ROOTFrameReader crash when reading multiple edm4hep files about podio

[tmadlener]

Thanks for reporting this. We will try to have a look as soon as possible, but it will potentially take until after next week since many of us will be at CHEP.

[wdconinc]

@lkosarz @c-dilks Can you add a more reproducible example, maybe connected with a fake PR that modifies tests/read-multiple.cpp to use ROOTFrameReader, if that reproduces the issue?

[lkosarz]

I added a test example with the modified version of podio tests. This crashes even with a simple data type.
Just run:
make
./write
./read
testROOTFrameReader.zip

[tmadlener]

Thanks for providing a reproducer. We will have a look ASAP.

[tmadlener]

I just had a look and this is

• a problem in the API usage of the ROOTFrameReader (which is entirely undocumented, so also technically a bug)
• and a genuine bug that needs fixing

API usage issue

In the reproducer you have

auto reader = podio::ROOTFrameReader{};
reader.openFile("example_frame1.root");
reader.openFile("example_frame2.root");
reader.openFile("example_frame3.root");

Instead the correct usage would be to use the openFiles method for this use case and simply pass in the list of files that should be read in one go:

auto reader = podio::ROOTFrameReader{};
reader.openFiles({"example_frame1.root", "example_frame2.root", "example_frame3.root"});

(or pass a std::vector<std::string> instead of building it on the fly)

Internal bug

There is an additional bug inside ROOTFrameReader::readEventMetaData that is only triggered with multiple files, but that is in the end easy to fix by properly handling the switching of files in the TChain.

[lkosarz]

Thank you! My initial test was with openFiles() and passing of a pre-defined vector, so not on the fly. I did see the same crash happening in both openFiles() and multiple openFile() cases. I did also see memory leaks, but haven't checked it for this standalone code, I sent you. Maybe we should double-check for these too.