Boost logo

Boost :

From: Reece Dunn (msclrhd_at_[hidden])
Date: 2003-12-08 17:00:24

Jonathan Turkanis wrote:

>I've looked at your library in detail now, except for the State Outputters.
>Good work!


>1. Surprisingly, partial template specialization is not necessary. See
> for a version
>with no partial spec. A good deal of the library now works on VC6. As a
>effect, output deduction now works fully on borland 5.6.4. (I don't know
>about 5.5.1.) I've also tested it on Intel 7.1.

It also works on borland 5.5.1 :). I was thinking that it would be possible
to rewrite the deduction code using Boost.MPL, but didn't know that much
about it. Great work!

>The headers type_deducer.hpp and output_deducer.hpp had to be mostly
>rewritten; otherwise, I made some small changes to format_traits and
>pair_traits. A detailed list of changes is included with the zip.

I have had a look at the changes you have made. They have been included into
the library.

I have also modified the way hash containers are detected, improving the
detection method: it should now work for any SGI/dinkumware supported
compilers. I haven't looked at the slist container yet, but I agree that it
should be supported.

I have also added support for compilers that have older I/O stream libraries
without std::basic_ostream support. This means that GCC 2.95.3-5 is now
(partially) working. There seems to be a problem in
<boost/io/detail/output_deducer.hpp>, and the hash container stuff is not
working for some reason in <boost/io/detail/type_traits.hpp>.

>Because boost::remove_bounds requires partial spec, array deduction does
>work properly on VC6. There are apparently workarounds for common array
>types in boost-sandbox\boost\string_algo\detail, which you might wan't to
>look at.

I will do :).

>2. My gut feeling tells me that the library can be simplified a bit. I'm
>particularly worried about the large number of object generators. Perhaps a
>more extensive tutorial would clear up some of my doubts. With your
>permission, I'd like to hack on the code a bit to see if I can make it a
>little simpler, without losing functionality.

Feel free. My reasoning as to having the three variants was thus:

[1] The base generator should supply default formatting type (char *) and
[2] Custom formatting types are supported as a function template parameter.

   boost::io::formatlistex< wchar_t * >( l );

Since you can't define a default value, this has to be a seperate generator.

[3] Custom outputters are supported as a specialist generator.

   boost::io::formatobout( o, ... );

These have different names because of problems with locating the correct
overloaded variant.

Each of these variants is required for each outputter class to allow maximum

>First, please let me know what you think of my changes.

They rock!

My updated code is at:

Reece H Dunn

On the move? Get Hotmail on your mobile phone

Boost list run by bdawes at, gregod at, cpdaniel at, john at