[Cocci] Replacing one (specific!) type with another

[Nikolaus@rath.org (Nikolaus Rath)]

On Oct 08 2016, Julia Lawall <julia.lawall@lip6.fr> wrote:
> On Fri, 7 Oct 2016, Nikolaus Rath wrote:
>
>> On Oct 05 2016, Julia Lawall <julia.lawall@lip6.fr> wrote:
>> >>
>> >> 2. ..and how would I go about if instead of the type, I want to replace
>> >>    a variable name? (my_type *ptr --> my_type *pointer).
>> >
>> > I'm not completely sure what the issue is here.  Do you specifically want
>> > to convert ptr to pointer?  Is the type important?  To make exactly what
>> > you have written, you could put:
>> >
>> > @@
>> > typedef my_type;
>> > idexpression mytype * p1;
>> > @@
>> >
>> > - ptr at p1
>> > + pointer
>> >
>> > This checks for the word ptr, and also checks that it is an identifier of
>> > the right type.  I haven't tested it, so let me know if there is any
>> > problem.
>>
>> It workes somewhat... but not completely.
>>
>> Here's what I wanted to do: I merged two structures (struct fuse_session
>> and struct fuse_ll) into one (struct fuse_session). I've first replaced
>> all the type names, and then manually fixed the cases where this
>> resulted in bogus/redundant code (typically in functions that used to
>> work with both structs).
>>
>> Now my project compiles and runs fine, but I the variable naming is
>> inconsistent: in some cases the struct fuse_session pointer is called
>> *se (these were the variables that were always of type struct
>> fuse_session), and in other cases the pointer is called *f (these were
>> the variables that were previously of type struct fuse_ll).
>>
>> I'd like to fix this too, and always refer call fuse_session pointers
>> "se" (except where this name is already used for something else, but
>> I'll just fix this up by hand afterwards). I tried the following patch:
>>
>> $ cat se-name.cocci
>> @@
>> idexpression struct fuse_session *p1;
>> @@
>> - f at p1
>> + se
>>
>> but it resulted in these changes:
>>
>>  	struct fuse_session *f = req->se;
>> -	struct cuse_data *cd = f->cuse_data;
>> -	size_t bufsize = f->bufsize;
>> +	struct cuse_data *cd = se->cuse_data;
>> +	size_t bufsize = se->bufsize;
>>
>> So it seems to replace the variable where its used, but not where it's
>> defined.
>>
>> Is there a way to catch the definitions too?
>
> Write separate rules for that.  You would need one case for a local
> variable and one case fora parameter.  You can actually probably just drop
> the rule you have currently.  I would imagine something like the
> following:
>
> @@
> symbol f, se; // avoid unneeded warnings from Coccinelle
> @@
>
> struct fuse_session *
> -f
> +se
>  ;
> <...
> -f
> +se
> ...>

Could you explain how this works (in particular the effect of the angle
brackets)? I also can't resist to point out that "symbol" is not
included in the list of metavariable types from the tutorial slides :-).

> @@
> identifier fn;
> @@
>
> fn(...,struct fuse_session *f,...) { <...
> -f
> +se
> ...> }
>
> I think that the symbol declaration has effect in the rest of the semantic
> patch, and does not have to be repeated.  If you get warnings for the
> second rule, just copy it down.

Not sure what you mean with "copy it down". I don't get Coccinelle
warnings, but if I just use the two rules as you gave them, then
it looks as if the second one isn't working:

@@ -584,9 +584,9 @@ static struct fuse_ll_pipe *fuse_ll_get_
 
 static void fuse_ll_clear_pipe(struct fuse_session *f)
 {
-	struct fuse_ll_pipe *llp = pthread_getspecific(f->pipe_key);
+	struct fuse_ll_pipe *llp = pthread_getspecific(se->pipe_key);
 	if (llp) {
-		pthread_setspecific(f->pipe_key, NULL);
+		pthread_setspecific(se->pipe_key, NULL);
 		fuse_ll_pipe_free(llp);
 	}
 }


Is the problem that "...," does not match nothing, i.e. *f must not be
the first argument of the function?


Best,
-Nikolaus

-- 
GPG encrypted emails preferred. Key id: 0xD113FCAC3C4E599F
Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F

             ?Time flies like an arrow, fruit flies like a Banana.?