Re: [BUG REPORT] use of unreachable() masks uninitialized variables warnings

[Jeff Merkey]

On 2/11/16, Theodore Ts'o <tytso@xxxxxxx> wrote:
> On Thu, Feb 11, 2016 at 11:12:12AM -0700, Jeff Merkey wrote:
>>
>> That's good to know, they could be false positives, but it was kind of
>> wierd behavior caused by that macro.
>
> If it is true, it sounds more like a compiler bug to me.  Any
> statements a BUG() call are unreachable.  If the compiler assumes that
> in the case of:
>
> 	if (a)
> 	        BUG();
> 	printf("foo bar\n");
> 	
> That the printf is not reachable, that's a compiler bug.  And not a
> problem in the BUG() macro.
>
> It might be that it's worthwhile to use other static code analysis
> tools.  Many people will look at warnings from Coverity and clang to
> find potential problems, since these tend to find more warnings than
> just using gcc.  The problem with some of these, including Coverity,
> is that they can be __too__ noisy, and if 90% of the warnings are
> false positives, most people won't take the time to weed out several
> dozen bogus warnings to find the one good one.
>
> In your case, for example, I looked through several dozen warnings,
> and they were ***all*** bogus.  Keep in mind that this might make me
> less inclined to pay attention to complaints from you in the future.
> The story of the buy who cried wolf too often comes to mind.
>
> Perhaps you could actually take a close look at the warnings, before
> you fire off an e-mail?  If at least one of the warnings were valid
> and pointed at an actual bug, it wouldn't have been a complete waste
> of my time....
>
> Best regards,
>
> 						- Ted
>

Well, it's nice to know I wasn't imagining what I was seeing.  It
wasn't a complete waste of my time or yours since it revealed a
problem with gcc issuing warnings.  Not all of them were bogus BTW,
just inert would be better word.  I noticed it while enabling the
ability of the BUG() macro to emit an int3 instead of a ud2
instruction.

Jeff