[cfe-dev] Warning when calling virtual functions from constructor/desctructor?

[Arthur O'Dwyer via cfe-dev]

On Tue, Jan 8, 2019 at 1:45 PM Arthur O'Dwyer <arthur.j.odwyer at gmail.com>
wrote:

> On Tue, Jan 8, 2019 at 1:05 PM John McCall via cfe-dev <
> cfe-dev at lists.llvm.org> wrote:
>
>> On 8 Jan 2019, at 8:37, Arnaud Bienner wrote:
>> >
>> > I realized I didn't put "DefaultIgnore" on this warning.
>> > I'm not experienced enough with clang to know what's the best way to
>> deal
>> > with new warnings, but my feeling is that it would be sensible to have
>> this
>> > new warning DefaultIgnore for now, in -Wall, and make it default once we
>> > have some feedback from the community: while not all C++ projects use
>> -Wall
>> > (or -Wextra) I believe enough do to give us a chance to get some
>> feedback.
>> >
>> > What do you think?
>>
>> We generally don't add things to -Wall.  That's why I went into my whole
>> spiel
>> about versioning: I think it's a conversation we need to have before we're
>> ready to accept this as a warning that's anything but hidden permanently
>> behind its own opt-in flag.
>>
>
> John: Wha? Clang *frequently* adds things to -Wall!  -Wall includes
> -Wmost which includes a bunch of other categories, so while we don't often
> put new diagnostics *directly* under -Wall, pretty much every
> "reasonable" diagnostic eventually winds up in there somehow — which is the
> intent.
>
> However, Arnaud: I don't think there's much difference between
> "on-by-default" and "on only with -Wall", because the received wisdom has
> always been to always compile with -W -Wall [-Wextra].  You're saying
> "don't worry, my change will affect *only* the people who pass -Wall" —
> but that's actually a huge percentage of the world, like maybe 90% (wild
> unscientific guess).  If you're worried that your diagnostic might be
> "unreasonable," then it doesn't belong in -Wall any more than it belongs
> on-by-default.
>
> I think Richard Smith's suggestion was right: the decision of where to put
> this diagnostic will become simpler if the diagnostic itself becomes more
> targeted and loses some of its current false positives.
>

OOPS: sorry, Richard's comment that I was thinking of was on
https://reviews.llvm.org/D56405, which is also about a "virtual"-related
diagnostic but not directly relevant to Arnaud's diagnostic IIUC.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20190108/334157fc/attachment.html>