[cfe-dev] literal suffixes for std::basic_string?

[Richard Smith]

On Mon, Jul 22, 2013 at 12:05 PM, Marshall Clow <mclow.lists at gmail.com>wrote:

>
> On Jul 22, 2013, at 11:59 AM, Eli Friedman <eli.friedman at gmail.com> wrote:
>
> > On Mon, Jul 22, 2013 at 11:22 AM, Marshall Clow <mclow.lists at gmail.com>
> wrote:
> >> I'm trying to implement N3642 in libc++ (User-defined literals for
> standard library types) and I'm running into something I don't understand.
> >>
> >> I tried to implement them in a standalone program, and got errors from
> clang:
> >>> string_literal.cpp:6:21: warning: user-defined literal suffixes not
> starting
> >>>      with '_' are reserved; no literal will invoke this operator
> >>
> >> Ok, fine. I developed my (and tested my code) using underscores. That
> all works (see attached "string_literal.cpp")
> >>
> >> [ All the code here was built with TOT clang and "-std=c++1y
> -stdlib=libc++ -I $LLVM/libcxx/include"  ]
> >>
> >> Then I dropped the code into libc++'s <string> header and removed the
> underscores (patch attached).
> >> No warnings - so apparently clang can tell when the code is in a system
> header.
> >
> > Yes... but it's probably just suppressing the warning, not actually
> > changing the behavior.
> >
> >> But if I uncomment the code that uses those operators (starting about
> line 46 of string_literal.cpp), I get the following errors:
> >>
> >>> string_literal.cpp:46:48: error: invalid suffix on literal; C++11
> requires a
> >>>      space between literal and identifier
> [-Wreserved-user-defined-literal]
> >>>        static_assert ( std::is_same<decltype(   "hi"s),
> std::string>::v...
> >>>                                                     ^
> >>>
> >>> string_literal.cpp:46:48: error: expected ')'
> >>> string_literal.cpp:46:40: note: to match this '('
> >>>        static_assert ( std::is_same<decltype(   "hi"s),
> std::string>::v...
> >>>                                             ^
> >>> string_literal.cpp:46:3: error: static_assert failed ""
> >>>        static_assert ( std::is_same<decltype(   "hi"s),
> std::string>::value, "...
> >>>        ^
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >>
> >> which, it seems to me, are just wrong.
> >>
> >> Maybe I screwed up putting the stuff into <string> (but that looks ok
> to me).
> >> I thought I might be using the wrong set of libc++ header files, but I
> checked, and I don't think that's happening, either.
>

That's not how it works -- it'd be a massive layering violation for the
lexer to care about which names have been declared. See below.


> >>
> >> Any suggestions?
> >
> > User-defined literals without underscores are reserved... and so
> > without any guidance from the standard as to what they are supposed to
> > do, clang just ignores them outright.
>
> Um, I don't think that's right.
>
> They are reserved for standard library implementations, and the C++14
> draft defines several such suffixes that library implementers must
> implement.


Right. The lexer change to support UDLs breaks a large set of programs
(including most users of the PRI* and SCN* macros added from C99 in C++11),
so since all such suffixes were reserved in C++11, we chose to lex them as
if in C++98 mode for compatibility. The intent was that once some standard
suffixes were added, we would modify the lexer to treat those suffixes as
UDL suffixes; we're not got around to doing that yet.

If you can provide a list of string suffixes that C++14 defines, it should
be straightforward to teach the lexer to treat those ones as UDLs. I think
it's just "s" we need to handle, but I've not checked.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20130722/e0788aca/attachment.html>