Geometry Boost List: [ggl] read_wkt and write_wkt

Subject: [ggl] read_wkt and write_wkt
From: Barend Gehrels (Barend.Gehrels)
Date: 2009-04-28 10:23:46

>   
>> The read_wkt function currently returns true/false if it is successfully or
>> not, while inside lexical_cast<> is used which might throw an exception. I
>> would like to change this by returning void and always throw an exception if
>> it is not successful. Is this OK for everyone?
>>
>> This exception is already defined as struct read_wkt_exception : public
>> geometry_exception, we need to move that geometry_exception to core. Is that
>> OK ?
>>     
>
> Both points sound fine to me.
>   
OK, I'll change it then.
>   
>> Boost defines an "exception/exception.hpp" which is not used by boost
>> libraries. At least not by e.g. spirit. So I derive from std::exception,
>> like most boost libraries do, is that OK?
>>     
>
> Boost folks usually like to avoid intra-dependencies, so I guess they
> don't have the intention to use Boost.Exception too much inside Boost.
> I propose to leave it like that for now. The second reason is that we
> don't know this library much for now. We can always integrate it in
> the future if we know it better and we realize it can bring us
> something.
>   
OK, makes sense. Indeed it is written somewhere that boost libraries 
should not use other boost libraries. I think in general this is an old 
guideline, or applicable to specific libraries. We're using many things 
of Boost which are specific for library programmers (remove_const) or 
very handy (ranges).

>   
>> About 2)
>> The "make_" prefix is normally reserved for object generators. This is one
>> but actually it is, more strictly, generating an manipulator (which is an
>> object). Is the "make_" prefix appropriate here? I thought I asked this
>> earlier but cannot find that mail so probably I didn't. Alternatively it
>> would be:
>> 1) cout << wkt_manip<G>(geometry) << endl
>> 2) cout << wkt(geometry) << end // because this will be most widely used
>>     
>
> I'd say that the make_prefix is used:
> - when the function generates an object whose constructor would
> otherwise need one or several template parameters
> AND
> - when the object is known by the user and usually explicitly
> manipulated by him (e.g. std::pair).
>
> Am I right to say that the wkt<> class is rather a "helper" class
> whose only purpose is to pass a geometry to a stream, and will never
> be used explicitly by the user for any other purpose? If it's the case
> then we should forget the make_ prefix and stick to wkt(geometry).
>   
OK, makes also sense. I already did the veshape like this. So changing 
the wkt is a small job, it is slightly more work to update the samples. 
But it looks better indeed.
>   
>> About 3)
>> The last one is prone to errors if geometry is not a geometry or not a tag.
>> It is not living in namespace ggl to enable the library to stream custom
>> geometries which are living outside the namespace ggl. So it is not in any
>> namespace. Bruno recently mailed me that this behaviour will never be
>> accepted by Boost so we have to re-add the namespace there, or remove it at
>> all.
>>     
>
> I really don't know what's allowed and what's not in this regard. We
> really should have a discussion on the Boost list about "is it allowed
> to enable operator<< for any class satisfying a given concept" and if
> yes, what are the rules to follow.
>
>   
OK, the problem is actually also that it tries to stream everything not 
streamable and then complains that it is not satisfying the concept. 
I'll have a look again, didn't touch this part for months.

Regards, Barend

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.osgeo.org/pipermail/ggl/attachments/20090428/79894348/attachment-0001.html