We've had a lot of problems getting our project detected as a module during registration/packaging, and have tried a few configurations now that were very slow to build. We've just discovered the 'default' ImageSpec builder that was added to flyte recently, which has sped up our builds massively 🚀 But when we try to import the main repository module into our task code (following packaging), we hit the dreaded "ModuleNotFoundError: no module named 'mycoolproject'". Our folder structure is as below, and the commands we use for packaging are:

pyflyte --pkgs mycoolproject package -f
flytectl register files --archive flyte-package.tgz --version custom-version

The tasks work fine during

pyflyte run --remote

, but fail with the ModuleNotFound error after the production packaging process. We can get around the ModuleNotFoundError by building a custom builder that copies the project directory to the image - but ideally we'd like to use the default builder. Is there an extra command or step we're missing to ensure that flyte tasks correctly have access to the surrounding module? An example project structure is below....

/my-cool-project
|_ src
   |_coolproject
       |_ flyte
              image_specs.py
       |_ tasks
              tasks_1.py
              tasks_2.py
       |_ utils
              common.py
       |_ workflows
              workflows.py
       constants.py

tasks_1.py
from coolproject.constants import USEFUL_CONSTANT
from coolproject.flyte.image_spec import CUSTOM_IMAGE
from coolproject.utils.common import handy_helper

@task
def task_1(container_image=CUSTOM_IMAGE) -> None:
    handy_helper(USEFUL_CONSTANT)

task_1 would fail with the 'ModuleNotFound' error following the packaging steps...