[Patch] Disable -Wtaulological-compare for substituted template types

[Richard Trieu]

Committed in r193887 and r193888


On Fri, Nov 1, 2013 at 1:02 PM, David Blaikie <dblaikie at gmail.com> wrote:

>
>
>
> On Fri, Nov 1, 2013 at 12:15 PM, Arthur O'Dwyer <arthur.j.odwyer at gmail.com
> > wrote:
>
>> On Thu, Oct 31, 2013 at 5:52 PM, Richard Trieu <rtrieu at google.com> wrote:
>> > On Thu, Oct 31, 2013 at 4:25 PM, David Blaikie <dblaikie at gmail.com>
>> wrote:
>> >>
>> >> Could we just restrict these warnings to only be done on template
>> >> patterns, not template specializations? (I know Ted's had some concern
>> about
>> >> doing work on template patterns in the past because he believed that
>> might
>> >> create too much work analyzing templates that are never instantiated,
>> >> though)
>> >
>> > I think that would be possible and a cleaner way of approaching this.
>> > Currently, Clang warns in both template patters and specializations.
>>  That
>> > means if you have :
>> >
>> > template<int Num>
>> > bool greater(unsigned Val) {
>> >   return Val >= 0;
>> > }
>> >
>> > ... greater<0>(42); ...
>> > ... greater<1>(42); ...
>> > ... greater<2>(42); ...
>> >
>> > There would be four warnings on the return line.  Ignoring the three
>> > warnings in the specializations and only showing the one from the
>> template
>> > pattern would be best on cutting down the noise.
>>
>> Only for -Wtautological-compare, though, right?
>>
>> I imagine that there are plenty of warnings that would be useful to
>> produce during template instantiation, but impossible to detect by
>> just looking at the pattern (without checking dependent
>> types/expressions). A trivial example would be
>> -Wuninitialized for this code:
>>
>>     template<typename T> T newval() { T t; return t; }
>>
>> Here, newval<int>() has undefined behavior, but newval<std::string>()
>> is perfectly correct.
>>
>> So whereas -Wtautological-compare empirically gives almost 100% false
>> positives on dependent expressions, -Wuninitialized probably gives
>> 100% *true* positives.
>
>
> Certainly. There are many warnings in each of these buckets, but generally
> a given warning falls into only one - either it should be done on
> instantiations or patterns. (there are some similar issues with macros and
> other things (sizeof/decltype of implementation-provided types, etc))
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131101/aed17f49/attachment.html>