Allow ADL lookup of parse and toStream

Currently, if one introduces a Gaudi::Property<T>, then there must be a corresponding StatusCode Gaudi::Parsers::parse(T&,const std::string&) in the namespace Gaudi::Parsers and a corresponding void Gaudi::Utils::toStream(const T&, std::stream&) in the namespace Gaudi::Utils.

This patch explicitly enables ADL lookup (with a fall-back to Gaudi::Parser::parse and Gaudi::Utils::toStream) so that it is sufficient to either define parse and toStream in the namespace in which T is defined, or add them as friend functions to T.

Mentioned in merge request lhcb/LHCb!458 (merged)

Mentioned in merge request lhcb/Gaudi!96 (merged)

Mentioned in commit lhcb/Gaudi@aeed153c

Mentioned in merge request lhcb/LHCb!466 (merged)

@graven - I don't see where this is an improvement? I'd agree with that change if that would be part of new coding conventions. Maybe we should do that first?

It is actually a bigger deal than one would suspect -- injecting symbols in Gaudi::Parse and Gaudi::Utils seems to be very dependent on the order of things. It seems that in complex cases you must forward declare prior to any Gaudi include, and define downstream, after including some relevant Gaudi headers. By using ADL this problem disappears, and makes things a lot easier...

But did you really encounter it in this case or is it just a cleanup you are working on? If the latter, it really should find its way into conventions.

If you take a look at HltConfigSvc.h and HltConfigSvc.cpp, you will see an extremely early forward declaration in the header file and a very late definition in the .cpp file -- and getting this right was a matter of a lot of trial and error. And with the proposed change here, the name lookup is much easier. Note that this matches (not by coincidence) the Standard Library usage of eg. swap, begin, end and operator<<(std::ostream&, T). So no, it's not just cleanup, it simplifies actual problems and it is more than just a convention and yes, it also looks cleaner.

Agreed. And my implicit message was - put it into the conventions as well! ;-)

Approved this merge request

@clemenci - can you trigger a test?

Reassigned to @hegner

Validation started with lhcb-gaudi-merge#87

Status changed to merged

Mentioned in commit 291ba412