Second attempt to perform debug registration

Cling
1

I am compiling on-line generated source code in a loop:

std::unique_ptrcling::Interpreter gCling;
for ( … ) {
gCling.reset(new cling::Interpreter(argc, argv, “…”));
…generate source code, compile and run it…
}

Seemingly at some random iteration i end up with:
GDBRegistrationListener.cpp:177: virtual void {anonymous}::GDBJITRegistrationListener::NotifyObjectEmitted(const llvm::object::ObjectFile&, const llvm::RuntimeDyld::LoadedObjectInfo&): Assertion `ObjectBufferMap.find(Key) == ObjectBufferMap.end() && “Second attempt to perform debug registration.”’ failed.

Can this be some kind of data race?

2

Could you use another variable name (just in case). gCling is reserved for other purposes.

3

Changed the name (although i don’t see how can that have any effect).

Is there some way to restart the interpreter (so i do not have to destruct and re-create it)?
I have tried unlink(), but i have no way to tell how many transactions need to be rolled back.

4

I think this is a bug sft.its.cern.ch/jira/browse/ROOT-7741
In the meanwhile you could try something like:

[cling$] #include "cling/Interpreter/Interpreter.h"
[cling$] gCling->process("...autogenerated_code..."); // this will create 1 transaction
[cling$] gCling->unload(1);
5

The generated code is complete source with includes, multiple functions etc so i think process() which would attempt to wrap it into a function is not going to work. I am using declare() but as far as i can tell the end result is not a single transaction, and i have not found a way to figure out the number of transactions generated (and unload() does not return any status so i cannot run it in a loop).

Currently the preamble (includes, typedefs etc) does not change between compilations, so i can try to compile the static part once, then at each iteration add functions then unload(# of functions added) – assuming function definitions add one transaction each. Would that work?