[Cocci] transforming arguments to statement macros?

[Cocci] transforming arguments to statement macros?

[David Young]

I have a semantic patch that renames parameters and local variables
called `index` to `idx`.  It renames most occurrences, however,
it does not know how to handle a macro that expands to a block:

#define HGOTO_ERROR(maj, min, ret_val, ...) {                           \
   do_something(maj, min, __VA_ARGS__);                                 \
   ret_value = ret_val;							\
   goto done;								\
}

I'd like for every occurrence of `index` in the HGOTO_ERROR() arguments
to change to `idx`, HGOTO_ERROR(..., index, ...) -> HGOTO_ERROR(..., idx, ...),
but spatch leaves those occurrences alone.

Can I write an isomorphism or something to force spatch to process each
occurrence of HGOTO_ERROR(...) as if it was either the function call
`hgoto_error(...);` or the block `{ (void)(...); goto done; }` ?


Dave

-- 
David Young
dyoung@pobox.com    Urbana, IL    (217) 721-9981
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[Julia Lawall]

--- Please note the new email address ---


On Thu, 2 Jan 2020, David Young wrote:

> I have a semantic patch that renames parameters and local variables
> called `index` to `idx`.  It renames most occurrences, however,
> it does not know how to handle a macro that expands to a block:
>
> #define HGOTO_ERROR(maj, min, ret_val, ...) {                           \
>    do_something(maj, min, __VA_ARGS__);                                 \
>    ret_value = ret_val;							\
>    goto done;								\
> }
>
> I'd like for every occurrence of `index` in the HGOTO_ERROR() arguments
> to change to `idx`, HGOTO_ERROR(..., index, ...) -> HGOTO_ERROR(..., idx, ...),
> but spatch leaves those occurrences alone.
>
> Can I write an isomorphism or something to force spatch to process each
> occurrence of HGOTO_ERROR(...) as if it was either the function call
> `hgoto_error(...);` or the block `{ (void)(...); goto done; }` ?

I think that the problem is that there is no ; in the uses of your macro.
So Coccinelle knows that it is some strange code, and it refuses to change
it.  If you run spatch --parse-c on a file that uses the macro, you will
see something able the code being passed.

The proper way to write such a macro, independent of Coccinelle, is as a
while do(0) loop, so that the uses can end in a semicolon.  Then there is
no possibility of strange mistakes if someone actually does put a
semicolon.  Would that be feasible to do?

julia
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[David Young]

On Fri, Jan 03, 2020 at 07:32:09AM +0100, Julia Lawall wrote:
> I think that the problem is that there is no ; in the uses of your macro.

I was afraid of that.

> The proper way to write such a macro, independent of Coccinelle, is as a
> while do(0) loop, so that the uses can end in a semicolon.  Then there is
> no possibility of strange mistakes if someone actually does put a
> semicolon.  Would that be feasible to do?

I would prefer that the macros were written with the do-while pattern,
but it's a legacy codebase that uses HGOTO_ERROR() no fewer than 12,000
times, and most occurrences have no semicolon.  I will have to automate
the conversion, and I guess that I cannot use spatch to do it. :-) I may
be able to write a suitable vim macro.

Thank you for your help.

Dave

-- 
David Young
dyoung@pobox.com    Urbana, IL    (217) 721-9981
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[Markus Elfring]

> I would prefer that the macros were written with the do-while pattern,
> but it's a legacy codebase that uses HGOTO_ERROR() no fewer than 12,000
> times, and most occurrences have no semicolon.  I will have to automate
> the conversion, and I guess that I cannot use spatch to do it.

Would you like to try out how good semicolons can be added behind
these macro calls by a small script for the semantic patch language?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[Julia Lawall]

On Sun, 5 Jan 2020, Markus Elfring wrote:

> > I would prefer that the macros were written with the do-while pattern,
> > but it's a legacy codebase that uses HGOTO_ERROR() no fewer than 12,000
> > times, and most occurrences have no semicolon.  I will have to automate
> > the conversion, and I guess that I cannot use spatch to do it.
>
> Would you like to try out how good semicolons can be added behind
> these macro calls by a small script for the semantic patch language?

The problem is that the arguemnt list is considered to be comments.  THere
is no way to specify to put a semicolon at the end of a comment.

