On 10 April 2018 at 23:06, Even Rouault <[email protected]> wrote: > On mardi 10 avril 2018 22:34:56 CEST Mateusz Loskot wrote: >> On 10 April 2018 at 22:06, Even Rouault <[email protected]> wrote: >> > >> > Just wanted to advertize the recent new introduction of C++ iterators >> > (...) >> >> Good one, thanks! >> >> > for( auto&& poLayer: poDS->GetLayers() ) >> >> Do the iterators return a proxy reference hence the tutorial prefers >> auto&& over plain auto&? >> (ie. like std::vector<bool> iterator does) > > Not completely sure to understand what a proxy reference is (C++11 stuff is > still very new to me, and I don't pretend to master C++98, so ...), but I feel > that must be close to what I have used in some cases, particularly wen > iterating over fields of a features, or points of a linestring.
I didn't look at the iterators impl. yet. I just looked at your example and wondered if there are proxies - for GDAL/OGR, those could be necessary, I imagine. I mean a proxy as an object that 'looks' like a reference. It's often necessary to for (proxied) collections of objects that can not be accessed directly, via regular pointers/references. > I read in some tutorial that using auto&& worked in all situations, including > when the returned reference was const, so now I just use that everywhere. auto&& is most useful when container<T>::reference returns a proxy object, which would not bind to auto& (it would bind to auto const& though). > In fact the nature of the object which is returned depends on the iterator, > and as documented in the API, for the sake of implementation simplicity, those > iterators do not necessarily respect the whole semantics of iterators. > > That is if you do > > auto iter = poLineString->begin(); > OGRPoint& oPoint = *iter; > ++iter; > OGRPoint& oAnotherPoint = *iter; > In fact &oAnotherPoint == &oPoint. At first, it seems like they model the input iterators, single-pass iterator category http://en.cppreference.com/w/cpp/concept/InputIterator > So basically you should only use them in range based loops, where you don't > really manipulate the iterator and cannot do stuff like the above. Or, use like any input iterators, seems safe. > I guess the "correct" solution would have been to return a OGRPointProxy that > would keep a link to the linestring and the point index. But not sure how you > would do that for a const iterator where the operator * is supposed to > returned a const reference... Or reference-like proxy like std::vector<bool> does. Best regards, -- Mateusz Loskot, http://mateusz.loskot.net _______________________________________________ gdal-dev mailing list [email protected] https://lists.osgeo.org/mailman/listinfo/gdal-dev
