Messages in this thread | | | Date | Fri, 2 Jan 2015 12:46:15 +0100 (CET) | From | Julia Lawall <> | Subject | Re: [PATCH 0/3 RFC] Coccinelle: completion API checking |
| |
On Fri, 2 Jan 2015, Nicholas Mc Guire wrote:
> > This little set of semantic patches is for partially checking the > completion API. It seems to be working correctly and has not yet > produced any false-positive. > > The cases being detected are: > > 1/3 - duplicate init_completions. > 2/3 - incorrect static initialization of completion on stack. > 3/3 - re-initialization of completion with init_completion() rather > than reinit_completion(). > > semantic patch findings files confirmed > duplicate_init_completion.cocci 2 2 2 > false_declare_completion.cocci 6 5 3 > false_init_compltion.cocci 9 6 5 > > Note: false_declare_completion.cocci runs extremely slow on my system > roughly a factor 10 slower than the other two - not clear why.
The problem is the use of <... ...>. That is good because it allows DECLARE_COMPLETION not to appear on some execution paths, but it is not so good because it allows DECLARE_COMPLETION to appear on no execution paths at all. So the semantic patch is considered to be relevant to every file, and every file has to be parsed, which takes a lot of time.
To fix the problem, you can first check whether the call actually exists in each function:
@pre_e exists@ identifier f; declarer name DECLARE_COMPLETION; @@
f(...) { <+... DECLARE_COMPLETION(...); ...+> }
Then only run the second rules on functions for which pre_e is satisfied. That is, in the rule e, instead of putting identifier f, you can put identifier pre_e.f. You can do the same in the rule ep. Ep is already looking for the existence of an execution path, so for ep you could also just replace the <... ...> by <+... ...+> instead of relying on pre_e.
julia
| |