[cfe-commits] r124279 - in /cfe/trunk: include/clang/Basic/DiagnosticSemaKinds.td test/Preprocessor/pragma_diagnostic_sections.cpp test/Sema/uninit-variables.c test/SemaCXX/uninit-variables.cpp

[Ted Kremenek]

Hi Chandler,

My focus this week was trying to address the clear false positives that Clang's -Wunitialized should be able to handle (and other user feedback) prior to posting on cfe-dev.  You are right that we should go ahead and have this discussion on cfe-dev now.

I'm less concerned than you are about providing different flags than you are (and have somewhat negative opinion about doing so), but I'm more immediately concerned with discussing expectations of how the compiler's "uninitialized variables analysis" should behave (regardless of the flags).

Later today I'll go ahead and write an introductory message for a cfe-dev thread.  I'll try and summarize most of the issues I've seen from user feedback.

Sent from my iPad

On Feb 3, 2011, at 12:46 AM, Chandler Carruth <chandlerc at google.com> wrote:

> (FYI, I've been waiting to follow up here hoping for a cfe-dev targeted thread. Should I just start one?)
> 
> I think we're actually talking a bit past each other, especially given Chris's comment. My expectations are *not* those of GCC's flag. I really dislike that flag's behavior for many of the reasons you've both outlined.
> 
> I like the ability to warn on potentially uninitialized variables, but that has proven to be very hard to turn on for a large code base. I'd rather see two flags, one targeted just as you've described at these potential errors with reasonable tradeoffs on false positives, and a separate flag that only flags definitively uninitialized variables. We've asked GCC to split their warnings in this way as well. I believe the names they're going with are "-Wmaybe-uninitialized" and "-Wuninitialized" resp., but I don't care that much about the names used.
> 
> Anyways, I'd can write this proposal up more fully on cfe-dev, including examples, motivations from our experience with GCC's -Wuninitialized, or whatever would help folks weigh the various options.
> 
> To answer a couple of other specific questions...
> 
> On Mon, Jan 31, 2011 at 8:39 AM, Ted Kremenek <kremenek at apple.com> wrote:
> I am completely open to changing the flag, but as Nico pointed out the point of turning it on this way was to find these regressions and to measure people's expectations.  This is TOT, not a shipping version of Clang.
> 
> Just so it's clear, while this is TOT that's what we cut our releases from and we'll have to work around this internally for now. Not a big deal, but we try to minimize how often we have to do that to ensure we keep upstreaming everything.
> 
> We try to release roughly bi-weekly from TOT, and are moving to weekly as soon as we can. I'm also hoping we can turn around and offer to qualify "beta" or other bleeding-edge-named releases on a similar time table. The automation for this should be wrapped up in the next couple of months.
>  
> We frequently turn on warnings and features, see how they work, and reevaluate.  Only a limited set of people would have provided feedback on the warning if I had not done this.
> 
> You're also welcome to send a note to cfe-dev asking for feedback on experimental flags, and I'll try to promptly get some detailed feedback. Maybe I've not made it clear in the past, but we have a great setup for evaluating flag changes across a large body of both open source and closed source code very very quickly. I would love to help anyone in the community take advantage of it however we can. Feel free to directly ping if we can help in this way.
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20110203/6f6d66d2/attachment.html>