Previously, desping a warning message stating that the TFriendElement
"could not be added", TTree was adding the friend anyway, even if the
actual friend TTree object was not found in the file. This could lead
to missing data or stale data being read.

* universal time in pdf file

* changes after universal time in pdf file

* update release notes universal time in pdf file

* add name

Method provides special handling of colors, palette, gStyle

When custom colors created in ROOT (in most cases - due to custom
palette), list of custom colors will be temporary inserted into canvas
list of primitives. Also current palette will be placed there.
And gStyle object inserted into list of primitives - it has defaults
which can affect produced JSROOT output

In particular, ignore dots in the path to libcppyy_backend.

[CMake][Exp PyROOT] Keep PYTHON_VERSION_STRING with '.' and set PYTHON_UNDER_VERSION_STRING to build PyROOT instead

- Keep PYTHON_VERSION_STRING as set by CMake
- Set PYTHON_UNDER_VERSION_STRING to build PyROOT
- Remove from Cache not needed OTHER... variables
- Revert 2082d7ce
- Some documentation added

This is to fix ROOT-10549.

This is to fix ROOT-10549.

In MVSC it is enabled automatically. More about: https://software.intel.com/en-us/blogs/2014/08/28/exact-exception-propagation-in-the-intel-threading-building-blocks-library-and-c11

Now testing:
- writing trees with different names to the same file (only case already tested)
- writing trees with same name
- writing tree with same name and RSnapshotOptions::fOverwriteIfExists

All cases are tested in single- and multi-thread mode.

...and add RSnapshotOptions::fOverwriteIfExists to give users the chance
to still write a TTree to a TFile opened in UPDATE mode, even if a TTree
with the same name is already present, at the cost of deleting the previous
TTree from the file.

This fixes ROOT-10573.

std::numeric_limits<epsilon> is 2 larger than eps computed before in Minuit2. Use then *4 instead of *8.

suggested by Hans Dembinski.

Co-Authored-By: Philippe Canal <pcanal@fnal.gov>

Even if something within the event loop throws, we still need to call
CleanUpTask to make sure SnapshotHelperMT::FinalizeTask gets called,
to avoid teardown issues due to input and output TTrees of a Snapshot
being deleted concurrently.

Update documentation of DataLoader constructor by explaining the usage of the passed name to indicate the location of the
training results and how to use a separate location, as suggested in
https://root-forum.cern.ch/t/specify-absolute-path-for-dataloader-output/38541/4

When a RooArgProxy owns its argument, but the proxied object is
redirected, the proxy loses ownership. It was, however still deleting
its argument.

When multiple batches are requested from a PDF (e.g. with probabilities
or log values or normalised over different observables), batch memory was
always overwritten with the last request.
- BatchData.h has been refactored to pay attention to normSets and
caller IDs.
- When asking for batches that are larger than the actual stored
batches, BatchData now returns smaller batches instead of signalling
that there is no batch.

RooAbsData::reduce() will stop processing events after a hard-coded
number of events that's close to max int. This has been extended to
std::size_t to prevent surprising results.

RooEffProd, RooProdPdf and RooProjectedPdf were all intercepting
getValV() calls to save a norm set. This has been removed, as
RooAbsPdf does the same. Eventually, setting the _normSet member should
be removed entirely.

- Remove unnecessary RooFitCore_LinkDef.h
- Small cleanups in RooAbsData, RooDirItem and remove unsafe casts in
RooAbsRealLValue.
- Change type of loop variable in RooGaussian.
- Remove a std::pow to make clang-tidy happy.
- Remove unnecessary include from RooDataSet.

- Start grouping the objects in RooGlobalFunc.h into doxygen groups
to make them a bit easier to find.
- Improve progress messages of RooMCIntegrator.
- Several small documentation fixes.

Python used to build

Since right now we only build one TPython library, we cannot
use TPython from CppCallable and expect that it works on
both Python 2 and 3.

However, the same functionality is provided by Cppyy, and
libcppyy is built per Python version, so we are good.

When iterating over a std::vector<std::string>, new cppyy returns
std::strings. If we want to use Python str features like strip
or sorted, we need to convert the std::strings to Python strs.

Placing _rdf_utils inside its own package (utils) to exclude it from
pythonization imports interferes with walk_packages in
ROOT/__init__.py, which imports utils anyway. This causes the
failure of two JupyROOT doctest tests.

Solution: blacklist _rdf_utils so that it is not imported.