[Linux-kernel-mentees] [PATCH v3 1/3] parsing_c: Align C AST and Cocci AST for pointer

Julia Lawall julia.lawall at inria.fr
Sun Feb 9 12:18:33 UTC 2020


On Sun, 9 Feb 2020, Jaskaran Singh wrote:

> For a pointer, the C parser constructed an AST dissimilar from that
> of the Cocci AST. This caused failures in matching with certain
> pointer types. For example, for the following case:
>
> char *1 const *2 id;
>
> The C AST constructed would be:
> const Pointer1 -> Pointer2 -> char
>
> The Cocci AST constructed would be:
> Pointer2 -> const Pointer1 -> char
>
> Change the pointer rule in the C parser so that an AST similar to the
> Cocci AST is constructed.
>
> Make necessary changes in the C pretty printer so that the pointer type
> is printed correctly.
>
> Signed-off-by: Jaskaran Singh <jaskaransingh7654321 at gmail.com>
> ---
>  parsing_c/parser_c.mly      |  4 ++--
>  parsing_c/pretty_print_c.ml | 12 ++++++++----
>  2 files changed, 10 insertions(+), 6 deletions(-)
>
> diff --git a/parsing_c/parser_c.mly b/parsing_c/parser_c.mly
> index 8d7b761e..4c74f15a 100644
> --- a/parsing_c/parser_c.mly
> +++ b/parsing_c/parser_c.mly
> @@ -1333,14 +1333,14 @@ pointer:
>   | tmul                   { (Ast_c.noattr,fun x -> mk_ty (Pointer x) [$1]) }
>   | tmul pointer
>       { let (attr,ptr) = $2 in
> -       (attr,fun x -> mk_ty (Pointer (ptr x)) [$1]) }
> +       (attr,fun x -> ptr (mk_ty (Pointer x) [$1])) }
>   | tmul type_qualif_list
>       { let (attr,tq) = $2 in
>         (attr,fun x -> (tq.qualifD, mk_tybis (Pointer x) [$1]))}
>   | tmul type_qualif_list pointer
>       { let (attr1,tq) = $2 in
>         let (attr2,ptr) = $3 in
> -       (attr1 at attr2,fun x -> (tq.qualifD, mk_tybis (Pointer (ptr x)) [$1])) }
> +       (attr1 at attr2,fun x -> ptr (tq.qualifD, mk_tybis (Pointer x) [$1])) }
>
>  tmul:
>     TMul { $1 }
> diff --git a/parsing_c/pretty_print_c.ml b/parsing_c/pretty_print_c.ml
> index a2e35588..ae02b513 100644
> --- a/parsing_c/pretty_print_c.ml
> +++ b/parsing_c/pretty_print_c.ml
> @@ -804,11 +804,13 @@ and pp_string_format (e,ii) =
>               (FunctionType (return=void, params=int i) *)
>            (*WRONG I THINK, use left & right function *)
>            (* bug: pp_type_with_ident_rest None t;      print_ident ident *)
> +          pp_type_left t;
>            pr_elem i;
> -          iiqu +> List.iter pr_elem; (* le const est forcement apres le '*' *)
> +          iiqu +> List.iter (function x ->
> +             (pr_space(); pr_elem x));(* le const est forcement apres le '*' *)

I'm not very fond of this coding style.  I would prefer:

iiqu +>
List.iter (* the comment can go here *)
  (function x -> pr_space(); pr_elem x)

Likewise below.

julia

>            if iiqu <> [] || get_comments_after i <> []
>            then pr_space();
> -          pp_type_with_ident_rest ident t attrs Ast_c.noattr;
> +          print_ident ident
>
>        (* ugly special case ... todo? maybe sufficient in practice *)
>        | (ParenType ttop, [i1;i2]) ->
> @@ -885,11 +887,13 @@ and pp_string_format (e,ii) =
>        match ty, iity with
>  	(NoType,_) -> failwith "pp_type_left: unexpected NoType"
>        | (Pointer t, [i]) ->
> +          pp_type_left t;
>            pr_elem i;
> -          iiqu +> List.iter pr_elem; (* le const est forcement apres le '*' *)
> +          iiqu +> List.iter (function x ->
> +             (pr_space(); pr_elem x));(* le const est forcement apres le '*' *)
>            if iiqu <> [] || get_comments_after i <> []
>            then pr_space();
> -          pp_type_left t
> +          ()
>
>        | (Array (eopt, t), [i1;i2]) -> pp_type_left t
>        | (FunctionType (returnt, paramst), [i1;i2]) -> pp_type_left returnt
> --
> 2.21.1
>
>


More information about the Linux-kernel-mentees mailing list