Måns Rullgård <[email protected]> writes:

> Diego Biurrun <[email protected]> writes:
>
>> On Tue, Mar 22, 2011 at 10:05:42PM -0400, Ronald S. Bultje wrote:
>>> From: Carl Eugen Hoyos <[email protected]>
>>> 
>>> __ICC is not supported on all platforms, this is an Intel documentation
>>> bug that is supposed to be fixed in a future release.
>>> (Intel issue 612289)
>>
>> OK
>
> No, it's not  OK.

I'll elaborate a bit on those tests.  The ifdefs are there to enable
various attributes for recent enough gcc versions.  They test the
version macros rather than a configure test since they are needed in
public headers.  So far all is well.

Enter icc.  Some misguided individual at Intel thought it a good idea to
make icc report itself as whatever gcc version happens to be on the
system, whether or not it actually supports all its extensions.  The
result is that with only a gcc version check, the compiler output is
flooded with warnings about unknown attributes.  Thankfully these are
only warnings, not errors, so they are merely an annoyance unless the
effect of the attribute is relied on for correct operation (not our case).

The good news in all this is that a simple compiler flag will disable
these warnings all at once with no added clutter to the code.  I even
added it to our builds once, only to have CE scream bloody murder and it
was removed again.

I suggest we once again shut that warning up and remove all the useless
icc version checks from the code.  This might give some warnings for
people using the public headers, but that's their own fault for using
icc in the first place.  If they don't like it, they can complain to
Intel.

-- 
Måns Rullgård
[email protected]
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to