r183597 - Debug info: An if condition now creates a lexical scope of its own.

[Robinson, Paul]

> From: cfe-commits-bounces at cs.uiuc.edu [mailto:cfe-commits-
> bounces at cs.uiuc.edu] On Behalf Of Eric Christopher
> Sent: Monday, June 10, 2013 4:05 PM
> To: Adrian Prantl
> Cc: Nadav Rotem; cfe-commits at cs.uiuc.edu
> Subject: Re: r183597 - Debug info: An if condition now creates a lexical
> scope of its own.
> 
> On Mon, Jun 10, 2013 at 3:56 PM, Adrian Prantl <aprantl at apple.com>
> wrote:
> > (CC'ing John because he understands the intricacies of LexicalScope
> better than I do)
> >
> > On the first glimpse LexicalScope appears to be a subclass of
> RunCleanupsScope that additionally emits a (DebugInfo-)LexicalScope. But
> looking at the destructors it appears that they have slightly different
> semantics: ~LexicalScope runs ForceCleanup and ~RunCleanupsScope
> apparently doesn't.
> >
> > I'm wary that switching to Lexicalscope in
> CodeGenFunction::EmitIfStmt() might lead to tricky ARC or EH-related
> problems because of that.
> >
> > Does anyone have an opinion on that?
> 
> That bit was added here:
> 
> commit 495cfa46300979642acde8d93a1f21c9291dac98
> Author: Nadav Rotem <nrotem at apple.com>
> Date:   Sat Mar 23 06:43:35 2013 +0000
> 
>     Make clang to mark static stack allocations with lifetime markers
> to enable a more aggressive stack coloring.
>     Patch by John McCall with help by Shuxin Yang.
>     rdar://13115369
> 
> 
> and oddly not to RunCleanupsScope.
> 
> Nadav?
> 
> -eric

Looks like ~LexicalScope() just wants RunCleanupsScope::ForceCleanup()
to happen before rescopeLabels().  All the right stuff happens in the right
order if you change the RunCleanupsScope instance to LexicalScope.
(RunCleanupsScope::ForceCleanup() does pretty much exactly the same thing
as ~RunCleanupsScope() so it works out.  It could be done in a more obvious
way, but functionally they're equivalent.)

And as long as the DWARF lexical scope tag is omitted if there aren't
any actual declarations, I have no debug-info-size concerns.
--paulr