[Cocci] please help me with a failing match

[Robert.Larice@t-online.de (Robert Larice)]

Julia Lawall <julia.lawall@lip6.fr> writes:

> On Sat, 17 Feb 2018, Robert Larice wrote:
>
>> Julia Lawall <julia.lawall@lip6.fr> writes:
>>
>> > On Sat, 10 Feb 2018, Robert Larice wrote:
>> >
>> >> Julia Lawall <julia.lawall@lip6.fr> writes:
>> >>
>> >> > On Sat, 10 Feb 2018, Robert Larice wrote:
>> >> >
>> >> >> Dear People,
>> >> >>
>> >> >>   I'm completely new here.
>> >> >>
>> >> >>   Attached is a small piece of .c and a .cocci file.
>> >> >>   There is a "return 41;" in both files, commented out.
>> >> >>   If I uncomment this "return 41;" in both files then
>> >> >>     spatch will not match the pieces any more.
>> >> >>
>> >> >>   Could you please help me to undertand and circumvent this issue ?
>> >> >
>> >> > I have not noticed this problem before, but I suspect that it is due to
>> >> > the fact that Coccinelle is matching the control-flow path and not the
>> >> > abstract syntax tree. In a control-flow graph, nothing follows a return.
>> >> >
>> >> > julia
>> >>
>> >> Thank You,
>> >> I tried to sneak around the problem with a second "rule" which
>> >>   translates "return 42" to "auxiliary(42)".
>> >> My intention was to first change the source in such a way
>> >>   that the "control-flow" graph does not end at the "return",
>> >> and then hope that the second (accordingly modified) rule would
>> >>   match.
>> >> This didn't work, I assume I would have to express the idea of
>> >>   first applying the first rule
>> >>     then to rebuild the control-flow graph
>> >>   then try the second rule.
>> >>   (and finally undo the changes of the first rule in a third rule)
>> >> I can not force "rebuild" without invoking spatch myself a second time.
>> >
>> > If you change all the returns to something else and then match your
>> > pattern, and then change them it should work.
>> >
>> >>
>> >> ---
>> >> I'm a bit of a maintainer for the "ngspice" project, which has a vast
>> >>   amount of very old files, and lots of semi duplicated stuff often crying
>> >>   for a thourough hair wash,
>> >> stumbled over this intresting tool, and am tying it for a certain
>> >>   rewrite I'm currently busy with.
>> >
>> > OK,feel free to ask more questions if you run into further issues.
>> >
>> > julia
>>
>> Hello,
>>
>>   Thank you for your help. I've four or five .cocci files now which
>>     do the job at hand for ngspice quite well. The rewrite of return
>>     to something else with two times invocation of spatch did work.
>>   Then I found another simpler way. So thats done.
>>
>>   Playing around, trying to better understand what it means
>>     to have more than one rule, their interaction etc ..
>>     I came to the attached example.
>>   Here I have basically tried to remove the whole body of a function
>>     iff the function matches two other @rules@
>>   It seems to work if I use the '*' notation, but doesnt if I use
>>     '+/-'
>>   Can you give me a hint which helps me to understand this ?
>
> For -+, Coccinelle requires that all control-flow paths from the starting
> point of the match match the complete rule. So for example a() ... b()
> would match the following:
>
> a();
> if (x)
>   b();
> else b();
>
> For *, it only requires the existence of such a path.
>
> If you have a -+ somewhere in the semantic patch, then all rules use the
> forall semantics.  If you have a * somewhere in the semantic patch, then
> all rules use the exists semantics.
>
> In your rule r2, you have ... Label: but this label is not reached on the
> execution path that ends in return 1.  When you have -+ somewhere in the
> semantic patch this rule is not satisfied.
>
> If you want to change the default for a rule, you can add exists for
> forall to the rule header, or put when exists or when forall on the ...
>
> julia

Thank You very much,

The behaviour, and what I have to do to fix it, is perfectly clear now,

Regards,
Robert