Former-commit-id: 51d79be71b9e86b2d0415857b09e2b7f1961a1f2

This reverts merge request !3963

Former-commit-id: 3f37a6f1f8c2fa667acf4b8c189d99e914b99d0b

Former-commit-id: d2dcd95d582f29f84574504b2c0dba1080be505a

Peter van Gemmeren authored 7 years ago and Johannes Elmsheuser committed 7 years ago

Former-commit-id: 68508a37dc3ab2e59f1db8c116a8574ec2e56fba

Peter van Gemmeren authored 7 years ago and Johannes Elmsheuser committed 7 years ago

Former-commit-id: f29eba9343ba4c46958d28388ff82f2f85cce68c

Peter van Gemmeren authored 7 years ago and Johannes Elmsheuser committed 7 years ago

Former-commit-id: f5e26949f8f98b8707cd1f8750614ce82e86d5f6

Peter van Gemmeren authored 7 years ago and Johannes Elmsheuser committed 7 years ago

Former-commit-id: 745afdfbb3035d05f02884a0af1db34ee923492c

Former-commit-id: fef00ed0655e82e48b41e7635e58457f25712445

Former-commit-id: 705565b0a99c889d7ef53a90835872633717c00e

EventLoop algorithms can change names after creation (and that would
be hard to fix), so I changed the code to ask the parent for their
name when the parent name is actually needed.

This update should also fix: https://its.cern.ch/jira/browse/ATLASG-1218


Former-commit-id: 841283749a47b73b73168cef7ca457d2fe39cf49

This is mostly needed to kill some warnings about missing dictionaries
when running EventLoop.


Former-commit-id: bfe3f61ffabdd7fe00a58450c0adcb8e503a906d

2017-04-19 Nils Krumnack <Nils.Erik.Krumnack@cern.ch>
        * added a test that the message level gets set correctly when the
          "OutputLevel" property is set via the AnaToolHandle.  See
          https://its.cern.ch/jira/browse/ATLASG-1165
        * tagging as AsgExampleTools-00-00-23


Former-commit-id: 0e53e5bc8630a482dc30bb32eb07c222ca84f76f

2017-05-15 Nils Krumnack <Nils.Erik.Krumnack@cern.ch>
        * add INamedInterface for RootCore (to use for the parent pointer
          for tool handles)
        * tagging as AsgTools-00-01-12

2017-05-11 Nils Krumnack <Nils.Erik.Krumnack@cern.ch>
        * added the new unit tests to CMakeLists.txt
        * tagging as AsgTools-00-01-11


Former-commit-id: 1bbaf8ec63e56bd49a9ef22202c7f3fcd3fd2069

To bring the branch in sync with the 2.6.X nightlies.


Former-commit-id: 8a854675ce4b2115a9fa0e2ca059cb895286a4c5

"offline mode" as well after all the recent code updates.


Former-commit-id: 4cdfd8c1bc1dab9a45ef73fd853edc8ca327dfec

branch. As some of the analysis tool packages already depend on features
that were not part of the code in this branch before.

The one thing that I didn't include from the master branch is the
const-correctness of the standalone implementation of ToolHandle<T>.
Since Gaudi doesn't force const-correctness for its own ToolHandle in
this branch, that just caused a lot of issues for the analysis code.
For no good reason. So the const-correctness issues will only be cleaned
up in the master branch properly.


Former-commit-id: 64b397379b90c683a755da250eed31c56508a218

correctly as part of the AnalysisBase project.


Former-commit-id: 9597ac6c18d7aee6957bbdad96d5e4c3b86158b2

With these changes AthenaMP will send a new set of error messages to the pilot
when running in the Event Service mode. These messages inform the pilot about
non-critical errors, which may occur during processing of a given event range
(e.g. the requested event range was not found in the input file). As a result
of such errors only a given range needs to be declared as failed, while the
AthenaMP job will keep running and process new event ranges.


