[Cocci] … adding missing NULL checks

[julia.lawall@lip6.fr (Julia Lawall)]

On Thu, 22 Dec 2016, Thomas Adam wrote:

> On Tue, Dec 20, 2016 at 08:45:26PM +0100, SF Markus Elfring wrote:
> > * A filter criterium like ?T = strdup(...);? was specified.
> >   How do you think about to use a SmPL disjunction (or an advanced SmPL constraint
> >   on a metavariable with the type ?identifier? instead)?
> >
> > * Can the function call ?pkg_emit_errno? become a bit more generic by passing
> >   a string from previously matched data?
> >
> > * Would you like to check any more cases so that the technology ?data flow analysis?
> >   will be also needed here?
>
> You're right.  The next step I'm looking at is making my checks automatable so
> that they can truly patch code properly.  Right now, what I have is merely a
> "best guess" placeholder which is inserted, leaving the programmer having to
> go in and correct the compilation.
>
> Consider the following rule:
>
> 	T = strdup(...);
> 	+ if (T == NULL) {
> 	+ 	pkg_errno("strdup", __func__);
> 	+	return (EPKG_FATAL);
> 	+ }
> 	... when != (T == NULL)
> 	    when != (T != NULL)
> 	    ? T = strdup(...);
>
> The problem is that if this rule matches, the inserted code will always be:
>
> 	pkg_errno("stdup", __func__);
> 	return (EPKG_FATAL);
>
> That return should *not* be there if the return type of the function the code is
> being inserted to is anything other than "int".  Likewise, if the function
> returns void, it should be "return;", or a pointer should be "NULL".
>
> Is there a clever way of conditionally making this change?

Doing something based on the type would be possible.  For example, you can
have a pattern like (simplified version):

void f(...) { <...
   x = foo();
+  if (!foo) return;
   ...> }

Choosing real error handling code that would free the previously allocated
resources is probably beyond the abilities of Coccinelle.

julia