Re: [Qemu-devel] [PATCH 3/3] qtest: kill QEMU process on g_assert() failure

[Paolo Bonzini <pbonzini@redhat.com>]

Il 17/02/2014 17:49, Markus Armbruster ha scritto:
> Assumes zero-initialization has the same effect as
> sigemptyset(&sigact.sa_mask).  Quoting POSIX:
>
>     The implementation of the sigemptyset() (or sigfillset()) function
>     could quite trivially clear (or set) all the bits in the signal set.
>     Alternatively, it would be reasonable to initialize part of the
>     structure, such as a version field, to permit binary-compatibility
>     between releases where the size of the set varies.  For such
>     reasons, either sigemptyset() or sigfillset() must be called prior
>     to any other use of the signal set, even if such use is read-only
>     (for example, as an argument to sigpending()).
>
> Looks like you better sigemptyset() here, for maximum portability.
>

Certainly memset of struct sigaction or sigset_t * is common enough that 
no one in their right minds would do this.  Is there really an OS that 
does it?  Also, the above justification is quite feeble; it would work 
for binary compatibility of sigset_t* arguments, but not for embedded 
sigset_t structs.  I'm CCing our resident POSIX experts in hope that 
this paragraph can be eliminated from the standard. :)

Related to this, there are a bunch of Coverity reports where we use 
uninitialized fields of a struct sigaction.

Paolo