When does Cling require headers to be present with a compiled library?

[tadej]

I compile a ROOT-based C++ library using CMake. Until now, I could copy the library and the binary using it to our cluster and run it there directly (I copied bin and lib folder).

Naively I would think I do not need a dictionary at all (as all my code should be compiled). Unexpectedly, my jobs started failing today when I made some changes.

The error I get is

Error in cling::AutoloadingVisitor::InsertIntoAutoloadingState:
   Missing FileEntry for common/LorentzChargeVector.h
   requested to autoload type LorentzChargeVector

This type is actually present in the dictionary I generate and the dictionary is present in the same folder as the library. I generate it using root_generate_dictionary CMake macro.

Is there a way to avoid including header files with the binary when putting on the cluster? I suppose something wants to re-generate the dictionary, can I check what?

Thanks in advance!

[Axel]

Hi Tadej,

We currently need to find the header to load the relevant library when that header is included in the interpreter.

@tadej

I do not need a dictionary at all

Then don’t generate one Dictionaries are needed for doing I/O on your types - if you don’t need that, don’t generate the dictionary.

Cheers, Axel.

[tadej]

Hi Axel, thanks for the answer! Will try without the dictionary for now and see how it goes.

I need the dictionary for local prototyping with PyROOT (which I also use to generate the configuration). When I submit the jobs to the cluster, I do not use any interpreter so I expect it is not needed (and indeed it was working until today).

[Axel]

Good! FYI, PyROOT does not need dictionaries; but PyROOT does need headers - because PyROOT uses cling, and cling needs to be told about your classes by including the header.

Cheers, Axel.

[tadej]

This makes things much clearer! Thanks!

It seems actually that I have an obscure crash in my code that also provoked cling to start doing strange things. Anyway I will cleanup our code to try to be as light as possible on dictionary generation.