Re: [Qemu-devel] [RFC PATCH v2 30/30] trace: Force compiler warnings on trace parameter type mismatches

[Eric Blake]

On 03/13/2017 02:55 PM, Eric Blake wrote:

> ---
> 
> RFC for two reasons:
> 1. the Makefile issue documented above means that incremental
> builds won't benefit from this patch without manual intervention
> (fresh builds, including docker, manage to test it, though)

Dan's posted the fix for this one.

> 2. there are still failures under 'make address@hidden'
> due to more type mismatches that still need to be squashed. I'm
> still working on fixing those, but wanted to at least post this
> series for initial review, especially so the maintainer can weigh
> in on how much (or little) belongs in 2.9

mingw is a nightmare.  Things like ntohl() having non-standard behavior
on 32-bit mingw can only be worked around by hairy redefinitions in
osdep.h or by casts at the callsites. See the conversation in 6/30, but
I'm hoping gcc will give us a new knob to tone down the level of
-Wformat warnings for two types that have different rank but the same
width; if gcc gives us that, then we'd add a configure test, and
conditionally generate the type-mismatch checking code ONLY when gcc is
smart enough to not care about rank mismatch.  Until then, this RFC
30/30 proved useful for flushing out the earlier cleanups (many of which
you can/should still accept), but should NOT be incorporated as-is, not
even when 2.10 opens.

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

signature.asc
Description: OpenPGP digital signature