Libguestfs - May 2021 - [libnbd PATCH 3/5] libev: Check for old compiler

On Fri, May 07, 2021 at 07:14:32AM -0500, Eric Blake
wrote:
>On 5/6/21 6:30 AM, Martin Kletzander wrote:
>> At some point GCC added some optimisations and checks that started
emitting
>> warnings on harmless code:
>>
>>   ...dereferencing type-punned pointer will break strict-aliasing
rules...
>>
>> This was later fixed, but can still break the build (or report
warnings) on old
>> GCCs (e.g. 7.5.0) when building the libev example.
>>
>> Signed-off-by: Martin Kletzander <mkletzan at redhat.com>
>> ---
>>  configure.ac | 24 ++++++++++++++++++++++++
>>  1 file changed, 24 insertions(+)
>>
>> diff --git a/configure.ac b/configure.ac
>> index 19b7bfdb5d2a..3b51354346cc 100644
>> --- a/configure.ac
>> +++ b/configure.ac
>> @@ -247,6 +247,30 @@ PKG_CHECK_MODULES([LIBEV], [libev], [
>>          AC_MSG_WARN([ev.h not found, some examples will not be
compiled])
>>      ])
>>  ])
>> +
>> +AS_IF([test "x$LIBEV_LIBS" != "x"], [
>> +    old_CFLAGS="$CFLAGS"
>> +    CFLAGS="-Werror=strict-aliasing -O2"
>> +    AC_MSG_CHECKING([if the compiler is new enough for good aliasing
rules])
>> +    AC_COMPILE_IFELSE([
>> +        AC_LANG_PROGRAM([
>> +            #include <ev.h>
>> +
>> +            static void cb (struct ev_loop *l, ev_timer *t, int e) { }
>> +            static ev_timer timer;
>> +        ], [
>> +            ev_timer_init (&timer, cb, 0, .1);
>> +        ])
>> +    ], [
>> +        AC_MSG_RESULT([yes])
>> +    ], [
>> +        AC_MSG_RESULT([no])
>> +        AC_SUBST([LIBEV_CFLAGS], [""])
>> +        AC_SUBST([LIBEV_LIBS], [""])
>> +        AC_MSG_WARN([compiler is probably too old to compile with
libev without errors, some examples will not be compiled])
>
>Do we really need to skip compiling the example altogether, or can we
>just modify CFLAGS to add -Wno-strict-aliasing to prevent the failure?
>
This is one of the results that I was not sure about.  We could just add
no-strict-aliasing for that one particular example of course.  I figured
it is just an example and it only causes an issue on CentOS 7, OpenSUSE
Leap 15.2, Ubuntu 18.04 and some similarly old systems if I remember
correctly.  But changing the LIBEV_CFLAGS seems more fit for the purpose.
>--
>Eric Blake, Principal Software Engineer
>Red Hat, Inc.           +1-919-301-3226
>Virtualization:  qemu.org | libvirt.org
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL:
<http://listman.redhat.com/archives/libguestfs/attachments/20210507/3b85bf33/attachment.sig>

On 5/7/21 8:23 AM, Martin Kletzander wrote:
>>> +??? AC_MSG_CHECKING([if the compiler is new enough for good
aliasing
>>> rules])
>>> +??? AC_COMPILE_IFELSE([
>>> +??????? AC_LANG_PROGRAM([
>>> +??????????? #include <ev.h>
>>> +
>>> +??????????? static void cb (struct ev_loop *l, ev_timer *t, int e)
{ }
>>> +??????????? static ev_timer timer;
>>> +??????? ], [
>>> +??????????? ev_timer_init (&timer, cb, 0, .1);
>>> +??????? ])
>>> +??? ], [
>>> +??????? AC_MSG_RESULT([yes])
>>> +??? ], [
>>> +??????? AC_MSG_RESULT([no])
>>> +??????? AC_SUBST([LIBEV_CFLAGS], [""])
>>> +??????? AC_SUBST([LIBEV_LIBS], [""])
>>> +??????? AC_MSG_WARN([compiler is probably too old to compile with
>>> libev without errors, some examples will not be compiled])
>>
>> Do we really need to skip compiling the example altogether, or can we
>> just modify CFLAGS to add -Wno-strict-aliasing to prevent the failure?
>>
> 
> This is one of the results that I was not sure about.? We could just add
> no-strict-aliasing for that one particular example of course.? I figured
> it is just an example and it only causes an issue on CentOS 7, OpenSUSE
> Leap 15.2, Ubuntu 18.04 and some similarly old systems if I remember
> correctly.? But changing the LIBEV_CFLAGS seems more fit for the purpose.
Sounds even better!

