Mark Stockton authored 4 years ago and Walter Lampl committed 4 years ago

Requires update Cost workflow for Trig_reco_tf (ATR-20974)
- Define COST BS output as DRAW_TRIGCOST (rather than a second BS to avoid conflicts in ProdSys)
- Split full BS to fill CostMon stream into the DRAW (before splitting other streams as requested)

- DRAW_TRIGCOST needs registering before can be used in ProdSys
- athena arguments for RAWtoCOST updated, but need further changes when scripts are ready

 - Pass class instances by const reference, not by value.
 - assert() with side effects.
 - unused value.
 

Thread-safety annotations should match between declarations and definitions.

Joerg Stelzer authored 4 years ago and Vakhtang Tsulaia committed 4 years ago

When running from the DB in athenaHLT, the HLTPSK needs to be read from
the database directly, `HLTPrescaleCondAlg.Source` needs to be changed
to `DB` to avoid trying to read the key from COOL.

Relates to ATR-22143.

- try to release early possible locks in tbb::concurrent_unordered_map (see also ATR-22112)
- put clearing and size allocation for output arrays of the eventCache_checkRobListToCache function
  at the begin of the function

Always put ROOT into batch mode to avoid additional GUI threads being
created that could cause problems during forking (ATR-21890).

!36223 introduced a slight change in behavior when reading
`DF_Enabled_ROB_IDs`, i.e. it was always overwriting the list of enabled
ROBs even if a value was set explicitly via job options.

Fix bug when parsing job properties of the type `MyAlg.MyTool.MyProp`.
The client name is in this case `MyAlg.MyTool` and not only `MyAlg`.

Package contains old-style data-scouting example for the muon
calibration stream. If we ever get around to using this online, this
could be revived and updated.

Standalone application to write trigger configuration to COOL/IS is no
longer needed (see ATR-19823, ATR-20641).

Add a new `ITrigEventLoopMgr::prepareForStart` method to the
`HltEventLoopMgr` that is being called by the `Psc` before calling the
start method of the `ApplicationMgr`. This will allow the EventLoopMgr to
setup anything that may be required during `start` (e.g. setting up
a partial EventContext, ATR-21331).

Patch sent by @wiedenma

Make sure getoutput() is in the subprocess module.
We still have a python 2 build, so we want thigs to work with both 
py2 and py3.

Fix or remove `atlas_install_*` statements that do not reference any
files.

Make test timeout shorter in opt build.

After some cleanup in upstream headers, the omissions showed up
in the build.

 - Suppress spurios ubsan warnings.
 - The internal timeout was frequently being hit in debug builds.
   Make it a bit longer.

`ITrigROBDataProviderSvc` was specific to L2. Remove the obsolete code
relying on this.

With atlas-tdaq-software/HLTMPPU!14 we made `SkipFinalizeWorker` the
default. In order for the child histograms to be saved with the regular
THistSvc they need to saved already at the end of the stop transition
because finalize never happens. Call `THistSvc::finalize` for the
workers at the end of `stopRun`. This also works in online mode with the
`TrigMonTHistSvc` as the finalize() method of that service is empty.

If run with `SkipFinalizeWorker=0` this also seems to work as the
THistSvc gracefully handles the case where it is being finalized
twice. However, this may of course change at any time. In that case we
should make this behavior dependent on the value of
`SkipFinalizeWorker=0`.

Relates to ATR-21715.

The work-around is no longer needed as the histograms are now saved
already during the stop transition in the PSC.

With atlas-tdaq-software/HLTMPPU!14 we made `SkipFinalizeWorker` the
default as this was the default operations mode at P1 since a long time.
However, in athenaHLT this leads to the child histograms not being saved
to the final ROOT file (ATR-21715). As a workaround set
`SkipFinalizeWorker=0` in athenaHLT until we have a proper fix.

(see also issue in ATR-21615)