[cfe-commits] [Patch] Checker for assignment of non-Boolean value to Boolean variable

[David Blaikie]

On Thu, Dec 8, 2011 at 9:14 PM, Anna Zaks <ganna at apple.com> wrote:

>
> On Dec 8, 2011, at 6:23 PM, David Blaikie wrote:
>
> > On Thu, Dec 8, 2011 at 6:03 PM, Eli Friedman <eli.friedman at gmail.com>
> wrote:
> >> On Wed, Dec 7, 2011 at 5:53 PM, Ryan Govostes <rzg at apple.com> wrote:
> >>> Hello,
> >>>
> >>> Please find attached a new checker for the static analyzer that warns
> on the assignment of a non-Boolean value to a variable with a Boolean data
> type. It supports Boolean data types defined by C++, C99, and Objective-C
> as well as those in stdbool.h and MacTypes.h.
> >>
> >> Some related work:
> >>
> >> http://blog.mozilla.com/mwu/2011/07/28/the-twelve-booleans-of-mozilla/
> >
> > Awesome.
> >
> > Yeah - now I understand the problem, I wonder again if this could be
> > implemented (in a simpler but more strict manner) as a front-end
> > warning in the same way that null-conversions are warned about (though
> > it'd have to be a bit smarter - perhaps detecting the macro/typedef
> > that was used to declare a variable or assign a constant to it - not
> > to mention that the current Clang null-conversion support is broken
> > (I've another thread pending some discussion on that)). But in general
> > this should be good for any >2 state value masquerading as an int -
> > for C++ bool or anything that's as good as it, a lesser warning (if
> > any at all) might be appropriate.
> >
>
> My understanding is that you are suggesting to add a compiler warning when
> a constant, which is not 0 nor 1, is being assigned to a boolean variable.
> Such warning would be weaker then this checker, so the checker should still
> go in.
>

I'm certainly not sufficiently informed to make any claim to veto this
checker - just tossing around some ideas.

I assume the "weakness" you're referring to, based on the compiler warning
you described is that it wouldn't catch this:

unsigned i = 3;
BOOL j = i;

That could be dealt with by checking that all assignments to/from BOOL (or
whatever other pseuod bool types exist) are from other BOOLs - this might
be somewhat noisy, I agree - the checker would be more surgical though the
general warning might suffice, especially for new code.

This might still be too expensive, I don't know - looking up function &
variable declarations whenever a BOOL expression is encountered to see how
the name was spelled. In which case something more like GNU's __null could
be used - a custom, hidden type that looks just like char or whatever the
intended underlying type is.

- David


>
> Cheers,
> Anna.
>
> > The usual warning with Win32 BOOL is to never mix BOOL & bool - but
> > with this kind of warning/checking you could be confident that BOOL is
> > only 0 or 1 and so "someBOOL == some_bool" would work sensibly (&
> > never be "2 == 1" => false, for example)
> >
> > - David
> >
> > _______________________________________________
> > cfe-commits mailing list
> > cfe-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20111208/747d8e3f/attachment.html>