apply this:
https://github.com/root-project/root/pull/7193#discussion_r585660675

Cloned histogram should not be assign to current directory

The standard does not require that std::string's default constructor
be noexcept until C++17. For C++14, all reasonably new STL
implementations do add the noexcept specifier anyway, but e.g. STLs
coming with very old gcc versions might not.

Co-authored-by: Jonas Hahnfeld <jonas.hahnfeld@cern.ch>
(cherry picked from commit 842bbf85e0c611e00baf724e2c2c3e233b08ef20)

Old code was try to manipulate TString::Data() directly and
has several caviats. Now just check if some special options are present

Avoid streaming a copy of the TPaveStats parent during the TH1 cloning process.

Connection id can be used to provide client-dependent output
Also there is always master connection flag.
Use connection info in RFrame and RStatBox when processing ranges
requests

It is central place where it used and applied.
Move also Request and Reply classes into correspondent drawables

RFrame keep track of selected zoom range for each client.
Every client sends update when zooming is changed.
Statbox request changes when zooming is changed on client side.
On server side statbox gets necessary range settings, which was
transferred already by frame

Visible range may depend on different components like RFrame or
RHistDrawable. Such range required to calculate statistic for the
statbox or implement improved histogram display.

This object provided to RDrawable::Display() method as argument and
give contexts - canvas, sub-pad, primitive index. Same context made
available for RDrawableRequest

Fixes error about missing dictionary for function type.

Co-authored-by: Claire Guyot <claire.eloise.guyot@cern.ch>

- Use by default same tratment of point data error options when fitting  a MultiGraph or any of the Graph.
This fixes ROOT-10706

* Zmin and Zmax did not take into account the Z errors

* Update release notes.

* [histv7] Overflows are stored in another vector, and return -1 and -2 for RAxis::FindBin

* [histv7] still WIP

* [histv7] The under-/overflow bins should be refactored, separated from the actual bins

* [histv7] The binning tests work both for under-/overflow bins and for actual bins

* [histv7] RHistData is initialised with correct number of bins (actual and overflow)

* [histv7] Tests for filling and constructing are correct

* [histv7] Tests for axes are correct

* [histv7] The function names and the documentation correspond to the refactoring, and all tests are working

* Prefer kInvalidBin constant to magic zero

* Test the whole (normal) floating-point range

* "Basic" axis binning tests are now taken care of by the axis tests

* "Epsilon" equidistant binning tests are now taken care of by the axis tests

* On second thought, drop epsilon-binning tests

They're hard to test, and not testing a useful usage scenario.

* Make irregular axis test more generic

* Fix integer comparison warning

* Add forgotten cassert header

* Fix code typo

* Rework the binning tests using the new binning convention

* Comment fix

* Better ASCII art

* [histv7] The overflow refactoring is indexed negatively in row-major, binning tests are more complete but are still failing

* [histv7] The new row-major refactoring is working as expected, and the binning tests are passing

* [histv7] The documentation corresponds to the changes made. Enum bin types are replaced by constants where needed.

* [histv7] Some functions like GetBinIndex or GetBinCenter have some equivalents to be called with local per-axis bin indices.

* [histv7] Some documentation is corrected and changes have been made to RHistData, to resolve PR comments.

* [histv7] Apply functions are removed, as they are not needed anymore with the refactoring.

* [histv7] The opposite of GetBinIndexFromLocalBins() is available to the user.

* [histv7] Only the FindBin() function is available to the user, instead of 3 different functions.

* [histv7] Complementary tests for GetBin[From/Center/To] in binning.cxx

* [histv7] The higher edge of the last bin is correctly handled in AdjustOverflowBinNumber()

* [histv7] Documentation and implementation changes for RHistImpl (and related files) following PR review.

Co-authored-by: Claire Guyot <claire.eloise.guyot@cern.ch>
Co-authored-by: Hadrien G <knights_of_ni@gmx.com>

* Protect against division by zero
* Protect against null pointer

* Change compiler flags for Visual Studio v16.4.0 & v16.5.0

 - core/base: Since Visual Studio v16.5.0 the /O2 optimization flag makes most of the roofit/roostats tests failing
 - hist/histv7/test: Using /O2 compiler flag prevent the followin error when building in debug mode:
   axis.obj : fatal error LNK1179: invalid or corrupt file: duplicate COMDAT '??$?8DU?$char_traits@D@std@@@__ROOT@experimental@std@@YA_NV?$basic_string_view@DU?$char_traits@D@std@@@012@0@Z'

Exact info about lines delivered with object - can be adjusted for each
statbox subclass in the future

Let extract RPadLength directly from the css code.
Simplify implementation of custom attributes classes.

Assign RCanvas, RPadBase and RDrawable to requests members,
which can be accessed in RDrawableRequest::Process() methods.
Use pointer for pad and canvas while shared_ptr is not possible for
RCanvas

Client can submit request to the server and can assign callback
to get reply. Requests are derived from dedicated class and via
sub-classing can get arbitrary access to RDrawable API

As first example, use RHistStatBox for this. When working online,
statistic will be filled by the server through such special requests

Use version to detect which drawable need to be transfer to client
Idea to reduce transfer when only attributes of some objects are changed

It only reference RDrawable attributes, but not drawable itself.
Idea to be able transfer all graphical attributes of drawable, 
but be able to provide absolutely different data compared to the
drawable

First example is RHistStatBox. It will reference histogram, but this
histogram not required on client side. All statistic should be
calculated on the server side and send as ready-to-use data to client

Provide minimal demo in draw_rh2_colz.cxx tutorial

Start with dedicated stat box for histograms
Later one can introduce base class to have generic stat box,
which can be reused for other classes

- The documentation for the TGrah's EX0 fit option is clarified  as suggested in
  https://root-forum.cern.ch/t/fitting-tgraphasymerrors-difference-between-w-and-ex0/38522

- Expalin better the  meaning of  the documentation of the W and WW fit option for TH1 and TGraph's

In additon fix in the code the case of using the fit options W or WW together with the option P.
The option P is Pearson chi2fit  and it does not make sense to ignore errors in that case.
Add a warning message to signal the user this case