Hi Rene and roottalk,
I remember your reply to an email from Thomas Bretz about memory leak,
nov. 9, 2001, when reading arrays. I'm currently working with the
v3.02/07 and am facing very bad memory leak when reading a tree (when
processing/writing it, I do not encounter any increase in memory
ressources which are used). I have a default constructor in all involved
classes, which do not allocate any memory.
I suspect that I am at the origin of the problem. But, as I can't find
any mistake, I was also thinking that ROOT could possibly be at the
origin of it, and that the AutoDelete is not able to deal with a
structure like: a class object (event) having containers of class
objects (tracks), having themself containers of further class objects
(hits relating track objects), having themselves a member pointing to a
class object (a hit)
I explain a bit different the last sentence:
- to simplify a class event has a container of tracks and a container of
hits
- the next step is to associate some hits to a track (a subset of the
hits), and other additional informations coming from the association of
this subset of hits to a specific track. Then, to a track object belongs
a container of objects which are relating hits to tracks. Those objects
are then made out of member, whose one of them is a pointer to a hit
This structure seems to cause problem, I can hardly draw 10'000 event,
when I draw 1000 events, the memory which is used is about 160 MB,
issuing some plots (5) looking like the following statement (it should
also help to understand the structure):
event_tree->Draw("event->GetTrack(0)->GetNumberOfHits()")
where GetTrack returns a ptr to a track
The file size, for 1000 events, not compressed is 32 MB. Then each plot
seems to increase the used memory by about the file size ???
The problem disappear in case I don't stream the container of hits
relating track objects (or at least, I have no problem to deal with
10'000 events plots)
Then the question is to know if such a scheme can be handled by ROOT or
not. If yes, I have to look further to identify the problem more clearly
Thanks for any help,
Mathieu
This archive was generated by hypermail 2b29 : Sat Jan 04 2003 - 23:50:52 MET