[cfe-dev] -Wunused-* where macros declare multiple things

[David Blaikie]

On Mon, Nov 5, 2012 at 4:54 PM, Matt Beaumont-Gay <matthewbg at google.com> wrote:
> On Mon, Nov 5, 2012 at 4:46 PM, David Blaikie <dblaikie at gmail.com> wrote:
>> While evaluating the viability of Clang's -Wunused-* warnings
>> (specifically -Wunused-variable, -Wunused-function, and
>> -Wunused-member-function*) we've run in to some fairly noisy but
>> common idioms involving macros.
>>
>> Most obviously about 30% of -Wunused-member-function warnings came
>> from uses of GMock. When defining a mocking member, GMock's macros
>> define two member functions (one being, commonly, the function
>> overriding the function to mock) and a member variable. In a
>> reasonable number of cases the second member function (the one used to
>> set expectations) is not called.
>>
>> While not being deeply familiar with gmock, I believe this is a
>> reasonably valid use case & would cause the test to fail should that
>> member function ever be called (essentially an implicit expectation of
>> "never call").
>>
>> (I've seen a smaller number of cases of -Wunused-function/variable due
>> to non-gmock but similar macros that define multiple things some of
>> which are used/important)
>>
>> In the interests of making this warning viable for use in such a
>> codebase, I'm considering a few of options:
>>
>> 1) ignore any -Wunused-* when the thing being ignored is written inside a macro
>
> +1. I was going to write just such a patch for -Wunused-value; macros
> are the only significant source of false positives I've seen for this
> warning.

Agreed (fwiw I got a buildbreak in gdb because of this - a legit macro
that was just trying to evaluate to 'true', essentially: #define
FOO(bar) (fun(bar), true) )

Though I don't think the solution to unused file scoped decls will be
immediately applicable to -Wunused-value.

Attached is a trivial, possibly hackish, solution to unused file
scoped decls. If anyone feels like there's an important collection of
true positives this suppresses or has other concerns about such a
simplistic approach, I'd love to hear about it.

Thanks,
- David
-------------- next part --------------
A non-text attachment was scrubbed...
Name: unused.diff
Type: application/octet-stream
Size: 903 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20121113/91b7fee4/attachment.obj>