r224012 - Emit warning if define or undef reserved identifier or keyword.

[Nico Weber]

On Thu, Dec 18, 2014 at 2:01 PM, Richard Smith <richard at metafoo.co.uk>
wrote:

> On Wed, Dec 17, 2014 at 7:58 PM, Nico Weber <thakis at chromium.org> wrote:
>>
>> On Wed, Dec 17, 2014 at 7:05 PM, Richard Smith <richard at metafoo.co.uk>
>> wrote:
>>>
>>> On Tue, Dec 16, 2014 at 1:17 PM, Nico Weber <thakis at chromium.org> wrote:
>>>>
>>>> Ok, I moved -Wkeyword-macro into -pedantic in r224371. I also removed
>>>> -Wreserved-id-macro in the same revision, as it's not covered by the
>>>> standards bit you cited
>>>>
>>>
>>> This seems backwards?
>>>
>>> -Wreserved-id-macro is (was) warning on code that is ill-formed.
>>> -Wkeyword-macro is warning on code that has undefined behavior (only) if a
>>> standard library header is #included.
>>>
>>> C++ [lex.name]p3 (introduced by DR, so applying retroactively to all
>>> C++ standards) says:
>>>
>>> "[S]ome identifiers are reserved for use by C ++ implementations and
>>> shall not be used otherwise; no diagnostic is required.
>>> — Each identifier that contains a double underscore __ or begins with an
>>> underscore followed by an uppercase letter is reserved to the
>>> implementation for any use.
>>> — Each identifier that begins with an underscore is reserved to the
>>> implementation for use as a name in the global namespace."
>>>
>>> Aaron's quote of [macro.names]p2 applies only to "C++ programs that use
>>> the facilities of the C ++ standard library." (per
>>> [constraints.overview]p1), and "If a program declares or defines a name in
>>> a context where it is reserved, other than as explicitly allowed by
>>> this Clause, its behavior is undefined." (per [reserved.names]p1).
>>>
>>
>> I'm not sure I understand the point here – do you mean the warning should
>> only fire if a standard library has been included? In pedantic mode, I
>> would expect that people would want to know about stuff that's undefined
>> behavior even if they don't happen to include a standard library header at
>> the moment.
>>
>
> The point is that -Wreserved-id-macro is *more* breaking-the-rules than
> -Wkeyword-macro, so I disagree with your argument for removing
> -Wreserved-id-macro and keeping -Wkeyword-macro.
>

The bit you quoted says "some identifiers are reserved for use by C ++
implementations and shall not be used otherwise", but a C++ implementation
can choose that is going to use a __identifier as part of it's public
interface, no? Something like "Define __no_exceptions if you want a
standard library without exceptions" or similar – in this case, user code
needs to define this macro and it'll be standards-compliant since it's
using the __identifier as intended by the C++ implementation it uses. Due
to this, it's difficult to predict which __identifiers are ok to define and
which aren't. The bit you quoted at least doesn't say "don't define
__idents", it just says "use __idents only as intended by your C++
implementation" as far as I can tell.

The other paragraph explicitly says "shall not #define or #undef names
lexically identical to keywords […]" (, in "C++ programs that use the
facilities of the C ++ standard library."). I agree that the warning would
be More Correct if it checked for that too, but it seems like a minor
detail in practice, as most programs use the C++ standard library – minor
enough to not revert the warning over it, at least.

(But this is discussion is somewhat pointless now that r224512
added-Wkeyword-macro back :-) )


>
>
>> In your commit message for r224371 (which hasn't yet hit the mailing
>>> list), you say:
>>>
>>> "-Wreserved-id-macro warns on
>>>
>>>     #define __need_size_t
>>>
>>> which is more or less public api for glibc headers."
>>>
>>> I don't agree with your example; these __need_* macros exist only to
>>> allow communication between the /usr/include headers and the
>>> compiler-provided <stddef.h>, because /usr/include needs to include a
>>> subset of <stddef.h> depending on which of its headers is included. They're
>>> not for end-users, and these warnings should be suppressed in system
>>> headers.
>>>
>>
>> Some end user code uses them, for example to work around bugs in
>> /usr/include headers. I know of one example in Chromium (
>> https://code.google.com/p/chromium/codesearch#chromium/src/chrome/browser/ui/libgtk2ui/gtk2_ui.cc&q=__need_Null&sq=package:chromium&type=cs&l=72
>> ) and I found
>> https://github.com/asfernandes/firebird/blob/35b06c7892e2440c4f4528b9ecc99969927728e0/src/common/utils.cpp#L35
>> (there might be more, but github's search is somewhat hard to operate.)
>>
>> Having said that, I'm happy with the warning coming back in an improved
>> form.
>>
>>
>>> That said, there are a bunch of reserved macro names that *is* part of
>>> the glibc public API, such as the _*_SOURCE macro family. If we reinstate
>>> this warning, we should whitelist those ones at least.
>>>
>>> On Mon, Dec 15, 2014 at 10:39 AM, Joerg Sonnenberger <
>>>> joerg at britannica.bec.de> wrote:
>>>>>
>>>>> On Mon, Dec 15, 2014 at 10:24:22AM -0800, Nico Weber wrote:
>>>>> > For the "final(a,b,c)" define: should we warn on that? It doesn't
>>>>> change
>>>>> > the meaning of the keyword "final" since that doesn't have
>>>>> parameters.
>>>>>
>>>>> For C++11 and newer: IMO yes. Otherwise, no.
>>>>>
>>>>> Joerg
>>>>> _______________________________________________
>>>>> cfe-commits mailing list
>>>>> cfe-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>>>
>>>>
>>>> _______________________________________________
>>>> cfe-commits mailing list
>>>> cfe-commits at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20141218/fc8500a1/attachment.html>