The only short term suggestion I have is the following:

@r@
statement s : script:ocaml() { List.hd (Str.split (Str.regexp " *(") s) =
"ASSERT" };
statement s1,s2;
position p1;
@@

s
s1@p1

@@
statement r.s1;
position r.p1;
@@

+;
s1@p1

-------------------

This uses ASSERT as an example, because standard.h already contains

#define ASSERT(x) MACROSTATEMENT

It uses a regular expression to detect that a statement starts with the
word ASSERT, and matches the next statement as well, and then in another
rule adds a ; before the subsequent statement.  If you add the ; after the
ASSERT statement it comes out after ASSERT, not after the commented
argument list. But with this semantic patch, the ; comes out on a line of
its own, and would have to be moved up manually.  So I don't know if that
gives any real benefit.  It obviously also doesn't work if the ASSERT is
at the end of a function.

julia
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[Markus Elfring]

> The problem is that the arguemnt list is considered to be comments.  THere
> is no way to specify to put a semicolon at the end of a comment.

This information needs further clarification.

Are there any other challenges to reconsider for the addition of a semicolon
after a closing parenthesis for the relevant macro call?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[Markus Elfring]

> It uses a regular expression to detect that a statement starts with the
> word ASSERT, and matches the next statement as well, and then in another
> rule adds a ; before the subsequent statement.

Can related improvements for the supported programming interfaces help
to split the available information into more useful pieces?


> If you add the ; after the ASSERT statement it comes out after ASSERT,
> not after the commented argument list. But with this semantic patch,
> the ; comes out on a line of its own, and would have to be moved up manually.

Can this software behaviour be avoided if the semicolon would be inserted
by using the function “make_stmt” of Coccinelle's library in
a SmPL script rule?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] Adding semicolons after macro calls

[Markus Elfring]

> The only short term suggestion I have is the following:

I wonder about the specification of two statement variables here.


Under which circumstances can a transformation approach (like the following)
work as desired?

@addition@
identifier action;
statement s;
@@
 <+...
 action(...)
+;
 s
 ...+>

elfring@Sonne:~/Projekte/Coccinelle/Probe> spatch --parse-cocci add_semicolon_behind_call1.cocci
init_defs_builtins: /usr/local/bin/../lib/coccinelle/standard.h
minus: parse error:
  File "add_semicolon_behind_call1.cocci", line 8, column 1, charpos = 70
  around = 's',
  whole content =  s


How will further improvements be achieved for data processing around macro calls?

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] Adding semicolons after macro calls

[Julia Lawall]

On Mon, 6 Jan 2020, Markus Elfring wrote:

> > The only short term suggestion I have is the following:
>
> I wonder about the specification of two statement variables here.
>
>
> Under which circumstances can a transformation approach (like the following)
> work as desired?
>
> @addition@
> identifier action;
> statement s;
> @@
>  <+...
>  action(...)
> +;
>  s
>  ...+>

Currently none.  The SmPL language doesn't support statements that look
like function calls but have no following ;

julia

>
> elfring@Sonne:~/Projekte/Coccinelle/Probe> spatch --parse-cocci add_semicolon_behind_call1.cocci
> init_defs_builtins: /usr/local/bin/../lib/coccinelle/standard.h
> minus: parse error:
>   File "add_semicolon_behind_call1.cocci", line 8, column 1, charpos = 70
>   around = 's',
>   whole content =  s
>
>
> How will further improvements be achieved for data processing around macro calls?
>
> Regards,
> Markus
>
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] Improving support for data processing around macro calls

[Markus Elfring]

> The SmPL language doesn't support statements that look like function calls
> but have no following ;

A missing semicolon can trigger the source code interpretation in different ways.

* Can macro calls be treated similar to attributes?

* How challenging will it become to change the mentioned software limitation?
  https://github.com/coccinelle/coccinelle/issues/140

Regards,
Markus
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

Re: [Cocci] transforming arguments to statement macros?

[Julia Lawall]

On Fri, 3 Jan 2020, David Young wrote:

> On Fri, Jan 03, 2020 at 07:32:09AM +0100, Julia Lawall wrote:
> > I think that the problem is that there is no ; in the uses of your macro.
>
> I was afraid of that.
>
> > The proper way to write such a macro, independent of Coccinelle, is as a
> > while do(0) loop, so that the uses can end in a semicolon.  Then there is
> > no possibility of strange mistakes if someone actually does put a
> > semicolon.  Would that be feasible to do?
>
> I would prefer that the macros were written with the do-while pattern,
> but it's a legacy codebase that uses HGOTO_ERROR() no fewer than 12,000
> times, and most occurrences have no semicolon.  I will have to automate
> the conversion, and I guess that I cannot use spatch to do it. :-) I may
> be able to write a suitable vim macro.

You should now be able to update the arguments, despite the lack of a
semicolon.

julia
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci