sometimes we have both a host address for the freeze dried bits, and we might have this mirrored down in the process. But the value for that will usually be handed out as a load address.

If a value has a host address, we should assume it is host only.


> 
>  1825         {
>  1826             if(address_type)
>  1827                 *address_type = m_value.GetValueAddressType ();
>  1828             return m_value.GetScalar().ULongLong(LLDB_INVALID_ADDRESS);
>  1829         }
>  1830         break;
>  1831     }
>  1832     if (address_type)
>  1833         *address_type = eAddressTypeInvalid;
>  1834     return LLDB_INVALID_ADDRESS;
>  1835 }
> 
> Another one from ValueObject.cpp:
> 
>  3802         case eAddressTypeHost:
>  3803             {
>  3804                 ClangASTType clang_type = GetClangType();
>  3805                 if (clang_type)
>  3806                 {
>  3807                     std::string name (1, '&');
>  3808                     name.append (m_name.AsCString(""));
>  3809                     ExecutionContext exe_ctx (GetExecutionContextRef());
>  3810                     m_addr_of_valobj_sp = ValueObjectConstResult::Create (exe_ctx.GetBestExecutionContextScope(),
>  3811                                                                           clang_type.GetPointerType(),
>  3812                                                                           ConstString (name.c_str()),
>  3813                                                                           addr,
> 
>      Why is the host address being treated as the value's address when the address type is eAddressTypeHost.

We should not allow getting the address of something that is a host address. We currently can though because the pointer itself is in the Value's Scalar which means we would take the address of the integer containing the pointer if we wanted to. But I don't think it is a good idea.

> 
>  3814                                                                           eAddressTypeInvalid,
>  3815                                                                           m_data.GetAddressByteSize());
>  3816                 }
> 
> 
> http://reviews.llvm.org/D9421
> 
> EMAIL PREFERENCES
>  http://reviews.llvm.org/settings/panel/emailpreferences/
> 
>