r198461 - Debug info: Ensure that the last stop point in a function is still within

[dblaikie at gmail.com]

On Mon Jan 06 2014 at 8:29:19 AM, Adrian Prantl <aprantl at apple.com> wrote:

>
> On Jan 5, 2014, at 8:22 PM, dblaikie at gmail.com wrote:
>
> >
> >
> > On Fri Jan 03 2014 at 4:00:30 PM, Adrian Prantl <aprantl at apple.com>
> wrote:
> >
> > On Jan 3, 2014, at 15:53, Eric Christopher <echristo at gmail.com> wrote:
> >
> >> There's not enough information in the commit message, the comments, or
> the test cases to know what's broken or fixed here. Why is forcing the
> block the correct solution here? What is it working around?
> >>
> >>
> >
> > Let’s use this as an example:
> >
> >> ============================================================
> ==================
> >> --- cfe/trunk/test/CodeGenObjC/arc-linetable.m (original)
> >> +++ cfe/trunk/test/CodeGenObjC/arc-linetable.m Fri Jan  3 17:34:30 2014
> >
> > - (int)testNoSideEffect:(NSString *)foo {
> > int x = 1;
> > return 1; // Return expression
> > }
> >
> > Before this fix, a user would set a breakpoint at the return expression
> and then attempt to print x and it would fail, because the instruction that
> is at that line would be in the DW_TAG_subprogram lexical scope (instead of
> the lexical scope of the function body compound statement which contains x).
> >
> > I'm confused by this - Clang doesn't emit a separate lexical scope for
> the top level compound statement of a function...
> >
>
> At least for Objective C this is not true. From CGObjcC.cpp:
>
> /// Generate an Objective-C method.  An Objective-C method is a C function
> with
> /// its pointer, name, and types registered in the class struture.
> void CodeGenFunction::GenerateObjCMethod(const ObjCMethodDecl *OMD) {
> StartObjCMethod(OMD, OMD->getClassInterface(), OMD->getLocStart());
> EmitStmt(OMD->getBody());
> FinishFunction(OMD->getBodyRBrace());
> }
>

OK - would the issue be fixed if we didn't emit the extra lexical block?
(compare this to GCC's behavior - I think at least in C++ they don't emit
the extra top level lexical block. They might do it in C, but I'm not sure
why (there may be a reason)). I think I changed Clang to not do this for
C++, hopefully for C too - but may've missed this particular ObjC case.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140106/35e10ec7/attachment.html>