On Fri, Feb 25 2022, Johannes Schindelin wrote:
> Hi Julia,
>
> On Tue, 22 Feb 2022, Julia Lawall wrote:
>
>> [I]f there are some cases that are useful to do statically, with only
>> local information, then using Coccinelle could be useful to get the
>> problem out of the way once and for all. Coccinelle doesn't support
>> much processing of strings directly, but you can always write some
>> python code to test the contents of a string and to create a new one.
>>
>> Let me know if you want to try this. You can also check, eg the demo
>> demos/pythontococci.cocci to see how to create code in a python script and
>> then use it in a normal SmPL rule.
>>
>> If some context has to be taken into account and the context in the same
>> function, then that can also be done with Coccinelle, eg
>>
>> A
>> ...
>> B
>>
>> matches the case where after an A there is a B on all execution paths
>> (except perhaps those that end in an error exit) and
>>
>> A
>> ... when exists
>> B
>>
>> matches the case where there is a B sometime after executing A, even if
>> that does not always occur.
>>
>> If the context that you are interested in is in a called function or is in
>> the calling context, then Coccinelle might not be the ideal choice.
>> Coccinelle works on one function at a time, so to do anything
>> interprocedural, you have to do some hacks.
>
> Right. The code in question is not actually calling a function, but a
> macro, and passes a literal string to the macro that we would want to
> check statically.
>
> I did have my doubts that it would be easy with Coccinelle, but since Ævar
> seemed so confident, I tried it, struggled, and decided to follow up with
> you.
>
> Thank you for confirming my suspicion!
> Johannes
In case it's not clear from the upthread (and I thought my [1] explained
it well enough) I never thought it would be easy or even possible to do
this particular thing with coccinelle.
I.e. I mentioned in [1]:
Aside: if we did want to do the "parse C" method the right way to do it
would be to have a coccinelle script do it
So it's intended as a side-note to explain to a new contributor that
*if* we do end up wanting to parse or transform C we have coccinelle,
and it's a great tool for those cases where such static transformations
are easy. I and others have added some in-tree in the past where
appropriate.
But I then went on to say (and elaborated on later in [2]) that in this
case the right thing to do is runtime checking.
So, I'm sorry if you wasted time on it. In either case it seems we've
ended up in agreement in this case about appropriate uses of coccinelle.
I.e. it's a fantastic tool as a semantic patch engine, but it
understandably has limitations where you'd effectively need it to
execute your program to decide what to do, as is the case with the
parse_options() API and the eventual parse_options_check() etc. doing
assertions depending on flags that got passed down.
1. https://lore.kernel.org/git/220221.86tucsb4oy.gmgdl@xxxxxxxxxxxxxxxxxxx/
2. https://lore.kernel.org/git/220222.867d9n83ir.gmgdl@xxxxxxxxxxxxxxxxxxx/
