[Cocci] Matching function pointer typedef

[Cocci] Matching function pointer typedef

[Jerome Glisse]

Hello,

I am trying to modify an function pointer typedef something like:

@@
@@
- typedef void (*toto_t)(int a, int b);
+ typedef void (*toto_t)(int a, int b, int c);

But it seems spatch or the semantic does not handle function pointer.
Or simply that typedef is not well handled in the first place. Thing
like:

@@
@@
- typedef int nombre;
+ typedef unsigned nombre;

also fails to work. But if typedef is use with struct then it works.
For instance:

@@
@@
- typedef struct {int a;} nombre;
+ typedef struct {unsigned a;} nombre;

Do work. Looking at declaration grammar i do not see why the former
does not work. I am using fedora 27 coccinelle 1.0.6 if that matters.

Is this a known limitation or am i writting it wrong ?

Thank you for any input on this,
J?r?me

[Cocci] Matching function pointer typedef

[Håkon Løvdal]

It's been a while since I used coccinelle, but I think to remember that you must
(and in any case should) keep the non-changing parts outside of the
+/- lines, e.g.

@@
@@
         typedef void (*toto_t)(int a, int b
+        , int c
         );

etc

BR H?kon L?vdal

On 15 May 2018 at 22:36, Jerome Glisse <jglisse@redhat.com> wrote:
> Hello,
>
> I am trying to modify an function pointer typedef something like:
>
> @@
> @@
> - typedef void (*toto_t)(int a, int b);
> + typedef void (*toto_t)(int a, int b, int c);
>
> But it seems spatch or the semantic does not handle function pointer.
> Or simply that typedef is not well handled in the first place. Thing
> like:
>
> @@
> @@
> - typedef int nombre;
> + typedef unsigned nombre;
>
> also fails to work. But if typedef is use with struct then it works.
> For instance:
>
> @@
> @@
> - typedef struct {int a;} nombre;
> + typedef struct {unsigned a;} nombre;
>
> Do work. Looking at declaration grammar i do not see why the former
> does not work. I am using fedora 27 coccinelle 1.0.6 if that matters.
>
> Is this a known limitation or am i writting it wrong ?
>
> Thank you for any input on this,
> J?r?me
> _______________________________________________
> Cocci mailing list
> Cocci at systeme.lip6.fr
> https://systeme.lip6.fr/mailman/listinfo/cocci

[Cocci] Matching function pointer typedef

[Julia Lawall]

On Tue, 15 May 2018, H?kon L?vdal wrote:

> It's been a while since I used coccinelle, but I think to remember that you must
> (and in any case should) keep the non-changing parts outside of the
> +/- lines, e.g.
>
> @@
> @@
>          typedef void (*toto_t)(int a, int b
> +        , int c
>          );

I think that typedefs of function pointers just don't work.  It is looking
for typedef type name;.  I can try to fix this.

julia

>
> etc
>
> BR H?kon L?vdal
>
> On 15 May 2018 at 22:36, Jerome Glisse <jglisse@redhat.com> wrote:
> > Hello,
> >
> > I am trying to modify an function pointer typedef something like:
> >
> > @@
> > @@
> > - typedef void (*toto_t)(int a, int b);
> > + typedef void (*toto_t)(int a, int b, int c);
> >
> > But it seems spatch or the semantic does not handle function pointer.
> > Or simply that typedef is not well handled in the first place. Thing
> > like:
> >
> > @@
> > @@
> > - typedef int nombre;
> > + typedef unsigned nombre;
> >
> > also fails to work. But if typedef is use with struct then it works.
> > For instance:
> >
> > @@
> > @@
> > - typedef struct {int a;} nombre;
> > + typedef struct {unsigned a;} nombre;
> >
> > Do work. Looking at declaration grammar i do not see why the former
> > does not work. I am using fedora 27 coccinelle 1.0.6 if that matters.
> >
> > Is this a known limitation or am i writting it wrong ?
> >
> > Thank you for any input on this,
> > J?r?me
> > _______________________________________________
> > Cocci mailing list
> > Cocci at systeme.lip6.fr
> > https://systeme.lip6.fr/mailman/listinfo/cocci
> _______________________________________________
> Cocci mailing list
> Cocci at systeme.lip6.fr
> https://systeme.lip6.fr/mailman/listinfo/cocci
>

