On Tue, Apr 28, 2020 at 08:41:08PM +0200, Julia Lawall wrote:
> 
> 
> On Tue, 28 Apr 2020, Paul Chaignon wrote:
> 
> > Thanks for the quick answer!
> >
> > On Tue, Apr 28, 2020 at 07:44:15PM +0200, Julia Lawall wrote:
> > >
> > >
> > > On Tue, 28 Apr 2020, Paul Chaignon wrote:
> > >
> > > > Hi all,
> > > >
> > > > I am working on a small semantic patch to annotate specific variables in
> > > > our codebase with __attribute__((aligned(8))). The following program 
> > > > works
> > > > fine.
> > > >
> > > >   @r@
> > > >   expression e1, e2;
> > > >   identifier x;
> > > >   @@
> > > >   (
> > > >     struct \(icmphdr\|icmp6hdr\) x
> > > >   + __attribute__((aligned(8)))
> > > >     ;
> > > >   |
> > > >     struct \(icmphdr\|icmp6hdr\) x
> > > >   + __attribute__((aligned(8)))
> > > >     = ...;
> > > >   )
> > > >     ... when exists
> > > >     ctx_load_bytes(e1, e2, &x, ...)
> > > >
> > > > However, when I replace __attribute__((aligned(8))) with our internal
> > > > macro __align_stack_8, it fails with the following error:
> > > >
> > > >   plus: parse error:
> > > >     File "/home/paul/cilium/contrib/coccinelle/aligned.cocci", line 7, 
> > > > column 2, charpos = 77
> > > >     around = '__align_stack_8',
> > > >     whole content = + __align_stack_8
> > > >
> > > > I've tried adding '#define __align_stack_8' in a file passed with
> > > > --macro-file, without success. Is this a known limitation for macros or
> > > > am I missing something?
> > >
> > > Try adding the "metavariable" declaration:
> > >
> > > attribute name __align_stack_8;
> >
> > Awesome, that worked. And I think I understand: undeclared identifiers are 
> > by
> > default considered symbols, leading to the parse error.
> >
> > Unfortunately, my semantic patch now leads to the following changes:
> >
> >   - struct icmphdr icmphdr __align_stack_8;
> >   + struct icmphdr icmphdr __align_stack_8 __align_stack_8;
> >
> > I would normally add a first case to my conjunction to match on
> > already-present attributes, but Coccinelle can't match on attributes yet.
> > Any workaround?
> 
> Jaskaran is working on it.

Great! I'll keep a watch on new releases :-)

> A hackish solution would be to put a position variable on the variable
> name and put a position variable on the ; and then use python to see if
> they are not adjacent to each other...
> 
> Do you get a lot of occurrences of the problem?  In the short term it
> could be simpler to just clean it up by hand.  It should be easy to search
> for at least.

We don't have a lot of variables that need this annotation overall, but
I'm hoping to run the semantic patch as part of our CI pipeline.

In the medium term, I think I can just keep my current version (as above
but extended with positions and a Python error message). Because it
doesn't try to add missing attributes, it doesn't need the attribute name
and doesn't have the double annotation problem.

Thanks a lot for your help!

Cheers,
Paul

> 
> julia
> 
> 
> >
> > Paul
> >
_______________________________________________
Cocci mailing list
[email protected]
https://systeme.lip6.fr/mailman/listinfo/cocci

Reply via email to