Hi all, I’ve got some google cloud functions that rely on the

psycopg2

package. I’d like the ability to build + deploy my cloud functions from my MacOS machine, however I’m running into current limitations of pants:

16:44:33.76 [WARN] Google Cloud Functions built on macOS may fail to build. If your function uses any third-party dependencies without binary wheels (bdist) for Linux available, it will fail to build. If this happens, you will either need to update your dependencies to only use dependencies with pre-built wheels, or find a Linux environment to run ./pants package. (See <https://realpython.com/python-wheels/> for more about wheels.)

It looks like the only way to get a wheel is via

psycopg2-binary

(which works) but the documentation recommends avoiding this in production and instead using the source distribution. I’m wondering if anyone has any recommendations/advice to work around this limitation? Is having a Linux pipeline to build/deploy cloud functions really my only option? Or is it possible to temporarily override/replace psycopg2->psycopg2-binary for one-off deployments?

Hmmm, so this isn't exactly a limitation of Pants itself per se.

psycopg2

is an sdist, so it has to be built on the platform you're going to deploy it on. That would be true regardless of which build system was orchestrating the building. However in Pants 2.15.0rc7 you can use the new Environments feature to run the necessary processes in a Docker container on macOS, so you end up with a linux artifact.

Yep, its arm

So you'll need to build it for linux x86 on macos ARM...

Yep

but it will be slow...

I can have a look at doing it in docker thanks

Ah! That's the difference