Re: [Qemu-devel] [PATCH v2 04/11] translate: downgrade IRQ BQL asserts to tcg_debug_assert

[Eric Blake]

On 03/03/2017 01:47 PM, Eric Blake wrote:
> On 03/03/2017 01:35 PM, Richard Henderson wrote:
>>
>> Which is exactly the point when you have a condition like (X > 0);
>> letting the compiler have the same information for the production build
>> that it would have gleaned from the debug build.
>>
>> But that's not the same as dropping the assert, which is what you wanted.
>>
>> On the other hand, isn't "assert" instead of "g_assert" exactly what you
>> wanted?  Don't we define NDEBUG for production builds?
> 
> No - no one in their right mind defines NDEBUG for qemu.  For better or
> worse, there are too many places where we liberally use assert(), and
> where the code WILL crash and burn when it falls through to subsequent
> code if a failed assert() is not equivalent to a fatal exit.  (I still
> try to make sure we avoid any new side-effects in assert in my code
> reviews, but there's no way you'll convince me to audit the code base
> for NDEBUG-safety violations).

A quick git grep shows, among others:

hw/scsi/mptsas.c:     * When we do, we might be able to re-enable NDEBUG
below.
hw/scsi/mptsas.c:#ifdef NDEBUG
hw/scsi/mptsas.c:#error building with NDEBUG is not supported
hw/virtio/virtio.c:     * When we do, we might be able to re-enable
NDEBUG below.
hw/virtio/virtio.c:#ifdef NDEBUG
hw/virtio/virtio.c:#error building with NDEBUG is not supported


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

signature.asc
Description: OpenPGP digital signature