[llvm] r176341 - GCC thinks that this variable might be used uninitialized (it isn't).

[Duncan Sands]

Hi Michael,

On 22/03/13 18:46, Michael Liao wrote:
> LGTM. But I am not the owner of cmake. Could cmake owners take a look of
> it?
> Thanks for fixing it.

it isn't enough when compiling clang: it looks like clang adds more flags
to the command line, including another instance of -Wall.  It does seem to
do the trick for LLVM itself though.

Ciao, Duncan.

>
> - Michael
>
> On Fri, 2013-03-22 at 18:32 +0100, Duncan Sands wrote:
>> Hi Michael,
>>
>> On 21/03/13 23:21, Michael Liao wrote:
>>> On Thu, 2013-03-21 at 23:17 +0100, Arnaud de Grandmaison wrote:
>>>> I understand your point, but do we really need this back-compatibility here
>>>> --- for the '-W flags' ?
>>>>
>>>> the add_(llvm_)definitions is for sure convenient because it sets the flags
>>>> for C and C++ at the same time, whereas the append_if only set one at a time.
>>>>
>>>> My other concern is that as far as I know, the respective ordering in the
>>>> compilation command line of definitions and flags depends on the cmake
>>>> toolchain file : we simply can not rely on that, as it could even change from
>>>> one revision to the other. Beside, we may want to have cross-compilation one
>>>> day with cmake. On the other hand, the append_if is quite clear in its intent
>>>> and does not have this issue.
>>>
>>> I don't mean that's the issue. The issue is LLVM specific. Its cmake
>>> prepares command line options through different ways. '-Wall' is added
>>> through 'add_definition', and '-Wno-maybe-uninitialized' is added
>>> through 'CMAKE_CXX_FLAGS'. We should follow a consistent way adding
>>> them.
>>
>> I think it's reasonable to use append_if for all warning flags.  Patch attached.
>>
>> Ciao, Duncan.
>
>