[cfe-commits] [PATCH] [CFG] Put implicit destructors in separate block after ReturnStmt

[Ted Kremenek]

On Feb 6, 2012, at 1:09 AM, Erik Verbruggen <erik.verbruggen at me.com> wrote:

> On 31-1-12 20:20, Ted Kremenek wrote:
>> Hi Erik,
>> 
>> I know that we have been discussing changing the CFG around its handling of ReturnStmts, but it isn't clear to me what this accomplishes.  We aren't looking to enforce that ReturnStmt is the last *statement* in a CFGBlock.  The ReturnStmt in a CFGBlock currently represents binding the return value, not a control-flow transfer.  If we want to model the ReturnStmt as a control-flow transfer (which is subject to debate), we should consider also making it a terminator (which is preceded by all the cleanup destructors).  I'm also not certain what you mean by "this results in blocks ending in with a terminator statement."
>> 
>> I don't want to put the brakes on this, but if the goal is to fix how destructors are represented in the CFG, I'd rather we *first* do the upfront infrastructure work discussed in PR 11645, and then re-evaluate, as a whole, how destructors are currently represented in the CFG.  This all feels a bit piecemeal, which would be fine if we were fixing specific issues, but the general sense I have right now is that how destructors are currently represented in the CFG is possibly very different from where we need to go once the infrastructure work in PR 11645 is done.  It's also not clear to me what invariants this patch is trying to establish or change.
> 
> You are right of course. The patch was in my outgoing-patch-queue for some time, and I send it without too much thought. My fault.

No worries.  I think we're at the point where we are re-evaluating and re-designing some key parts of the CFG, so I think it is fine bringing up ReturnStmt in those discussions, but it should be part of a bigger picture of where we want to take the CFG.