Former-commit-id: f982a135

Former-commit-id: a82f5e672b054e2e955f69759bbf6fcd3888b9f2

Former-commit-id: e7a9dd855e51ed71a4c6b0be27d5e2881b0033d6

scott snyder authored 7 years ago and Johannes Elmsheuser committed 7 years ago

LoggedMessageSvc::initialize() was calling Service::initialize()
instead of AthService::initialize().

Former-commit-id: aed2b42fac280c9c0eab046d68102255ceac27e5

Former-commit-id: e42af8e305c09d58e7e2d3d72d3340772693716d

Former-commit-id: 82047f88868cafbee942762b9a0afcba9e925667

* Remember and compare process ID to detetect a child process, If a fork is detected, m_newLumiBlock is set to simulate an LB boundary and allow deriving classes to reset themselves. All LW histograms known the to the ManageMonitorToolBase class are also reset.


Former-commit-id: a31107fa249cc1c3f1491d6e0fe7ce68251b6861

AthContainers: Fix bug in setting the number of mantissa bits of a packed float.

See merge request !2941

Former-commit-id: f8e705bd

Former-commit-id: 6ef7fb5c

Add IAuxTypeVectorFactory::createFromData().



Former-commit-id: bd5150fb

Adding package AthContainersRoot.



Former-commit-id: 15768609

Implementing createVectorFromData.



Former-commit-id: aa83bfa3

Former-commit-id: fe129f9097ade9d25f0b204bd14c8ee2eb56a7de

Don't fail if AtlasProject is not defined.

Suppress an ubsan warning we get from cling.


Former-commit-id: 1f6767e24b3e6c647740231631fca390ccab3ac5

SG::ConstIterator was using a non-const dereference on the underlying
DataHandle, causing it to return null if the referenced object was locked.
This was leading to ubsan warnings in ControlTest.

Former-commit-id: addd0611

Boost has some known issues that trigger ubsan warnings
(Ex: https://svn.boost.org/trac/boost/ticket/11632)
Try to suppress these for athena jobs.


Former-commit-id: f88d3b1a055ebe184419c20695d87d9ee414a4b8

gcc6.2 expands the warnings from the undefined behavior sanitizer.
However, there are some false positives, and also real warnings coming
from external code that we don't want to change.

Some ubsan warnings can be suppressed through a file read at runtime,
not all can.  (It looks like gcc7 will be better in this regard, but i haven't
actually tested it.)

However, ubsan will not report a warning for a given source location
more than once.  So the hacky way of suppressing ubsan warnings is
to actually trigger the warning with standard error (fd 2) temporarily
redirected to /dev/null.

This change provides a function to do that.

I hope that this can be removed once we switch to gcc7.


Former-commit-id: a31239ba

A (deliberate) reinterpret_cast in copyAuxStoreThinned triggers a warning
from the undefined behavior sanitizer in gcc6.  For now, disable warnings
for this particular function.  This is done with a gcc-specific function
attribute, so add a macro in CxxUtils to make this visible only for gcc6
or later.

Also fix another ubsan warning in one of the tests.


Former-commit-id: b4de894f

This race condition was caused by synchronizing AthenaMP sub-processes
via two IPC objects: shared queue and shared memory segment. The shared
memory segment was dropped. As a result, the synchronization is now
implemented only via shared queue.

At the same time migrated two AthenaMP Tools to ATH_MSG macros.


Former-commit-id: e54ac4db5acdf3cbe8ad1a9de13a7bd952c5b88b

Former-commit-id: 78faa4dc76d4bb30975609876ae7501937760ee1

Update test reference file for PoolSvc changes.



Former-commit-id: 1a6b87feabe37d4e9695f77cac92a858acc87c05

Avoid warning about using Gaudi base class directly.



Former-commit-id: a452d8c33b7f6a269049c5044ea0bd57b7f0065a

* src/AthenaOutputStream.cxx: When writing an object with the
exact flag set, it is not sufficient to just cons up a new
DataBucket of the desired type.  The DataProxy is also used to
fill the DataHeaderElement.  So we also make a dummy DataProxy to
go with an AltDataBucket instance.



Former-commit-id: f871f7a148a6903bcfcd53c6eae50a8ed486b50b