Re: [PATCH v2] NVMe: silence GCC warning on 32 bit

[Paul Bolle <pebolle@tiscali.nl>]

Matthew,

Paul Bolle schreef op ma 24-03-2014 om 16:36 [+0100]:
> On Mon, 2014-03-24 at 09:31 -0400, Matthew Wilcox wrote:
> > I should try things myself before opening my big mouth.  Weird.  Using
> > gcc-4.8, I see the same thing.  Guess I should just apply the patch,
> > though it feels wrong to be initialising an entire struct just to silence
> > a bogus compiler warning :-(
> 
> I noticed this difference on a 32 bit x86 machine and a 64 bit x86
> machine that are both running Fedora 20. They both should be at
> gcc-4.8.2 for quite some time now (if I grepped the yum log correctly).
> 
> Anyhow, the warning on 32 bit is rather noisy, so I wanted it gone. But
> my comments should make clear I'm not really happy with this patch. 
> 
> And as this is now unlikely to be in time for v3.14, we might decide to
> dig deeper. It won't be the first time that a rather small change (say,
> converting a variable from signed to unsigned) turns out be enough to
> make GCC understand the flow of the code.

This noisy warning is still seen when compiling v3.15-rc4 for x86 (32
bit, that is). Assuming this patch is not queued anywhere: is the
unsophisticated approach of my v2 acceptable or would you like me to try
and find the cause of this warning?


Paul Bolle