Giulio Eulisse

Thanks. I am doing so right now. I should have something in an hour or so.

This seems to leak somewhere. I have:  where it used to be flat.

On the other hand, it seems to have fixed the leak and memory usage is back to expected.

I was enabling SetClusterPrefetch as part of the attempt to reduce read_calls when processing our AODs. Indeed I now notice that it's enough to simply do: ```C++ // Was affected...

> What is the change (increase of the cache size or explicit cache learning or both)? Doing the caching at all. I thought prefetching was part of it, but apparently...

Ok, thank you for your explanations. I can confirm that 1 works already for me. I will try 3. 2 I probably do not need, actually.

All the test does is: ``` 5 TTree noEntry("noEntry", "All branches correct but no entry"); 4 std::vector digits; 3 std::vector rofs; 2 noEntry.Branch("MCHDigit", &digits); 1 noEntry.Branch("MCHROFRecords", &rofs); 129 BOOST_CHECK_THROW(DigitTreeReader dtr(&noEntry),...

Indeed, I guess the invalid_argument comes from the DigitReader reacting at the empty file. While I agree that a 0 entry TTree should not be particularly dramatic, if something external...

The exeception was invoked because `value.GetSetupStatus() < 0` was previously true, now it's not anymore.

For what I am concerned, we (ALICE) have at least one bit of code which relayed on the API returning a negative number, so I would appreciate if that behaviour...