[cfe-dev] virtual method calls in ctor/dtor

[Jordan Rose]

For the record, there's an alpha static analyzer checker for this: alpha.cplusplus.VirtualCall. I'm not quite sure what state it's in; it's been a long time since anyone's worked on it. It's purely doing a syntactic check, though; if you want the check to be interprocedural, it should be converted to path-sensitive.

In either the warning or static analyzer case, I would suggest allowing the warning to be silenced if the user explicitly writes a non-virtual call, e.g. ThisClass::doSomething().

Jordan


On Dec 5, 2013, at 8:50 , David Blaikie <dblaikie at gmail.com> wrote:

> If you're thinking of doing this as a Clang warning just a few thoughts
> 
> * it won't be interprocedural - that's just not practical at
> compile-time. So you won't catch the indirect calls (dtor calls free
> function foo, passing 'this, foo then calls virtual function on the
> passed in pointer)
> * you probably don't want/need the CFG - that's for path sensitive
> analysis. In this case I don't think there's anything path-sensitive
> you need to worry about, so you can probably just do this warning on
> the fly - you see a call, you check whether we're currently parsing
> the body of a dtor, then you check if it's a member call to the same
> object (again, probably best to just catch simple cases like
> "this->x()" or "x()" and not worry about "foo *f = this; ... f->x()")
> * Getting such a warning into Clang has a somewhat high bar. We avoid
> having warnings in Clang that have a high false positive (where "false
> positive" is any case where the warning fires and the users code
> doesn't have a bug, generally speaking) rate and I'm not sure if this
> warning would meet that bar without some heuristics/improvements - but
> I haven't given it much consideration just yet, perhaps you have ideas
> there.
> 
> On Thu, Dec 5, 2013 at 6:55 AM, Olaf Krzikalla
> <Olaf.Krzikalla at tu-dresden.de> wrote:
>> Hi all,
>> 
>> before I implement it by myself I'd rather ask. Maybe there is already a
>> solution or it is not as simple as I thought.
>> The question is, whether or not it is possible to implement a warning about
>> direct or indirect calls to virtual methods within ctors or dtors.
>> My approach would be to generate the cfg of the examined ctor/dtor and then
>> traverse the execution paths and see if there is such a call. The
>> interesting/difficult thing could be to keep track of the "this" pointer,
>> since calls by other objects should not lead to such a warning.
>> 
>> What do you think?
>> 
>> Best Olaf
>> 
>> 
>> 
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev