[patch] [libcxx] #warning

[G M]

Hi Everyone

Here's a revision of my previous #warning patch that provides the
_LIBCPP_WARNING macro, to be used for MSVC only, since that compiler
doesn't support #warning.

This revision updates the #if/#else usage of the macro to accommodate
Howard's stated preference there.

Despite the clutter, given the verbosity and #pragma issues Howard mentions
that make this macro undesirable to be used in clang++ and g++, I've
removed the implementation of the macro for those compilers so as not to
encourage such undesirable use as #warning remains the preferred means to
achieve warnings for compilers other than MSVC.



On Sat, Sep 21, 2013 at 1:26 PM, Howard Hinnant <howard.hinnant at gmail.com>wrote:

> On Sep 16, 2013, at 11:49 PM, G M <gmisocpp at gmail.com> wrote:
>
> > Hello Everyone
> >
> > Attached is a patch to libcxx to remove the errors relating to the use
> of #warning when compiling using MS's compiler cl.exe
> >
> > It introduces a macro to __config _LIBCPP_WARNING for use with the MS
> compiler and updates all references of #warning to use it when compiling
> with the real MS compiler.
> >
> > The macro is compatible with clang++ and g++ and can be used there too
> if you understand and accept the slight incompatibilities, which I'll
> explain. From my investigations, if I am correct, the limitations of this
> macro and #warning and #pragma message are:
> >
> > * #warning doesn't work on any version of visual studio. In fact it
> means something different to MS and it's used extensively; so they will
> likely never standardize on the use case here as they can't IMHO.
> >
> > * #pragma message DOES work for clang++/g++/cl, BUT not completely
> compatibly:
> >
> > The differences in #pragma message are:
> >
> > * For g++, it generates a note, which is logical to me on the basis that
> if you want a warning, you have #warning from gcc's perspective.
> > * For MSVC, it generates a note like gcc (or appears to) at the command
> line, but the message does show up in the warning lists of Visual Studio
> but as far as I can tell it's not really a true warning. The message
> doesn't include a file or line number on the command line (and that may be
> to too true in the IDE I forgot to check) but on the command line, it's too
> brief and is easy to miss if used in a raw way.
> >
> > * For clang++, #pragma message generates a warning not a message/note
> like g++/cl. This is a very practical implementation, but a little
> misleading given the name and isn't quite as compatible with gcc as one
> might expect. Given these aspects, it's arguable that clang++ should be
> changed to issue a note instead of warning to match gcc, but I'm not
> advocating that change.
> >
> > * clang++ is however quite verbose in it's formatting of #pragma message
> and #warning compared to g++. I think too verbose. It's so verbose when
> used as a macro in fact that I think it's undesirable. IMHO, I think in
> this formatting aspect, clang++ should be changed to match g++.
> >
> > It would be better IMHO if it was tuned to remove the macro info and
> show it only if there is an actual error or problem in the formulation and
> use in the macro. If not, I think the use of the macro should appear in
> indistinguishable in use as it is in g++. And only be expanded in full if
> there is a real error, or another option is applied to force it for
> debugging purposes or something. For my purpose, it's arguable that the
> verbosity makes it somewhat useless. clang++ is just a little too helpful
> here, love it as I do. That is part of the reason I don't use this macro
> for anything other than VS.
> >
> > I've reported this verbosity as a undesirable feature to Richard Smith
> to see if he agrees but I (and I'm sure he) would be interested in seeing
> other opinions on here on the subject. I hope that "someone" will "fix" it
> and make clang++ match g++ more here.
> >
> > I've tested the attached patch with clang, Visual Studio and mingw / g++.
> > Thanks for your thoughts.
> >
> >
> > <warning.diff>
>
> The current verbosity on clang, combined with the technique for disabling
> the warning on clang makes this undesirable for me (-W#warnings vs
> -W#pragma-messages).  I'm guessing there are already projects out there
> that are disabling the libc++ #warnings with (for example):
>
> #pragma clang diagnostic ignored "-W#warnings"
>
> And I've insufficient motivation to rock their boat.  However I've no
> objection to something along the lines of:
>
> #if defined(_MSC_VER) && ! __clang__
>     _LIBCPP_WARNING("cl user triggered warning");
> #else
> #    warning user triggered warning;
> #endif
>
> Howard
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130922/ac64bc0a/attachment.html>