[gdal-dev] Minimum supported C and C++ standards

[Even Rouault]

Le lundi 09 mai 2016 19:28:48, vous avez écrit :
> Here is my current take on language standards
> 
> - I am working on a C++11/14 C99/11 proposal.  I despirately want to be
> able to use C++11 to make GDAL more robust

C++11 is probably OK, but C++14 support is really "new". For Linux distros, we 
should try to make GDAL compilable on the current LTS and probably in the case 
of Ubuntu last-1 LTS with their default compiler.

For example RHEL 7 ships with gcc 4.8.X, Ubuntu 14.04 too.
I've checked that gcc 4.8 refuses -std=c++14. It accepts -std=c++1y though, so 
some features of C++14 might be available.

At least for the sake of our CI environements : Travis has no immediate plans 
for now to ship with with Ubuntu 16.04 ( https://github.com/travis-ci/travis-
ci/issues/5821 )

> - As it stands I will vote >>>against<<< my proposal any time soon - until
> with have a bunch of proposals for changes to GDAL that require new
> language versions or someone rewrites my proposal to be much more
> compelling -
Which proposal exactly ? AFAICS in this thread, it was more about polling 
opinions.

> We have lots of work that we can do to make GDAL maintenance
> / debugging easier that does not require newer language versions
> - We have successfully ( at least in my opinion ) stuck a toe in the C++11
> world with CPL_FINAL, CPL_DISALLOW_COPY_ASSIGN, NULL_AS_NULLPTR, and
> my std::lock_guard<std::mutex> experiments.  We can probably do a bit more
> if we are careful.
> - I think having an experimental autotest2 C++11/14 only testing tree might
> be the way to start.  Those who want to can try it out or look through what
> is possible and it won't impact the main tree or the existing C++ tests.
> 
> 
> I would be happy to start contributing to
> https://trac.osgeo.org/gdal/browser/trunk/autotest2, especially if we call
> it experimental and I can get help with adding a build system.  I have
> 
> >5000 lines of C++ in 32 test files that mostly cover the port directory
> 
> right now.  I got hung up on trying to create a working initial version in
> a separate github tree.  Adding it to the existing svn tree would get me
> around that issue (it's an internal work thing).  Here is
> cpl_string_test.cc... it's super boring.
> 
>    https://gist.github.com/schwehr/02128959ee78d56b553defa0a527bdf2
> 
> It's written using the Google C++ style guide and is based on gunit, gmock
> and glog.
> 
> - https://google.github.io/styleguide/cppguide.html
> - https://github.com/google/googletest
> - https://github.com/google/glog

Are those available as ready to be installable packages ? (thinking about 
integration with CI)

> 
> On Mon, May 9, 2016 at 2:33 AM, Mateusz Loskot <mateusz at loskot.net> wrote:
> > On 7 May 2016 at 19:10, Kurt Schwehr <schwehr at gmail.com> wrote:
> > >> If we move to a later C++ standard, or even use features of C++98 we
> > >> currently
> > >> don't use, I'd advocate for using things that are obviously making the
> > >> code
> > >> better / more readable. Honestly who finds that
> > >> "std::unique_ptr<int *, std::function<void(char *)>>
> > >> Vals(CPLCalloc(256, 0),
> > >> CPLFree);" is obviously more readable, efficient and less error prone
> > 
> > than
> > 
> > >> "std::vector Vals(256,0)" ?
> > > 
> > > This is cart before the horse but... as fast as I can so expect typos.
> > 
> > Now
> > 
> > > just think of a ~1K long function or method with tons of instances and
> > 
> > lots
> > 
> > > of places to bailout successfully or as failures.  We have > 9K
> > > free/CPLFree/CPLdelete/CPLDestroys that could be < ~100.
> > > [...]
> > 
> > ...the very long story short, your desire is to introduce the RAII idiom
> > across GDAL codebase. Awesome!
> > I'm sure 99% of GDAL committers will welcome this idea.
> > Then, **next** question is how we want to implement it:
> > using C++11+ features, home-brewed smart pointer class(es), etc.
> > 
> > Upgrading to C++11+ just for the sake of upgrade, makes little sense to
> > me. Especially if folks are not certain about it, don't use those
> > features daily,
> > don't feel comfortable...it may cause more harm.
> > 
> > Best regards,
> > --
> > Mateusz Loskot, http://mateusz.loskot.net

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com