I thought it's going to say something like Couldn't open file: xxxx. Is that a python thing that the error message gets lost? Or is it you want ParmParse::addFile to be explicitly mentioned (instead of looking it up in Backtrace files)?

Anyway there is a function called amrex::FileExists that you can use. Also do we need bcast the information and print on all processes? If the I/O process aborts, it will abort other processes too.

So it could just

if (ParallelDescriptor::IOProcessor()) {
    AMREX_ALWAYS_ASSERT_WITH_MESSAGE(FileExists(filename), "xxxxxx");
}

Thanks! Yes the error message I got so far looks not as clean as it could be.

If we do the check & abort inside an if(IOProc) then technically another MPI process could front-run the check and maybe obfuscate the error message again at scale (or even trigger abort before the clear error messsage is thrown). The BCast is a bit safer... What do you think?

To prevent other errors later, a barrier should be enough.

There is already a barrier after this in reading and bcast the file, which also produce Couldn't open file: xxxx if I/O fails.

My point is the fewer lines of code, the better, unless the code is hard to read.