[Cocci] Matching function pointer typedef

[Jerome Glisse]

On Tue, May 15, 2018 at 10:53:36PM +0200, Julia Lawall wrote:
> 
> 
> On Tue, 15 May 2018, H?kon L?vdal wrote:
> 
> > It's been a while since I used coccinelle, but I think to remember that you must
> > (and in any case should) keep the non-changing parts outside of the
> > +/- lines, e.g.
> >
> > @@
> > @@
> >          typedef void (*toto_t)(int a, int b
> > +        , int c
> >          );
> 
> I think that typedefs of function pointers just don't work.  It is looking
> for typedef type name;.  I can try to fix this.
> 

Above does not work either. The error is same roughly spatch complains
that it matches whole content ... I have a workaround, namely abusing
gcc which accept:

typedef void toto_t(int a, int b);

For function pointer typedef and then coccinelle on function declaration
do work. Still it would be nice if coccinelle can understand function
pointer typedef.

J?r?me

[Cocci] Matching function pointer typedef

[Jerome Glisse]

On Tue, May 15, 2018 at 05:26:59PM -0400, Jerome Glisse wrote:
> On Tue, May 15, 2018 at 10:53:36PM +0200, Julia Lawall wrote:
> > 
> > 
> > On Tue, 15 May 2018, H?kon L?vdal wrote:
> > 
> > > It's been a while since I used coccinelle, but I think to remember that you must
> > > (and in any case should) keep the non-changing parts outside of the
> > > +/- lines, e.g.
> > >
> > > @@
> > > @@
> > >          typedef void (*toto_t)(int a, int b
> > > +        , int c
> > >          );
> > 
> > I think that typedefs of function pointers just don't work.  It is looking
> > for typedef type name;.  I can try to fix this.
> > 
> 
> Above does not work either. The error is same roughly spatch complains
> that it matches whole content ... I have a workaround, namely abusing
> gcc which accept:
> 
> typedef void toto_t(int a, int b);
> 
> For function pointer typedef and then coccinelle on function declaration
> do work. Still it would be nice if coccinelle can understand function
> pointer typedef.

Ok my work around does work that well, it seems gcc allow this abuse
only for function parameter type, it can not be use as a type inside
a structure and so it doesn't cover all cases i am dealing with.

Is there a way to modify code using python inside coccinelle ? I have
seen example that print thing inside coccinelle but i haven't not seen
an example that use coccinelle to match something and then do changes
to the code using python.

I am guessing i can open the file and use the position information to
do that but was wondering if there is already helper provided for that.

Cheers,
J?r?me

[Cocci] Matching function pointer typedef

[Julia Lawall]

On Wed, 16 May 2018, Jerome Glisse wrote:

> On Tue, May 15, 2018 at 05:26:59PM -0400, Jerome Glisse wrote:
> > On Tue, May 15, 2018 at 10:53:36PM +0200, Julia Lawall wrote:
> > >
> > >
> > > On Tue, 15 May 2018, H?kon L?vdal wrote:
> > >
> > > > It's been a while since I used coccinelle, but I think to remember that you must
> > > > (and in any case should) keep the non-changing parts outside of the
> > > > +/- lines, e.g.
> > > >
> > > > @@
> > > > @@
> > > >          typedef void (*toto_t)(int a, int b
> > > > +        , int c
> > > >          );
> > >
> > > I think that typedefs of function pointers just don't work.  It is looking
> > > for typedef type name;.  I can try to fix this.
> > >
> >
> > Above does not work either. The error is same roughly spatch complains
> > that it matches whole content ... I have a workaround, namely abusing
> > gcc which accept:
> >
> > typedef void toto_t(int a, int b);
> >
> > For function pointer typedef and then coccinelle on function declaration
> > do work. Still it would be nice if coccinelle can understand function
> > pointer typedef.
>
> Ok my work around does work that well, it seems gcc allow this abuse
> only for function parameter type, it can not be use as a type inside
> a structure and so it doesn't cover all cases i am dealing with.
>
> Is there a way to modify code using python inside coccinelle ? I have
> seen example that print thing inside coccinelle but i haven't not seen
> an example that use coccinelle to match something and then do changes
> to the code using python.
>
> I am guessing i can open the file and use the position information to
> do that but was wondering if there is already helper provided for that.

