Re: [liblo] getting reading for 0.28-rc

On Mon, Nov 25, 2013 at 10:09 AM, Stephen Sinclair <rad...@gm...> wrote:
> On Mon, Nov 25, 2013 at 2:01 PM, Felipe Sateler <fsa...@gm...> wrote:
>> I hit one snag: the configure script munges the CFLAGS to remove
>> -Werror. However, -Werror can take a parameter (eg,
>> -Werror=format-security), meaning you end up with an invalid line like
>> `gcc  =format-security conftest.c`.
>>
>> I'm not sure configure should munge the CFLAGS, nor why would it be
>> necessary (do some configure tests contain invalid code?). Removing
>> the munging I can build and make test ends successfully
>
>
> Okay, thanks.  Believe me I hated adding code to mess with the flags,
> but in fact if you pass -Werror, the configure script fails since some
> tests, in particular AC_CHECK_FUNC, do indeed produce warnings.

Ah, yes, because the function is possibly undeclared.

>
> I looked up how to deal with this and the recommendation seemed to
> generally be to ensure -Werror was removed from CFLAGS before any
> tests.  I would think that adding it back is a mistake rather than
> just recovering the old CFLAGS variable, but after some tests, new
> flags may be added if I understand correctly.  (Though now that I
> think about it, maybe only LIBS would be modified by AC_CHECK_FUNC)
>
> I admit I didn't realize Werror could take arguments.  I'll see if I
> this can be dealt with better, thanks for finding the problem.

Other options to consider:

1. Emptying up CFLAGS before configure and re set it afterwards. The
disadvantage is that someone may have put needed -I or other flags
here (although they should be in CPPFLAGS).
2. Adding -Wno-error=<whatever>. This is probably not very
future-proof, as new warnings get added.
3. Don't touch CFLAGS, leave the developer to wonder what happened
when the tests break after adding -Werror.
4. Print a warning when -Werror is detected, and then error out explicitly.


-- 

Saludos,
Felipe Sateler