Re: [Qemu-devel] qapi-commands.py generates code that uses uninitialized variables

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1326 bytes --]

On 04/11/2014 01:27 AM, Peter Maydell wrote:
> On 11 April 2014 02:40, Eric Blake <eblake@redhat.com> wrote:
>> We uncovered a real bug that would be fixed by this patch:
>> https://lists.gnu.org/archive/html/qemu-devel/2014-04/msg01745.html
> 
> No, that's a bug in the called code. The API here defines
> that for optional parameters, if the have_foo bool is false
> then the foo argument isn't set. The generated code
> can't know the correct default value (it just happens
> to be 0 in the case you point out, but what if the default
> speed were 100?) so this must be handled by the called
> code.

The called code ALSO needs a fix, but guaranteeing that
'have_foo==false' implies 'foo==0' is MUCH nicer than 'have_foo==false'
implies 'foo is indeterminate'.  For this particular caller, an
indeterminate foo had detrimental effects, and a known foo==0 happened
to be the right default.  I agree that we can't always predict the right
default for all callers, but avoiding random behavior can be considered
a bug fix in its own right, and if we make it part of the contract that
callers can rely on zero initialization, we could simplify a lot of
callers that ARE happy with a 0 default.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]