The interface is now more consistent, but sadly still a bit repetitiv.

An issue in getHist_i was fixed, where multiple matches returned a
nullptr, which was equivalent to no match in the THistSvc::exitsts
function. We return the first match now instead.

The Interface should deal with unique_ptr instead of the raw pointer.
The unique_ptr is released, once the service gets ownership. This
simplifies the integration with the existing code.

The verifier needs to Sort certain lines in the output before comparing
it to the reference file.

The CI job failed because of different clang-format versions.

duplicated Histograms were cleared after a successful merge, but they
were still associated to the output file, which resulted in empty
histograms overwriting the desired one.
SetDirectory(nullptr) before Reset() of these copies solved the issue.

Charles Leggett authored 8 years ago and Martin Errenst committed 7 years ago

added merging of hists from clones (on demand, not automatic)

TODO: which clones to write out?

modified:   GaudiKernel/GaudiKernel/ITHistSvc.h
modified:   GaudiSvc/src/THistSvc/THistSvc.cpp
modified:   GaudiSvc/src/THistSvc/THistSvc.h
modified:   GaudiSvc/src/THistSvc/THistSvc.icc

Charles Leggett authored 8 years ago and Martin Errenst committed 7 years ago

by default, alg clones get their own copies of hists.

if reg as a shared hist, get a LockedHandle which does a mutex lock/unlock
around each deref.

new interfaces:

  virtual StatusCode regSharedHist(const std::string& name, std::unique_ptr<TH1>&,
                                   LockedHandle<TH1>&) = 0;
  virtual StatusCode regSharedHist(const std::string& name, std::unique_ptr<TH2>&,
                                   LockedHandle<TH2>&) = 0;
  virtual StatusCode regSharedHist(const std::string& name, std::unique_ptr<TH3>&,
                                   LockedHandle<TH3>&) = 0;

  virtual StatusCode getSharedHist(const std::string& name,
                                   LockedHandle<TH1>&) const = 0;

THistSvc internals:
  - identifiers kept as unordered_map< vector<THistID>* >.
  - TObject map is unordered_map< TObject*, pair< vhid_t*, size_t >>
    to index within vector<THistID>*

  - THistID holds mutex for shared hists.

  - GlobalDirectoryRestore also acts as a lock_guard

modified:   GaudiKernel/GaudiKernel/ITHistSvc.h
new file:   GaudiKernel/GaudiKernel/LockedHandle.h
modified:   GaudiSvc/src/THistSvc/THistSvc.cpp
modified:   GaudiSvc/src/THistSvc/THistSvc.h
modified:   GaudiSvc/src/THistSvc/THistSvc.icc

Fixed use of messages in ApplicationMgr::i_startup

See merge request !436

See merge request !447

Adding DecisionNode negation mapping the "Invert" parameter of the GaudiSequencer

Since the introduction of Phoenix v3 (in boost 1.47) the 'result'
struct is no longer needed in spirit parser. See eg.
http://www.boost.org/doc/libs/1_47_0/libs/phoenix/doc/html/phoenix/what_s_new/phoenix_3_0.html
which points out the switch to the 'Boost::Result_of' protocol

Modernize genconf

See merge request !437

and rename the variables used to iterate over the values returned to
indicate they are names of factories.

Also, remove some pre-C++11 compatibility scaffolding

- prefer enum over std::string
- prefer freestanding function in anonymous namespace over member
  functions
- prefer direct member initialization
- prefer auto
- prefer range-based for loops
- use std::type_index to compare std::type_info objects (the latter
  are not guaranteed to be unique for a given type!)
- prefer std::unordered_map over Gaudi::HashMap
- prefer std::vector<std::pair<std::string,std::string>> over
  Gaudi::HashMap if no lookup is required

fix formatting

See merge request !442

tweak overload resolution of TupleObj::farray  for  gcc7

See merge request !430

Fix task-based conditions handling

See merge request !431

Fix various printouts

See merge request !416