Re: [Cocci] Removing the last return statement from a void function

[Julia Lawall <julia.lawall@inria.fr>]

On Thu, 18 Mar 2021, Thomas Adam wrote:

> Hello all,
>
> I've another Coccinelle question I'm hoping you can help me with.  The
> codebase I'm working on is old, and has some interesting styles which
> by themselves probably don't cause any problems, but newer C compilers
> are now starting to flag them.
>
> In particular, there seems to be a pattern in this code base of using
> explicit `return;` statements at the end of void functions.  Here's an
> example:
>
> static void broadcast_mini_icon(FvwmWindow *fw)
> {
>     if (!FMiniIconsSupported)
>     {
>         return;
>     }
>     if (fw->mini_pixmap_file && fw->mini_icon)
>     {
>         BroadcastFvwmPicture( M_MINI_ICON, FW_W(fw),
>             FW_W_FRAME(fw), (unsigned long)fw,
>             fw->mini_icon, fw->mini_pixmap_file);
>     }
>     return;
> }
>
> Here you can see the last return statement is not necessary.
>
> I'm trying to make coccinelle recognise this and remove such cases.
> Here's what I've tried:
>
> @@
> identifier f;
> @@
>
> void f(...) {
>   <...
> - return;
> ...>
>
> }
>
> ... which sort of works, but proceeds to remove *all* `return;`
> statements from void functions, rather than the last occurance in the
> function.
>
> Am I on the right track with this approach, or do I need to do
> something more creative?

The ... in Coccinelle is based on control flow, so it is a bit hard to
find the return at the bottom of the function.  Actually, from
Coccinelle's point of view, all returns are at the bottom of the function,
because one leaves the function after a return.

You can try the following:

@r@
position p;
identifier f;
}

f(...) {
<...
{ .. return@p; }
...>
}

@@
position p != r.p;
@@

- return@p;

Basically the first rule collects the position of all returns that are
inside a { }, and then the second rule removes the others.

However there is an isomorphism that makes a pattern with { ... S } match
just S, for any S, which you don't want.  So you can make an empty file
called empty.iso, and then run the rule with the command-line argument
--iso-file empty.iso

julia
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci