[cfe-commits] [PATCH] Thread-safety analysis: allow 'this' to be used as a valid lock expression

[Delesley Hutchins]

Thanks for pointing out the static data member issue; I'll definitely
need to fix that.  In some cases, clang expands the AST to better
reflect the semantics of the expression (e.g. inserting an implicit
cast, or converting myvar to this->myvar for an ordinary data member),
while in other cases it doesn't (e.g. not converting this->lock to
S::lock below).  I still haven't got my head around all the details.
:-)  However, it is clear to me that I'll need to do a substantial
overhaul of the mutexID stuff to do a better job of canonicalizing
lock expressions; that will resolve issues like this one.

  -DeLesley

On Tue, Jan 17, 2012 at 11:02 AM, Richard Smith <richard at metafoo.co.uk> wrote:
> Hi,
>
> On Fri, January 13, 2012 22:39, Delesley Hutchins wrote:
>> This patch modifies the thread safety analysis, so that 'this' can be
>> used as a valid lock expression.
>
> Using '0' to represent 'this' is a little subtle; I'd like to see a comment on
> either MutexID or MutexID::DeclSeq calling this out. With that, LGTM.
>
> One observation: this change breaks code which uses 'this->m' to name a mutex
> which is a static data member (such a reference no longer matches against
> naming the mutex as 'Class::m'). However, code which used 'p->m' was already
> broken in this way, and the fix for that more general case should take care of
> the 'this' case too, so I don't think that is an issue with this patch per se.
> Testcases:
>
> struct __attribute__((lockable)) S {
>  static S *lock;
>  static int n __attribute__((guarded_by(*lock)));
>  int f() __attribute__((exclusive_locks_required(*this->lock))) {
>    return n; // was accepted, now rejected
>  }
>  int g(S *p) __attribute__((exclusive_locks_required(*p->lock))) {
>    return n; // rejected, should probably be accepted
>  }
> };
>
> These may seem like strange ways to write such code, but the 'this->' form, at
> least, might plausibly be used in templates.
>
> - Richard
>



-- 
DeLesley Hutchins | Software Engineer | delesley at google.com | 505-206-0315