Use of importlib.metadata breaks import when metadata is stripped

[JP-Dhabolt]

Expected Behavior: Importing the module works without bundling Package Metadata in application
Actual Behavior: Stripping Package Metadata breaks imports

Cause:
The pattern of directly using importlib.metadata to set the version assumes that package metadata is included in the location where the package is being used. As a result, this breaks when using slimmed down deployment artifacts, as package metadata is not bundled. I have seen a standard approach to this to be to wrap this self-referential version within a try/except block, e.g.

try:
    __version__ = version("annotated-doc")
except Exception:
    __version__ = "unknown"

https://github.com/fastapi/annotated-doc/blob/7dd21fb75ea6882caf79725881690551ab28c5ff/src/annotated_doc/__init__.py#L5C1-L5C39

[DBS-ST-VIT]

We are also facing this issue. Especially, we are using FastAPI in AWS Lambda and since FastAPI is depending on this package, our application isnt starting anymore as of this Import Error.

@tiangolo May you can have a look on this?

[raghavankarthick]

Hi @tiangolo, We are also facing this issue with FastAPI in AWS Lambda, can you take a look on this?

[JP-Dhabolt]

There is a workaround available, which is to modify your Lambda Zip artifact generation to include the **/*.dist-info* glob pattern. This will slightly increase the artifact size, but does temporarily resolve the issue.