Hi Julia,
On Wed, 2016-05-11 at 11:51:43 +0200, Julia Lawall wrote:
>
>
> On Tue, 10 May 2016, Sören Brinkmann wrote:
>
> > On Wed, 2016-05-11 at 07:59:06 +0200, Julia Lawall wrote:
> > >
> > >
> > > On Tue, 10 May 2016, Sören Brinkmann wrote:
> > >
> > > > Hi Julia,
> > > >
> > > > On Wed, 2016-05-11 at 07:41:20 +0200, Julia Lawall wrote:
> > > > >
> > > > >
> > > > > On Tue, 10 May 2016, Sören Brinkmann wrote:
> > > > >
> > > > > > Hi,
> > > > > >
> > > > > > I'm trying to remove 'const' qualifiers from function arguments
> > > > > > that are
> > > > > > passed by value.
> > > > > >
> > > > > > e.g. I want to convert:
> > > > > >
> > > > > > void foobar(const int bar, const int *baz);
> > > > > > to
> > > > > > void foobar(int bar, const int *baz);
> > > > > >
> > > > > > I came up with this semantic patch:
> > > > > > /* Remove const from arguments passed by value */
> > > > > > @ rule1 @
> > > > > > identifier fi, I;
> > > > > > type T;
> > > > > > @@
> > > > > > fi ( ...,
> > > > > > - const T
> > > > > > + T
> > > > > > I
> > > > > > ,...
> > > > > > )
> > > > > > { ... }
> > > > > >
> > > > > > That seems to work fine on function declarations, but does not cover
> > > > > > function prototypes. Extending the patch to function prototypes I
> > > > > > ran
> > > > > > into some trouble. As complete separate patch, I evolved the above
> > > > > > to
> > > > > > (just replacing the function body with a ';'):
> > > > > > /* Remove const from arguments passed by value */
> > > > > > @ rule1 @
> > > > > > identifier fi, I;
> > > > > > type T;
> > > > > > @@
> > > > > > fi ( ...,
> > > > > > - const T
> > > > > > + T
> > > > > > I
> > > > > > ,...
> > > > > > ) ;
> > > > > >
> > > > > > But that is not parsed by spatch, causing this error:
> > > > > > init_defs_builtins: /usr/local/lib/coccinelle/standard.h
> > > > > > minus: parse error:
> > > > > > File "const_proto.cocci", line 9, column 1, charpos = 115
> > > > > > around = 'I',
> > > > > > whole content = ' I'
> > > > > >
> > > > > > Also, using the first version of a patch on a c-file with matching
> > > > > > function
> > > > > > declarations that also includes function prototypes, results in
> > > > > > both,
> > > > > > declarations and prototypes, while using it on a header with
> > > > > > prototypes
> > > > > > only doesn't result in any replacements.
> > > > > >
> > > > > > I attach the patches and some test files for reference.
> > > > > > const.cocci apparently correctly removes all const keywords from the
> > > > > > test.c file, but does not change the .h at all (which is correct,
> > > > > > just
> > > > > > weird that it changes the prototype in the .c).
> > > > > >
> > > > > > The const_proto.cocci file causes the above-mentioned error.
> > > > > >
> > > > > > Does anybody know what I'm missing here?
> > > > >
> > > > > By default, when you change a function header, Coccinell changes the
> > > > > prototype accordingly. But to be able to do that it needs to have
> > > > > access
> > > > > to the prototype. So you may need to give an argument like
> > > > > --all-includes
> > > > > (include all the .h files mentioned in the .c file) or
> > > > > --recursive-incudes
> > > > > (include all the .h files recurively included by any .h file
> > > > > mentioned in
> > > > > the .c file). --recursive-includes at least may be slow, but things
> > > > > will
> > > > > be done in a consistent manner.
> > > >
> > > > Interesting. I tried different variations of the include switches on the
> > > > real code base, but none resulted in the headers to be updated. Though,
> > > > with those switches I get 'different modification result for foo.h'
> > > > warnings.
> > > >
> > > > For the test files, including the header in the c file and running
> > > > spatch with --local-includes works as expected.
> > > >
> > > > Guess I'll dig into the headers and check if there are any unusual
> > > > constructs causing the problem. Any hints would be appreciated :)
> > > >
> > > > > I will check on why the expliciti prototype rule doesn't parse.
> > > >
> > > > Thanks for looking into it.
> > >
> > > Could you send an example of a .c and .h file where the result is not as
> > > expected?
> >
> > It works now... It seems something is handled significantly different
> > depending on how the target files are specified on the command line.
> >
> > Initially I ran:
> > spatch --sp-file const.cocci --local-includes --in-place --dir .
> > and now with your hints, I changed to
> > spatch --sp-file const.cocci --local-includes --in-place *.c
> >
> > That way headers are updated as expected and the warnings are gone.
> >
> > I can try to condense a small test case tomorrow, if you're
> > interested/necessary.
>
> --dir is the proper solution. Since it is not working, I would be
> interested in having a test case. I guess it should be a directory with
> multiple .c files.
>
> When you put multiple files on the command line, you are treating them all
> at once, ie they are all parsed and in memory at the same time. This is
> good if there are references between .c files that have to be taken into
> account. But obviously it doesn't scale. On the other hand, .h files
> should be treated at the same time as the .c files that reference them, so
> it should work.
I think I have a test case that is minimal.
I attach the following files:
- module_a.c -- c source
- module_a.h -- c header providing prototypes for functions in module_a.c
- module_b.c -- c source including module_a.h
- const.cocci -- semantic patch using fundecl syntax
- const_proto.cocci -- semantic patch using funproto syntax
The patch with funproto syntax fails as described above and is also
redundant when headers are included in the transformation the other
patch is doing.
When I run:
spatch --sp-file const.cocci --local-includes --in-place *.c
things work fine and both, module_a[ch], get updated as intended.
When I run
spatch --sp-file const.cocci --local-includes --in-place --dir .
then only the c files are updated (in-place). But, the patch output
includes the correct header transformation (I was expecting headers to
be updated in-place too, just as in the first case).
Also, removing module_b.c from the test, makes spatch update both
module.[ch] in place.
Thanks,
Sören
#include "module_a.h"
int function(const u32 arg0, u32* const arg1)
{ }
#ifndef MODULE_A_H_
#define MODULE_A_H_
int function(const u32 arg0, u32* const arg1);
#endif
#include "module_a.h"
/* Remove const from arguments passed by value */
@ rule1 @
identifier fi, I;
type T;
@@
fi ( ...,
- const T
+ T
I
,...
)
{ ... }
/* Remove const from arguments passed by value */
@ rule1 @
identifier fi, I;
type T;
@@
fi ( ...,
- const T
+ T
I
,...
) ;
_______________________________________________
Cocci mailing list
[email protected]
https://systeme.lip6.fr/mailman/listinfo/cocci
