There are some cases where using auto for function return type deduction is necessary to be able to state what you want generically in templates, and I think we should allow that use case. I can ping some friends for good examples if people would like. Definitely general use of return type deduction shouldn't happen, but inside templates a lot of the time the code is a lot simpler, more readable, and easier to maintain when using auto (Esp. c++14 return type deduction auto), than if a type has to be hand-deduced.
Also note at times we just can't say what the type is, because it varies in the template instantiation. See my use in https://reviews.apache.org/r/25525/diff/, slaveinfo_utils.cpp where writing explicit types for memberFunc is really hard if not impossible. On Mon, Sep 22, 2014 at 9:57 AM, Dominic Hamon <dha...@twopensource.com> wrote: > For reference: > http://google-styleguide.googlecode.com/svn/trunk/cppguide.html#auto > > We should be able to adopt that wholesale but please document anywhere you > think we would diverge from those examples. > > On Mon, Sep 22, 2014 at 7:28 AM, Alex Rukletsov <a...@mesosphere.io> > wrote: > > > We now start using auto in the code (among several other C++11 features). > > However we don't want to hamper reasoning about types. I would suggest we > > select several use cases where we all agree using auto is welcomed and > > several counterexamples. After the short conversation with BenH, we > agreed > > that iterators on one side and Try<>, Option<> on the other side are > first > > candidates for such list. I put the examples here > > <https://reviews.apache.org/r/25622/>. > > > > Any thoughts and ideas? > > > > > > -- > Dominic Hamon | @mrdo | Twitter > *There are no bad ideas; only good ideas that go horribly wrong.* >