Date:	Wed, 12 Dec 2012 18:48:12 -0600
From:	Daniel Santos <danielfsantos@....net>
To:	Jan Beulich <JBeulich@...e.com>
CC:	Andrew Morton <akpm@...ux-foundation.org>,
	Rusty Russell <rusty@...tcorp.com.au>,
	Arjan van de Ven <arjan@...ux.intel.com>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] utilize _Static_assert() for BUILD_BUG_ON() when the
 compiler supports it

On 11/06/2012 03:23 AM, Jan Beulich wrote:
>> This sort of logic is normally performed via the
>> include/linux/compiler*.h system.  And
>>
>> 	grep __GNUC include/linux/*.h
>>
>> indicates that we've been pretty successful.  Can we do that here too?
>>
>> (eg: suppose the Intel compiler supports _Static_assert?)
>> Yeah, there are a lot of goodies here:
>>
>> _Static_assert:
>>          We could define __ASSERT_STRUCT_FIELD(e) for this:
>>          #define BUILD_BUG_ON_ZERO(e) \
>>                  sizeof(struct { __ASSERT_STRUCT_FIELD(e); })
> I considered something like this too, but it wouldn't work well: The
> diagnostic issued from a failed _Static_assert() would preferably
> refer to the original source expression, not an already macro
> expanded variant of it. That, however, precludes passing it
> through multiple levels of macro expansion. Which would leave
> passing e and #e to __ASSERT_STRUCT_FIELD(), but that's again
> ugly as it opens the door for someone adding a use where the two
> arguments don't match up.
I was under the assumption that _Static_assert doesn't exist unless you 
are using -std=c11 or -std=gnu11, but I admit I didn't look into it 
much. I'm very glad it was added to the standard. IMO, it would be 
better to use such a feature (presuming it behaves well) when it is 
available but, of course, we still need the best possible functionality 
when it isn't.

In either case, the new BUILD_BUG_ON macro simply stringifies the 
condition, so it will emit an error indicating the exact expression that 
failed. If you pass it a macro, macro expansion will not be performed 
prior to stringification.
>
>> __compiletime_error():
>>          I blame Arjan for this.  It disappears if not implemented, which
>>          is just lazy.  BUILD_BUG_ON() does this right, and he didn't fix
>>          that at the time :(
> Again, the name of the macro made me not use it, as the obvious
> fallback is a link time error. The only thing I would be agreeable to
> is something like __buildtime_error().

Yeah, I agree, since we speak more in terms of a "build" rather than 
compiling.  But I didn't mess with it in my patches.

>
>> I'd say we have three patches here, really:
>>
>> 1) Add __ASSERT_STRUCT_FIELD(e) to compiler.h
There are actually a number of reasons for having multiple mechanisms to 
break the build.  For instance, BUILD_BUG_ON_ZERO is used outside of 
function bodies, where a compound gcc expression ({expr; expr;}) isn't 
permitted.
>> 2) Add __UNIQUE_ID().
>> 3) Use them (I can think of at least one other place for __UNIQUE_ID()).
>>
>> Jan, do you want the glory?  :) If not, I'll respin.
> Depending on the answers to the above (i.e. how much of it
> actually would need re-doing and re-testing), it may take me
> some time to get to this, but beyond that I'm fine with trying
> to improve this to everyone's satisfaction.
Personally, I would rather we start a cpputil.h (or some such) to 
encapsulate all of our preprocessor kludgery, err, I mean wizardry.  
There must be 14 thousand paste macros in the kernel.  We could move 
stringify into it as well, but I suppose that's another issue.

Daniel
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists