It’s for a FastAPI app that needs to connect to sql server with pyodbc. I’m building a linux container on mac, and have to provide a platform string for pex. No pyodbc matches that platform string, so we want to build from source in the image and have that available when the pex runs, hoping that solves the problem

@enough-analyst-54434 or someone else from the maintainer team will be able to guide you here when they wake up.

As to the issue with platform @icy-hair-30586 that is almost certainly orthogonal. Either the

PEX_*

technique ot the installing the PEX in a container approach should work without modifying platform strings. Do not though that in the PEX in a container approach you need to change the 1st example

RUN

line from

RUN PEX_TOOLS=1 /usr/local/bin/python3.10 /my-app.pex venv --scope=deps --compile /my-app

to

RUN PEX_TOOLS=1 /usr/local/bin/python3.10 /my-app.pex venv --pip --scope=deps --compile /my-app

. Without adding that

--pip

then created venv will not include

pip

to use to install more dependencies with.

More direct answers:

Is there a way to ignore the platform string only on the pyodbc dependency?

No. A PEX is a fully enclosed environment so Pex demands it is able to include all dependencies requested. As such, you can't say, "except for FastAPI". Instead you must take the work you did to build FastAPI from source in the container and save the resulting Linux platform specific FastAPI wheel outside the container. You then make that wheel availale to your build using

[python-repos] repos|indexes

https://www.pantsbuild.org/docs/reference-python-repos (whichever is easier for you to host internally, for most this is a

repo

which can be a simple directory with wheels in it that is shared over NFS or sshfs or some other network filesystem or else a simple webserver with index.html generation enabled for the directory with wheels in it). Finally, your

platforms

PEXes will now be buildable via Pants and the will include FastAPI since you pre-built the Linux wheel and made it available outside of PyPI. Its a bit of work to be sure, but does that make sense @icy-hair-30586?