Re: [Xen-devel] [PATCH for-4.8] ipxe: update to newer commit

[Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>]

On 10/11/2016 09:32 AM, Ian Jackson wrote:
> Boris Ostrovsky writes ("Re: [Xen-devel] [PATCH for-4.8] ipxe: update to 
> newer commit"):
>> On 10/11/2016 05:52 AM, Ian Jackson wrote:
>>> Maybe we could consider these as backports to earlier releases.
>>> However, I looked at the patch and it mostly seems to be removing null
>>> pointer checks.  I find this ... surprising.
>> That's because routines have __nonnull attribute (which tells compiler
>> that arguments are never NULL) and new gcc warns on non-NULL checks for
>> these arguments.
> *sigh*  Why don't we just disable that warning ?

We could but what if an old compiler doesn't support that option?
Although it looks like -Wno-<option>, which is what we'd use, may be OK:

ostr@workbase> gcc foo.c
ostr@workbase> gcc -Wfoo foo.c
gcc: error: unrecognized command line option ‘-Wfoo’; did you mean ‘-Wno-’?
ostr@workbase> gcc -Wno-foo foo.c
ostr@workbase>


>
>> Another interesting new warning that is fatal with -Werror is
>>     if(a)
>>         foo();
>>         bar();
>>
>> gcc warns that bar() is indented and maybe braces are needed.
> Do we actually have cases like this ?  Are they real bugs ?

Yes we have (for example igb_phy.c change in the patch that I sent) and
no, they don't look like bugs.

>
>> BTW, another option for backporting may be removing -Werror. If we know
>> we are not changing sources then we might consider this.
> Perhaps we could disable warnings more selectively.

I scanned the changes again and at least one appears to be fixing a
legitimate bug (buffer overrun). There is an upstream patch for that,
which is essentially what I have there, but not as a separate patch.

-boris



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel