[cfe-dev] Preprocessor operators for #error and #warning

[Richard Smith]

On Sun, Apr 13, 2014 at 1:55 PM, Kyle Sluder <kyle at ksluder.com> wrote:

> On Apr 13, 2014, at 1:29 PM, Richard Smith <richard at metafoo.co.uk> wrote:
>
> On Sun, Apr 13, 2014 at 12:38 PM, Kyle Sluder <kyle at ksluder.com> wrote:
>
>> On Apr 12, 2014, at 2:34 AM, David Chisnall <David.Chisnall at cl.cam.ac.uk>
>> wrote:
>> >
>> > For _Error, you can use C11's _Static_assert():
>> >
>> > #define fail(msg) _Static_assert(0, msg)
>> >
>> > There's no standard equivalent for #warning, unfortunately.
>>
>> That workaround is sufficient for my use case, but the output is noisy
>> and as you note it doesn’t work for those who treat warning differently
>> from errors.
>>
>> So, can anyone opine on whether such a patch would be accepted?
>
>
> I don't think we want _Warning and _Error. You can use #pragma message to
> emit a warning (as _Pragma("message \"...\"") within your macro).
>
>
> Such a macro definition is sadly impossible because the C99 spec is too
> strict: _Pragma only accepts a single string literal. No implicit
> concatenation of string literals is allowed.
>
> If the clang folks are fine with relaxing that, then _Pragma becomes
> sufficient for my purposes. But I presume it was so strictly defined in the
> spec for some reason.
>

You don't need string literal concatenation to make this work. Here's a
hint, from C11 6.10.9/2:

#define PRAGMA(x) _Pragma(#x)

I'd much prefer to add support for something like '#pragma clang error
> "blah"' rather than adding something like _Error(...), if _Static_assert is
> somehow insufficient, but I'd like to understand what's wrong with
> _Static_assert first.
>
>
> Well, for starters, it can’t be used to produce a warning. That alone
> makes it worth contemplating alternatives, in my view.
>

I explained how to get a warning above =) I was only talking about the
error case here.

But even for the error-producing use case, the error output contains a lot
> of noise about failing nonsensical assertion conditions. The best thing
> about #error and #warning is that I get to control the output, and that
> lets me match clang’s much-lauded user-friendliness in my own error
> messages.
>

OK, if you don't like the 'static assertion failed' text added by
_Static_assert, then use this:

#define pragma(...) _Pragma(#__VA_ARGS__)
#define warning(msg) pragma(GCC warning(msg))
#define error(msg) pragma(GCC error(msg))
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20140414/7d5f5f0b/attachment.html>