From: soren.brinkmann@xilinx.com (=?utf-8?B?U8O2cmVu?= Brinkmann) Date: Tue, 10 May 2016 22:55:15 -0700 Subject: [Cocci] remove redundant const qualifiers from function arguments In-Reply-To: References: <20160511031419.GY3766@xsjsorenbubuntu> Message-ID: <20160511055515.GA3766@xsjsorenbubuntu> To: cocci@systeme.lip6.fr List-Id: cocci@systeme.lip6.fr Hi Julia, On Wed, 2016-05-11 at 07:41:20 +0200, Julia Lawall wrote: > > > On Tue, 10 May 2016, S?ren 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. S?ren