Re: [git patches] two warning fixes

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Jul 19, 2007, at 14:04:29, Linus Torvalds wrote:
On Thu, 19 Jul 2007, Krzysztof Halasa wrote:
Jeff Garzik <[email protected]> writes:
My overall goal is killing useless warnings that continually obscure real ones.

Precisely, the goal should be to make must_check (and similar things) warn only in real cases.

.. the problem with that mentality is that it's not how people work.

People shut up warnings by adding code.

Adding code tends to add bugs.

People don't generally think "maybe that warning was bogus".

More people *should* generally ask themselves: "was the warning worth it?" and then, if the answer is "no", they shouldn't add code, they should remove the thing that causes the warning in the first place.

For example, for compiler options, the correct thign is often to just say "that option was broken", and not use "-fsign-warning", for example. We've literally have had bugs *added* because people "fixed" a sign warning. More than once, in fact.

Every time you see a warning, you should ask yourself: is the warning interesting, correct and valid? And if it isn't all three, then the problem is whatever *causes* the warning, not the code itself.

I agree that there are a fair number of things (like the sysfs calls) that should just WARN() when they hit an error, but I also think that we're currently missing a *lot* of __must_check's that we should have. For example a friend of mine was having problems with an HDAPS patch where it just kind of hung. Turns out the problem was that the code blithely called scsi_execute_async() and then put itself to sleep on a completion... except scsi_execute_async() returned failure and the completion would never complete.

For instance, I would bet that a fair number of the other int- returning functions in include/scsi/scsi_device.h want __must_check on them. That said, the person adding the __must_check should be REQUIRED to do at least a superficial audit of the code.

I'd propose a few simple rules:
(1) If it can return the only pointer to freshly-allocated pointer then it's __must_check (2) If it can return a hard error which the caller must handle specially, then it's __must_check (3) If the only possible error is a kernel bug then make the damn thing return void and give it a big fat WARN() when it fails.
  (4) For any other case (or if you are unsure), don't flag it.

And of course the burden of proof is on the person trying to add the __must_check.

Cheers,
Kyle Moffett

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

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux