From: skunk@iSKUNK.ORG (Daniel Richard G.) Date: Wed, 24 Jun 2015 23:10:05 -0400 Subject: [Cocci] Various minor parsing bugs In-Reply-To: References: <1434594839.2823366.298546905.56C27EC5@webmail.messagingengine.com> <1435098813.2577840.305986601.7F5C26BA@webmail.messagingengine.com> Message-ID: <1435201805.2992101.307114657.0D123073@webmail.messagingengine.com> To: cocci@systeme.lip6.fr List-Id: cocci@systeme.lip6.fr On Wed, 2015 Jun 24 07:40+0200, Julia Lawall wrote: > > OK, perhaps I was not quite correct. Overall, it seems to be related to > the goal of matching disjunctions on expressions from left to right. For > example, if you have a pattern like: > > ( > - &x > + y > | > - x > + z > ) > > and the code you have is &x, then you expect to get y. Maybe it is > throwing away matches with later terms that overlap with matches for > earlier ones. I would have thought that only smaller matches would > get thrown away, but whenever things overlap there is a danger of > inconsistent modifications, so maybe any overlap at all is the better > strategy. Hmm. I think I may need to find a better example with which to examine this behavior, one where the disjunction turns out to be necessary after all. (I'm a bit uneasy with not knowing how overlaps interact with disjunctions, and I suspect the issue will crop up again in a different form.) > > Coccinelle already does some preprocessor handling; there isn't a > > way to tell it to look at only one side of a particular conditional? > > The example I gave is a trivial one---the expression could be > > written out twice---but there are others less amenable to such a > > rewrite. > > There are some special cases where it ignores one side, but in general > it doesn't. I think that to make this decision in general you have to > parse, and the ifdefs block parsing. Could some ifdefs be selectively evaluated/ignored for parsing purposes? > Mostly Coccinelle has heuristics for identifying types. I'll have to > check again why they were not triggered in your case. If it helps, here are the minimal cases I put together where the C99 type flummoxed Coccinelle (when not mentioned in a macro file): #include #include size_t foo1(void *ptr) { return bar((int32_t)(size_t)ptr); } void foo2(char *array) { array[(int32_t)'.'] = '.'; } On Wed, 2015 Jun 24 09:36+0200, Julia Lawall wrote: > For #ifdef, are the terms in the two branches proper expressions > normally (like in your example)? Perhaps we could do something for > that. Or are there things like > > #ifdef XXX > 3 + > #else > 4 - > #endif > 5 Just about, yes. There are idioms like int #ifdef WIN32 WINAPI WinMain([...]) #else main(int argc, char *argv[]) #endif { and others where the conditional/surrounding parts can't stand on their own. The wisdom of writing code in this way may be questionable, but I would be happy for Coccinelle to evaluate the conditional itself, and process only one side. (In my case, the WIN32 side is irrelevant.) --Daniel -- Daniel Richard G. || skunk at iSKUNK.ORG My ASCII-art .sig got a bad case of Times New Roman.