[PATCH] [analyzer] Fix ObjC Dealloc Checker to operate only on classes with retained properties

[David Kilzer]

Thanks for the comments, Anna!  I will post a new patch soon.

================
Comment at: lib/StaticAnalyzer/Checkers/CheckObjCDealloc.cpp:108
@@ -82,3 +107,3 @@
 
   assert (LOpts.getGC() != LangOptions::GCOnly);
 
----------------
None of these checks is useful for ARC (-fobjc-arc), so I need to assert this is never called in ARC mode (which Jordan alluded to in a previous comment).


================
Comment at: lib/StaticAnalyzer/Checkers/CheckObjCDealloc.cpp:206
@@ -194,3 +205,3 @@
     bool requiresRelease = PD->getSetterKind() != ObjCPropertyDecl::Assign;
     if (scan_ivar_release(MD->getBody(), ID, PD, RS, SelfII, Ctx)
        != requiresRelease) {
----------------
zaks.anna wrote:
> Would it be possible to move the checking of the setter (to be not 'assign') into the body of isSynthesizedWritablePointerProperty() ?
I can, but it makes it harder to share code with the patch in D5042, since I need that bit of information separately.



================
Comment at: lib/StaticAnalyzer/Checkers/CheckObjCDealloc.cpp:255-256
@@ -237,4 +254,4 @@
                     BugReporter &BR) const {
     if (mgr.getLangOpts().getGC() == LangOptions::GCOnly)
       return;
     checkObjCDealloc(this, cast<ObjCImplementationDecl>(D), mgr.getLangOpts(),
----------------
None of these checks is useful for ARC (-fobjc-arc), so I need to return early for ARC mode as well (which Jordan alluded to in a previous comment).

http://reviews.llvm.org/D5023