[cocci] Matching function pointer casts

[cocci] Matching function pointer casts

[Mathias Krause]

Hi,

I have a hard time trying to come up with a semantic patch that would
match function pointer casts, specifically (void (*)(void)) in my case.

I want to get rid of these and replace them with a properly typed
functions that'd do necessary type conversions. (I already have that
part, though.)

Below are two attempts I tried and did not work;

1.
@pfunc_cast@
identifier fn;
fresh identifier fnthunk = "THUNK_" ## fn;
@@
- (void (*)(void)) &fn
+ &fnthunk

Above doesn't parse for me (spatch version 1.1.0).

2.
@pfunc_cast@
identifier fn;
fresh identifier fnthunk = "THUNK_" ## fn;
type T =~ "^void (*)(void)$";
@@
- (T) &fn
+ &fnthunk

as well as 3.
@pfunc_cast@
identifier fn;
fresh identifier fnthunk = "THUNK_" ## fn;
type T =~ "void \\(*\\)\\(void\\)";
@@
- (T) &fn
+ &fnthunk


This basically matches all address taking expressions, even without
casts, and replaces them by &THUNK_... -- not quite what I want. ;)
It looks like to completely ignoring the regular expression for the type
match beside complaining about syntax errors when one does some?

Is there a way to match function pointer casts in spatch at all?

Thanks,
Mathias

Re: [cocci] Matching function pointer casts

[Markus Elfring]

> Is there a way to match function pointer casts in spatch at all?

How do you think about to construct another test case variant
for the desired software clarification?

Regards,
Markus

Re: [cocci] Matching function pointer casts

[Markus Elfring]

> I meanwhile found a workaround but it's ugly -- not because cocci fails
> to match on function pointer casts (I implemented that by making use of
> its python script capabilities) but because of actual source code
> guards, the function pointer assignments are wrapped in.
> 
> It's an '#if ... #else ... #endif' block that, apparently, doesn't get
> processed correctly. The preprocessor condition is non-trivial as in not
> simply being "0" or "1" but still easy enough to handle (one level of indirection).

Thanks for such background information.

There are some development challenges remaining for better support
of conditional “compilation” (or analysis) also by the means of the Coccinelle software.

Regards,
Markus