Makes sense. Applied:
http://llvm.org/viewvc/llvm-project?view=rev&revision=113799 On Sep 13, 2010, at 2:38 PM, Marcin Świderski wrote: > W dniu 13 września 2010 23:17 użytkownik Ted Kremenek <[email protected]> > napisał: > Hi Marcin, > > If you think you will need to be using insert(), I'm dubious that a vector is > the right data structure. Doing inserts repeatedly might wreck havoc on CFG > construction time. What do you think? > > On Sep 13, 2010, at 12:51 PM, Marcin Świderski wrote: > > > Hi > > > > This is an implementation of insert method for BumpVector that is needed in > > my work on CFG. > > > > Cheers, > > Marcin > > <bump-vector-insert.patch> > > I use insert for adding CFGElements for implicit destructors. The procedure > is as follows: > - count objects for destruction (time linear w.r.t. depth of local scopes), > - make room for those objects with batch insert of empty CFGElements, > - set proper values of destructors for inserted CFGElements. > > This will cause data moving inside the vector only while back patching for > goto statements. In other cases this will only append data at the end of > vector, and can be optimized if will cause too much overhead. Currently I > would prefere to leave it this way.
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
