Howard Hinnant <hhinnant@apple.com> writes:
| On Dec 29, 2005, at 4:07 PM, Gabriel Dos Reis wrote:
|
| > Howard Hinnant <hhinnant@apple.com> writes:
| >
| > [...]
| >
| > | > As to overloads for std::cosf (note the suffix), I find the
| > | > justificaton for them as "C99 compatibility" as dubious.
| > |
| > | Interesting.
| >
| > cos() is "overloaded" in C99 through <tgmath.h>; not cosf. If the
| > motivation for a C++ library to add overloads for cosf is
| > compatibility then the motivation better finds a different, more
| > convincing, excuse.
| >
| > -- Gaby
|
| My mistake again. I thought Ed was simply asking for:
|
| float std::cosf( float );
|
| which is required by C99.
No C++ standard requires that signature (for C99 compatibility.)
What libstdc++ does is to try to mess minimally with your "C" headers
<xxx.h>, so that if you include <math.h> chances are that your cosf
() will
just work fine. There is no std::cosf() because no such thing is
mandated.
If I remember correctly (and my memory may be failing here), there is
a proposal to acknowledge that implementations could have <cxxx> put
everything in both global and std:: scope.
When the C99 compat problem was urgent, we tried several variations,
and at some point we put them in __gnu_cxx. The c_shawdow was an
effort to implement the standard letter, it never succeeded, partly
because for implementations like GCC, it was an unrealistic
requirement.
| Having that signature (not any overloads
| of it) in the name of "C99 compatibility" does not seem dubious to
| me.
Well, given that C99 does not have std::, the compatibility claim
needs
more explanation. If compability is a real issue, then the standard
should require that "C" headers are left *alone*, untouched. But the
standard did something else, causing more troubles, and now we need
more kludge in the name of compatibility?
| I'm not sure where we started discussing overloading cosf, nor
| with what signature.
-- Gaby