[wwwdocs, PATCH] C++ terminology: the One Definition Rule in diagnostics

Martin Sebor msebor@gmail.com
Tue Mar 14 15:26:00 GMT 2017


On 03/14/2017 08:18 AM, David Malcolm wrote:
> Looking at PR ipa/80000, which notes that ipa-devirt.c has two trailing
> spaces in:
>
> if (warning_at (
>       DECL_SOURCE_LOCATION(TYPE_NAME (DECL_CONTEXT (vtable->decl))),
>       OPT_Wodr,
>       "virtual table of type %qD violates "
>       "one definition rule  ",
>       DECL_CONTEXT (vtable->decl)))
>
> and thus emits:
>
>   foo.cc: virtual table of type 'foo' violates one definition rule
>
> ...I think the wording here is a little confusing to a novice C++
> coder: it could be misread that the compiler is complaining that there
> are "definition rules", and that one of them has been somehow violated
> ("why aren't you telling me which one?"), as opposed to the correct
> reading, that there is a rule that there must be one definition.

Heh.  I never would have thought of that.  But you're right.
It reminds me just how wide-spread the convention of dropping
articles is in diagnostics.  Would it make sense to put in
place a guideline to use them consistently?

   https://gcc.gnu.org/wiki/DiagnosticsGuidelines

> There are 5 such diagnostics, all in ipa-devirt.c.
>
> So I think the wording could be improved by adding a "the", and maybe
> capitalizing, to make it:
>
>   foo.cc: virtual table of type 'foo' violates the One Definition Rule
>
> (giving a big hint to the user that the "One Definition Rule" is the thing that they should be typing into their search engine, and that "ODR" is an acronym form of it).
>
> I've even seen it hyphenated in some places, which would make it:
>
>   foo.cc: virtual table of type 'foo' violates the One-Definition Rule
>
> or:
>
>   foo.cc: virtual table of type 'foo' violates the one-definition rule
>

The gcc.po file has these forms:

   msgid "Merge C++ types using One Definition Rule."
   msgid "virtual table of type %qD violates one definition rule"
   msgid "virtual table of type %qD violates one definition rule  "
   msgid "type %qT violates the C++ One Definition Rule"
   msgid "type %qT itself violate the C++ One Definition Rule"
   msgid "%qD violates the C++ One Definition Rule "

I like the last one.  My only question is about capitalization.
GCC is inconsistent about it (not just in this case but in many
others).  Most messages are all lowercase, but some capitalize
established terms or keywords (FORTRAN), some even start with
a capital letter (FORTRAN).  The GCC Diagnostic Guidelines don't
mention capitalization and the GNU Coding Standard recommends
only that messages issued by interactive programs start with
a capital letter and others don't.  It would be nice to have
clarity about what the convention is (or what we agree it
should be) and eventually converge on it in all sources.

Martin

>
> Here's a patch to the terminology part of our conventions, which suggests adding a "the" and capitalizing (but without the hyphen).
>
>
> Thoughts?
> Dave
>



More information about the Gcc mailing list