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org

On Fri, May 7, 2021 at 4:24 PM Martin Kletzander <mkletzan at redhat.com>
wrote:
>
> On Fri, May 07, 2021 at 07:14:32AM -0500, Eric Blake wrote:
> >On 5/6/21 6:30 AM, Martin Kletzander wrote:
> >> At some point GCC added some optimisations and checks that started
emitting
> >> warnings on harmless code:
> >>
> >>   ...dereferencing type-punned pointer will break strict-aliasing
rules...
> >>
> >> This was later fixed, but can still break the build (or report
warnings) on old
> >> GCCs (e.g. 7.5.0) when building the libev example.
> >>
> >> Signed-off-by: Martin Kletzander <mkletzan at redhat.com>
> >> ---
> >>  configure.ac | 24 ++++++++++++++++++++++++
> >>  1 file changed, 24 insertions(+)
> >>
> >> diff --git a/configure.ac b/configure.ac
> >> index 19b7bfdb5d2a..3b51354346cc 100644
> >> --- a/configure.ac
> >> +++ b/configure.ac
> >> @@ -247,6 +247,30 @@ PKG_CHECK_MODULES([LIBEV], [libev], [
> >>          AC_MSG_WARN([ev.h not found, some examples will not be
compiled])
> >>      ])
> >>  ])
> >> +
> >> +AS_IF([test "x$LIBEV_LIBS" != "x"], [
> >> +    old_CFLAGS="$CFLAGS"
> >> +    CFLAGS="-Werror=strict-aliasing -O2"
> >> +    AC_MSG_CHECKING([if the compiler is new enough for good
aliasing rules])
> >> +    AC_COMPILE_IFELSE([
> >> +        AC_LANG_PROGRAM([
> >> +            #include <ev.h>
> >> +
> >> +            static void cb (struct ev_loop *l, ev_timer *t, int
e) { }
> >> +            static ev_timer timer;
> >> +        ], [
> >> +            ev_timer_init (&timer, cb, 0, .1);
> >> +        ])
> >> +    ], [
> >> +        AC_MSG_RESULT([yes])
> >> +    ], [
> >> +        AC_MSG_RESULT([no])
> >> +        AC_SUBST([LIBEV_CFLAGS], [""])
> >> +        AC_SUBST([LIBEV_LIBS], [""])
> >> +        AC_MSG_WARN([compiler is probably too old to compile with
libev without errors, some examples will not be compiled])
> >
> >Do we really need to skip compiling the example altogether, or can we
> >just modify CFLAGS to add -Wno-strict-aliasing to prevent the failure?
> >
>
> This is one of the results that I was not sure about.  We could just add
> no-strict-aliasing for that one particular example of course.  I figured
> it is just an example and it only causes an issue on CentOS 7, OpenSUSE
> Leap 15.2, Ubuntu 18.04 and some similarly old systems if I remember
> correctly.
Do we really support Centos 7? libnbd is available since rhel 8 (2, 3?).
Why support older systems?

Time spent on older systems means less time working features,
performance, support for new systems, etc.
> But changing the LIBEV_CFLAGS seems more fit for the purpose.
Sounds like the best option.

Nir