[libvirt] [PATCH 0/8] Fix build with clang

[Ján Tomko]

On Wed, Jun 29, 2016 at 07:37:03PM +0200, Andrea Bolognani wrote:
>On Wed, 2016-06-29 at 15:53 +0200, Ján Tomko wrote:
>> I am not proud of the first patch.
>> 
>> I wrote the temporary bump for -Wframe-larger-than thinking
>> splitting it out would be too intrusive for the freeze, but
>> it did not turn out so bad.
>> 
>> Patch 7/8 does not fix anything.
>> 
>> Ján Tomko (8):
>>   virFirewallAddRule: exchange first two parameters
>>   virTypedParamsValidate: do not use STREQ_NULLABLE
>>   build: disable -Wdouble-promotion
>>   build: ignore some clang-specific warnings
>>   Introudce VIR_WARNINGS_NO_UNUSED_VALUE
>>   Temporarily bump maximum stack size
>>   test/Makefile.am: drop WARN_CFLAGS from LDFLAGS
>>   Split out -Wframe-larger-than warning from WARN_CLFAGS
>> 
>>  daemon/Makefile.am                        |  3 +
>>  m4/virt-compile-warnings.m4               | 15 ++++-
>>  src/Makefile.am                           |  1 +
>>  src/internal.h                            |  5 ++
>>  src/nwfilter/nwfilter_ebiptables_driver.c | 84 +++++++++++++--------------
>>  src/util/viratomic.h                      |  7 +++
>>  src/util/virebtables.c                    |  8 +--
>>  src/util/virfirewall.c                    |  8 +--
>>  src/util/virfirewall.h                    |  4 +-
>>  src/util/viriptables.c                    | 38 ++++++-------
>>  src/util/virtypedparam.c                  |  2 +-
>>  tests/Makefile.am                         |  3 +-
>>  tests/virfirewalltest.c                   | 94 +++++++++++++++----------------
>>  13 files changed, 150 insertions(+), 122 deletions(-)
>
>I have
>
>  clang 3.8.0 on Fedora rawhide
>  clang 3.8.1 on Debian sid
>
>and libvirt already builds fine in both cases.
>

The warnings were on a Gentoo system wtih sys-devel/clang-3.8.0-r100.

After rebuilding without the 'python' USE flag, it behaves the same as
3.8.0 in F24 for me (and I wish I knew why):
USE="static-analyzer -debug -multitarget -python"

At least the VIR_TYPEMATCH error seems to be present on Erik's F23 with
clang 3.7.1.

>The only failures I have are with virnetdevtest and
>qemuxml2argvtest, and both are when *running* the test cases,
>not when building them. Applying your series doesn't change
>that.
>

Yes, with "-g -O2" we use by default.

Without -O2, I get frame size errors in the test suite.

Jan