Update documentation for utilizing py_library from nested folder about rules_python HOT 5 CLOSED

Regarding the Python behavior of adding the main script's directory to the import path, this is something we'd like to workaround to eliminate a class of hermeticity problems. See bazelbuild/bazel#7091.

from rules_python.

Ok, I figured it out, it seems that because the name of my py_binary is analysis as well as the name of the file analysis.py, the import statement in the file could not find the submodules (folders) within because it seems to have been interpretting the actual analysis.py file.

If I simple do import analysis in my analysis.py file, it actually executes twice.

Is there some rationale as to why py_binary/py_library do this? I can escalate this into the actual bazelbuild/bazel repo since this is behavior exhibited by the native rules.

from rules_python.

@ulfjack Who owns the native Python rules?

from rules_python.

lberki does, although we're planning on transitioning them to the python team this year. @duggelz

from rules_python.

This should be fixed once this change is released in Bazel: bazelbuild/bazel@e9c885a

You'll need to add legacy_create_init=false to your py_binary and py_library targets.

Meanwhile, my practical advice is to rename the directory or the file. You'll keep running into similar problems, even in non-Bazel contexts, because of how Python implicitly adds the main script directory to the start of the Python import path.

from rules_python.