Rene,
Hi - after further thought it occurs to me that, in split level=2, we face a
major problem (unless I am completely misunderstanding something). If the
streaming is done by TBranchClones, then old datamember names can never be
'assigned' to the new name. (We are interested in changing many of our
datamember names.) This step is necessary to make the class's getters work
with the old files (our datamembers are protected) and, as I understand, is
usually taken care in the class streamer:
if (R__v < 2) {
R__b >> sector;
sect = sector;
} else R__b >> sect;
If the class streamer is never called, this assignment (sect = sector) is
never performed and the class getters cannot be used at all. How is this
taken care of in the old or new ROOT?
My only ideas are to write additional getter's for the old datamember names
(ugly solution!) OR to invent a local class which somehow makes all the
assignments between old and new datamember names. Or, we lose the ability
to read old files with a new class definition.
The most likely solution is that we (1) lose backward compatibility and
finish migrating to Root 3.01, (2) invoke the new BypassStreamer() function
of TClonesArray, and (3) make our datamember name changes and continue to
use custom streamers. Otherwise I do not see how we can ever do more than
simply reading trees once the class def changes for split level=2...
Thanks for the extra attention you've given to this,
Dan
-------------------------------------------
| Daniel Magestro +44-6159-71-2147 |
| magestro@gsi.de GSI/Kernphysik I |
| www.gsi.de/~magestro Planckstr. 1 |
| < last updated > 64291 Darmstadt |
| < Aug 6, 2001 > Germany |
-------------------------------------------
| From: Rene Brun [mailto:Rene.Brun@cern.ch]
| Sent: Tuesday, August 14, 2001 9:48 AM
| To: Magestro Daniel
| Subject: Re: [ROOT] Changing class definition in split level=2
|
| Hi Daniel,
|
| What you see seems to be TTree::Draw problems and not I/O problems.
| Could you send me (not to the list) a tar file with:
| - a copy of an old small file
| - a copy of a new small file
|
| Rene Brun
|
| Magestro Daniel wrote:
| >
| > Rene,
| >
| > Hi - as a test regarding the use of split level=2 and class
| def changes, I
| > have compiled our code with ROOT v3.01 including a simple
| change of one
| > datamember name (OLD=sector, NEW=sect). When I read in an
| old DST file that
| > was created with the original class def (we used ROOT
| v2.25), I see the
| > original datamember name 'sector' in the TBrowser. However,
| when I do a
| > TTree->Draw(), I get the following error message:
| >
| > root [2] T->Draw("HMdcCal1.fData.sector")
| > Error in <TTreeFormula::DefinedVariable>: sector is not
| a datamember of
| > HMdcCal1
| >
| > In other words, the old datamember name is not known any
| longer. Also,
| > although the new datamember name doesn't show up in the
| TBrowser, I *can* do
| > a TTree->Draw(), which returns only -1's (sect's initialized value):
| >
| > root [3] T->Draw("HMdcCal1.fData.sect")
| > <TCanvas::MakeDefCanvas>: created default TCanvas with name c1
| >
| > This seems quite strange to me. The TBrowser shows the old
| name which cannot
| > be drawn, but the new name is not shown but CAN be drawn.
| >
| > If I do a T->MakeCode() and look at the generated macro,
| the original
| > datamember name is listed as a leaf:
| >
| > ....
| >
| T->SetBranchAddress("HMdcCal1.fData.sector",HMdcCal1_fData_sector);
| > ....
| >
| > Clearly there is some difficulty in reading old files if
| the class def has
| > changed, even for tree drawing. Do you have any
| suggestions? Is this due
| > to the change in the schema for TBranchClones?
| >
| > Thanks again for your support, you've been really helpful...
| > Dan
| >
(see roottalk for further messages)
This archive was generated by hypermail 2b29 : Tue Jan 01 2002 - 17:50:57 MET