[cfe-commits] r158796 - in /cfe/trunk: include/clang/Driver/Options.td lib/Driver/Tools.cpp test/Driver/warning-options.cpp

[Chandler Carruth]

On Wed, Jun 20, 2012 at 11:47 AM, Chris Lattner <clattner at apple.com> wrote:

>
> On Jun 20, 2012, at 8:16 AM, Ted Kremenek wrote:
>
> > On Jun 20, 2012, at 12:33 AM, John McCall <rjmccall at apple.com> wrote:
> >
> >> On Jun 20, 2012, at 12:21 AM, Chandler Carruth wrote:
> >>> On Wed, Jun 20, 2012 at 12:03 AM, Ted Kremenek <kremenek at apple.com>
> wrote:
> >>> Author: kremenek
> >>> Date: Wed Jun 20 02:03:37 2012
> >>> New Revision: 158796
> >>>
> >>> URL: http://llvm.org/viewvc/llvm-project?rev=158796&view=rev
> >>> Log:
> >>> Revert "Provide a -no-pedantic to cancel out -pedantic."  This needs
> to be designed
> >>> a bit further.  We may wish to just have -Wno flags to silence
> warnings, and not have a -no-pedantic.
> >>>
> >>> Why? This seems to make using '-pedantic' fairly problematic...
> >>>
> >>> A lot of build systems essentially only allow appending flags to
> change the build flags in use, and it seems bad to leave flags around which
> "poison" the entire command because of a lack of a '-no-*' variant.
> >>>
> >>> Note that I'm not really endorsing the use of '-pedantic'; I generally
> prefer the -W flags (with their -Wno- variants). I'm just a bit concerned
> about not allowing appending an option to reverse the decision about
> '-pedantic'.
> >>
> >> I think Ted's position here is that he wants -pedantic to be an alias
> for -Wpedantic, which would be cancelled by -Wno-pedantic.  I continue to
> believe that, even if we provide -Wno-pedantic, the existence of -pedantic
> basically mandates an analogous -no-pedantic as well;  however, Ted asked
> me to revert while we discussed it, only he beat me to it.
> >
> > Thanks John.  That's sums it up well.  Right now the workflow people
> know is to pass -Wno- to silence a warning, and seeing the warning flag in
> the diagnostic.  This flag is completely different from that simple
> workflow.
> >
> > My understanding was that ever clang warning should be controllable
> under a -W flag.  That's not the case with all -pedantic warnings.
>
> I'd also like it if we never produced [-pedantic] in a diagnostic as the
> warning flag.  [-Wpedantic] would be much more consistent.


I completely agree with presenting the user *only* with '-Wpedantic' and
variants.

That said, for compatibility, I think we should support '-pedantic' and
'-no-pedantic' as aliases for '-Wpedantic' and '-Wno-pedantic'
respectively. I don't really like the flags either, but I'm not thrilled
about explaining that the solution to negate '-pedantic' is to pass
'-Wno-pedantic'. =/

(and yes, I would expect '-pedantic-errors' to expand to '-Wpedantic
-Werror=pedantic', and '-no-pedantic-errors' no expand to '-Wno-pedantic
-Wno-error=pedantic')

Yuck. Yay for commandline legacy... =/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20120620/fccbd7b2/attachment.html>