Frank Winklmeier authored 3 years ago and Frank Winklmeier committed 3 years ago

Remove `PackageInfo` (package name and version) from the CLID
database. This information was never correctly filled (ATEAM-731).

- use `extends` formalism for inheritance
- remove unnecessary calls to base-class init/finalize
- remove print of obsolete `PACKAGE_VERSION`
- add `virtual`/`override`

Add ITPCnvSvc::t2p_cnv_unique, which returns ownership to the caller,
and which doesn't print any warnings on failure

Modified `AthenaMPTools/SharedHiveEvtQueueConsumer` so that it can process
events concurrently. In order to achieve that this tool is now using
the `AthenaMtesEventLoopMgr` as the event processor, instead of the
previously used `AthenaHiveEventLoopMgr`.

Also introduced a helper abstract interface `AthenaKernel/IHybridProcessorHelper`
to allow the `SharedHiveEvtQueueConsumer` tool to call certain methods of the
`AthenaMtesEventLoopMgr`.

Move file closing funcionality that was earlier in IOVDbSvc::handle()
and was triggered by StoreCleaned incident to a new method:
IOVDbSvc::postConditionsLoad()
Make CondInputLoader call it after loading all conditions.
File closing releases TTreeCache memory (30MB * 3)

The Units header provides units objects for which division is automatically changed at compile-time
to multiplication by a reciprocal of the unit constant.  This relied on the compiler changing
(1./x) to multiplication by a reciprocal, if x is constant.  However, this does not work
with clang if the -ffp-exception-behavior=maytrap option is used.

Rewrite to fix this.  The price is that we need to have a separate class for each unit
(which can then return the value as a constexpr).

Frank Berghaus authored 4 years ago and Walter Lampl committed 4 years ago

Writing the output FileMetaData on preFinalize causes problems in
EventService: event ranges will be written out before the FileMetaData
object is created. This causes an error when the stream is told to write
an object that does not (yet) exist.

Enhancement: make the log messages from the FileMetaDataCreatorTool more
informative.

Fix a typo in comments.

    
Can't use full template-id to name a destructor.

Partially addresses ATEAM-677

Frank Berghaus authored 4 years ago and Frank Winklmeier committed 4 years ago

There should be no need to expose the header of the FileMetaDataTool to
other packages, it can be treated as a component and called through its
interface. Remove some unneeded headers. Enable checking for thread
safety.

Fix compilation failure with using fromStorable into a const pointer
(only showed up in debug builds).

Frank Berghaus authored 4 years ago and Adam Edward Barton committed 4 years ago

This has just laid dormant ever since the CMT->CMake migration. :-(

At the same time introduced some tests for the implicit copy-constructor
and copy-operator of xAOD::EventFormat.

When we cast from a DataBucket, first try the conversion using
the static SG_BASES information.  This can be unfolded and inlined
at compile-time using BaseList::foreach_.

Also need to move the CLID defininition for CondContBase from StoreGate
to where the class itself is declared; otherwise, we get errors
from cling.

A DataProxy holds a pointer to a DataObject.
When we were doing a casting operation from a DataProxy,
we were using dynamic_cast to convert from DataObject to DataBucketBase.
Nowadays, though, this DataObject is always an instance of a DataBucket.
So do a static_cast instead.
This speeds up the casting operation.

Change BaseInfo initialization to use the variable-length BaseList.
Remove fixed typedefs Base1, etc.
Add a varadic SG_BASES macro.

There have been two overloads for DataBucketBase::cast, one taking
a clid and one taking a type_info.  To cast from a DataProxy, we've
been using the type_info version.  However, doing it with a clid
can be a bit faster, so we'd like to use that.  There is a problem,
though, since xAODRootAccess also implements a DataBucket, but this
version only supports casting via type_info.

So add a third overload for cast that takes both a clid and a type_info
and give the callee the opportunity to choose which one to use.

Rework SG::Bases to (a) allow for an arbitrary number of base classes,
and (b) support compile-time iteration over the bases.

(This is working towards speeding up the casting operations performed
by DataProxy.)

    
Move the safe_clid function from SGTools to AthenaKernel.
(It currently has no clients.)

The Intel Clang compiler printed a warning every time it encountered
the SG::DataBucket destructor. This change in the source code silences
that warning.

Add abstract method IAthenaOutputTool::preStream().
To be called just before streaming objects.

Switch to using `RootType::ByNameNoQuiet` in `load_type` in order to
make this method thread-safe. In case of a missing dictionary, ROOT
would now print an error message, which we do not expect to happen.

ThinningCache holds a map of object keys to a set of thinned entries,
and is used during output.

Extend ThinningCache so that it now olds a ThinningInfo structure,
rather than just the set of thinned entries.
This allows us to extend ThinningCache to hold additional information.
Here, we're adding the set of vetoed variables per object.

Remove the old thinning interface IThinningSvc.   It is no longer used.

We were allowing ranges in CondCont to overlap one another.
This isn't just academic, as we can frequently get such ranges
from IOVDbSvc.
However, this was leading in some cases to a lookup failure.

Change the semantics of overlap/extend so that we no longer overlap ranges.
Rather, in case of a overlap, the new range is truncated to avoid the overlap.
(We also allow treating an overlapping final range as an extension.)
Note that it is in principle possible for the truncated part
of the new range to contain the current IOV key.  The caller
will need to handle that case.

Pass large structures by reference, not by value.

SealSvc does not have any useful function any more