No, python code doesn't have access to the ast, which is an ocaml
structure.

julia

>
> Cheers,
> J?r?me
>

[Cocci] Matching function pointer typedef

[Julia Lawall]

On Tue, 15 May 2018, Jerome Glisse wrote:

> Hello,
>
> I am trying to modify an function pointer typedef something like:
>
> @@
> @@
> - typedef void (*toto_t)(int a, int b);
> + typedef void (*toto_t)(int a, int b, int c);

This now works.  Currently, you need to remove the whole typedef and add
it back, not just add the third argument as H?kon suggested.

>
> But it seems spatch or the semantic does not handle function pointer.
> Or simply that typedef is not well handled in the first place. Thing
> like:
>
> @@
> @@
> - typedef int nombre;
> + typedef unsigned nombre;

Was this just an experiment, or is it something you need?  I would imagine
that the problem is that "unsigned" as a type by itself is not well
supported.

julia

> also fails to work. But if typedef is use with struct then it works.
> For instance:
>
> @@
> @@
> - typedef struct {int a;} nombre;
> + typedef struct {unsigned a;} nombre;
>
> Do work. Looking at declaration grammar i do not see why the former
> does not work. I am using fedora 27 coccinelle 1.0.6 if that matters.
>
> Is this a known limitation or am i writting it wrong ?
>
> Thank you for any input on this,
> J?r?me
> _______________________________________________
> Cocci mailing list
> Cocci at systeme.lip6.fr
> https://systeme.lip6.fr/mailman/listinfo/cocci
>

[Cocci] Matching function pointer typedef

[Jerome Glisse]

On Thu, May 17, 2018 at 09:28:00PM +0200, Julia Lawall wrote:
> 
> 
> On Tue, 15 May 2018, Jerome Glisse wrote:
> 
> > Hello,
> >
> > I am trying to modify an function pointer typedef something like:
> >
> > @@
> > @@
> > - typedef void (*toto_t)(int a, int b);
> > + typedef void (*toto_t)(int a, int b, int c);
> 
> This now works.  Currently, you need to remove the whole typedef and add
> it back, not just add the third argument as H?kon suggested.

Awesome ! :) Thank you for working on that.

> 
> >
> > But it seems spatch or the semantic does not handle function pointer.
> > Or simply that typedef is not well handled in the first place. Thing
> > like:
> >
> > @@
> > @@
> > - typedef int nombre;
> > + typedef unsigned nombre;
> 
> Was this just an experiment, or is it something you need?  I would imagine
> that the problem is that "unsigned" as a type by itself is not well
> supported.

This was just an experiment i did while trying to understand why my
function pointer typedef changes did not work. If i use long or any
basic ctype then it works.

Cheers,
J?r?me

[Cocci] Matching function pointer typedef

[Julia Lawall]

On Fri, 18 May 2018, SF Markus Elfring wrote:

> >> I am trying to modify an function pointer typedef something like:
> >>
> >> @@
> >> @@
> >> - typedef void (*toto_t)(int a, int b);
> >> + typedef void (*toto_t)(int a, int b, int c);
> >
> > This now works.
>
> Thanks for another software improvement.
> https://github.com/coccinelle/coccinelle/commit/80dcf751e411f1421fdcdb7119a95e50b841609c
>
>
> > Currently, you need to remove the whole typedef and add it back,
>
> I find this information inappropriate because such a source code replacement
> was attempted.

It was attempted before the change was made.

julia

>
>
> > not just add the third argument as H?kon suggested.
>
> It can be clearer to limit changes precisely while it can be also occasionally nice
> to extend the adjustment a bit so that the affected source code will be
> hopefully reformatted by the Coccinelle software in a desired way.
> https://github.com/coccinelle/coccinelle/issues/37
>
> Regards,
> Markus
>

[Cocci] Matching function pointer typedef

[Julia Lawall]

On Fri, 18 May 2018, SF Markus Elfring wrote:

> >>> Currently, you need to remove the whole typedef and add it back,
> >>
> >> I find this information inappropriate because such a source code replacement
> >> was attempted.
> >
> > It was attempted before the change was made.
>
> I get the impression from such feedback that another clarification
> would be useful about the granularity if each function parameter can be
> adjusted in data type definitions finally.

Concretely, the typedef name has to appear on the same line as all the
parameters.

julia