[Cocci] Multiple match versus single header result in conflicts

[jglisse@redhat.com (Jerome Glisse)]

On Mon, May 07, 2018 at 09:57:43PM +0200, Julia Lawall wrote:
> 
> 
> On Mon, 7 May 2018, Jerome Glisse wrote:
> 
> > On Sat, May 05, 2018 at 07:20:04AM +0200, Julia Lawall wrote:
> > >
> > >
> > > On Fri, 4 May 2018, Jerome Glisse wrote:
> > >
> > > > Following semantics does not update function prototype in header file:
> > > >
> > > > @S@
> > > > identifier I1, I2;
> > > > @@
> > > > struct myop I1 = { ..., .add = I2 , ... };
> > > >
> > > > @U depends on S@
> > > > identifier S.I2;
> > > > identifier A1, A2;
> > > > type T1, T2;
> > > > @@
> > > > int I2(T1 A1,
> > > > +int c,
> > > > T2 A2) { ... }
> > > >
> > > > run with spatch --in-place --sp-file test.spatch --dir . --all-includes
> > > > (tested various includes/headers combinations) on 3 files f1.h f1.c
> > > > f2.c (if f1.c and f2.c are merge together then thing works).
> > > >
> > > > f1.h: ----------------------------------------------------------------
> > > > struct myop {
> > > >     int (*add)(int, int);
> > > > };
> > > >
> > > > int myadd(int, int);
> > > > ----------------------------------------------------------------------
> > > >
> > > > f1.c: ----------------------------------------------------------------
> > > > #include "f1.h"
> > > >
> > > > int myadd(int a, int b)
> > > > {
> > > >     return a + b;
> > > > }
> > > >
> > > > struct myop myop = {
> > > >     .add = myadd,
> > > > };
> > > > ----------------------------------------------------------------------
> > > >
> > > > f2.c: ----------------------------------------------------------------
> > > > #include "f1.h"
> > > >
> > > > int myadd2(int a, int b)
> > > > {
> > > >     return a + b;
> > > > }
> > > >
> > > > struct myop myop2 = {
> > > >     .add = myadd2,
> > > > };
> > > > ----------------------------------------------------------------------
> > > >
> > > > If f1.c and f2.c are just one file than the header files is properly
> > > > updated. The error message is:
> > > >
> > > > different modification result for ./f1.h
> > > >
> > > > I am not sure if there is a way to make the semantic patch work against
> > > > such scenario. So is this expected ? Is my semantic patch wrong ? Or
> > > > is it a bug in coccinelle ?
> > >
> > > I believe that Coccinelle just doesn't make the effort to realize that the
> > > modifications are the same.  Probably things will be fine if you don't use
> > > --in-place.
> > >
> >
> > Yes it does work thank you for quick answer. By the way my next hurdle
> > is trying to match function prototype no matter if argument has a name
> > or not ie matching all:
> >
> > void toto(int,int);
> > void toto(int a, int);
> > void toto(int, int b);
> > void toto(int a, int b);
> >
> > This tie back to my original issue, when a function callback is use
> > in myop struct in one file and prototype is in header file, i want to
> > update prototype so that latter when coccinelle process the different
> > file in which the function is defined i can use the modified header
> > file to also update the function definition.
> 
> This should work already.  When you change the function definition, if it
> has access to the prototype it should change it as well.  If this is not
> working, please send an example.

Roughly same as above:

f1.h: ----------------------------------------------------------------
struct myop {
    int (*add)(int, int);
};

int myadd(int, int);
----------------------------------------------------------------------

f1.c: ----------------------------------------------------------------
#include "f1.h"

struct myop myop = {
    .add = myadd,
};
----------------------------------------------------------------------

f2.c: ----------------------------------------------------------------
#include "f1.h"

int myadd(int a, int b)
{
    return a + b;
}
----------------------------------------------------------------------

Semantic: ------------------------------------------------------------
@S@
identifier I1, I2;
@@
struct myop I1 = { ..., .add = I2 , ... };

@depends on S@
identifier S.I2;
identifier A1, A2;
type T1, T2;
@@
int I2(T1 A1,
+int c,
T2 A2) { ... }
----------------------------------------------------------------------

So because the function is declared in a different files this does not
work unless i group process all files. Sadly as i am working on the
linux kernel for the real case i am interested in i can not create
file group easily (a group would be hundreds of different files scatter
around different directories and with no easy way to find them).

So:
WORKS: spatch --in-place --sp-file t.spatch --all-includes *.c
NOOP:  spatch --in-place --sp-file t.spatch --all-includes f1.c
NOOP:  spatch --in-place --sp-file t.spatch --all-includes f2.c

Other solution might be to first use spatch to find all myops.add
function and then from collected function name find all the c files
which define them.

Hence why i wanted to update all header file in first pass (matching
on myop.add and function prototype) and then match function prototype
that with function declaration that are missing a parameter. This
seems like the easiest path.

Cheers,
J?r?me