PyMC Compilation Nesting with pytensor.scan

11

I’ve tried bumping it with the -fbracket-dept=N flag several different ways, but for some reason it keeps ignoring the flag during the clang++ call no matter how I try to update the flag. I’ve tried altering that flag in my OS environment variables, I’ve tried to do it in code like os.environ['CXXFLAGS'] = "-fbracket-depth=2048" and os.environ['CFLAGS'] = "-fbracket-depth=2048", I’ve tried to do it by modifying flags in pytensor like

    pytensor.config.cxxflags = "-fbracket-depth=2048"
    pytensor.config.cflags = "-fbracket-depth=2048"
    pytensor.config.gcc__cxxflags = "-fbracket-depth=2048"

Regardless, when clang++ get’s called during the model compilation, it just ignores the flag and then crashes due to nesting.

If I try to switch to gcc explicitly, I get this warning:
WARNING (pytensor.configdefaults): Only clang++ is supported. With g++, we end up with strange g++/OSX bugs.

So, it seems that it’s only an issue for Mac’s because we have to use clang++, but there’s a compilation nesting limit.