Blue Swirl <blauwir...@gmail.com> writes:
> On 3/18/10, Markus Armbruster <arm...@redhat.com> wrote:
>> Blue Swirl <blauwir...@gmail.com> writes:
>>
>> > On 3/18/10, Markus Armbruster <arm...@redhat.com> wrote:
>> >> Blue Swirl <blauwir...@gmail.com> writes:
>> >>
>> >> > On 3/17/10, Markus Armbruster <arm...@redhat.com> wrote:
>> >> >> Blue Swirl <blauwir...@gmail.com> writes:
>> >> >>
>> >> >> > On 3/17/10, Markus Armbruster <arm...@redhat.com> wrote:
>> >>
>> >> [...]
>> >>
>> >> >> >> +void monitor_set_error(Monitor *mon, QError *qerror)
>> >> >> >> +{
>> >> >> >> + assert(0);
>> >> >> >
>> >> >> > Please use abort().
>> >> >>
>> >> >>
>> >> >> Why?
>> >> >
>> >> > Because assert(0) does not abort when compiled with -DNDEBUG.
>> >>
>> >>
>> >> Why is that a problem? And why isn't it a problem for the 300+ other
>> >> assertions in the code?
>> >
>> > We didn't support -DNDEBUG build until recently. Therefore it's safe
>> > to assume that also on production builds assert(0) was equal to
>> > abort(). If the default for production builds changes to -DNDEBUG, we
>> > want to retain the same abort() behaviour.
>> >
>> > The assertions which perform debugging checks aren't a problem. But
>> > assert(0) clearly isn't a debugging check, if you ever reach this line
>> > of code, it's abort() time.
>>
>>
>> I *know* that this line cannot be reached. That's why I asserted it
>> cannot be reached.
>>
>> If you want "this can't ever happen" assertions to be checked, then you
>> shouldn't define NDEBUG.
>
> If you know for sure that this line can't be reached, why bother with
> any code at all?
>
>> Do you still want me to use abort() here?
>
> abort() or nothing at all, as you wish.
Okay, nothing it is.
>> By the way, a quick grep shows >50 uses of assert(0).
>
> Not anymore since 43dc2a645e00e6761a741e3d16c27c5b5a373b66.
I doubt the wisdom of such a wholesale conversion, but I it's hardly
worth arguing now.
