Installer Issues on Pi Zero (Bookworm)

[parandandrd]

While configuring the EnviroPi on a fresh installation of Bookworm Lite 32-Bit (5-5-25) via the included installer script, I ran into a couple of issues:

1. The installer attempts to build numpy rather than using a prebuilt wheel. Takes a bit on a Pi Zero, but it does eventually work.

2. When installing dependencies for the examples, the installer attempts to build pil. This, however, fails with an error.

I resolved both issues by using "sudo apt install python3-pil python3-numpy" before attempting the installation.

[Gadgetoid]

PIL and numpy are usually fetched from PiWheels (https://www.piwheels.org) which provides binary wheels (pre-compiled) for 32bit Raspberry Pi versions.

Unfortunately PiWheels is currently undergoing maintenance (piwheels/packages#530) and not building new packages. During this period newer versions of numpy and PIL have been released which now have no binary available for 32bit Pi's. (my bad for not pinning these at known good versions, to be fair)

Installing packages from apt is a reasonable compromise, but if you're manually installing/creating a virtual environment it must be created with --system-site-packages so the venv can "see" the system Python packages.

In theory this problem will fix itself, but there's no clear timescale on when. In the interim I ought to pin these packages at versions which have PiWheels binaries.

For numpy the safest release is 2.0.2 - https://www.piwheels.org/project/numpy/ - so you should be able to pip install numpy==2.0.2 into a virtual environment and not affect your system Python.

For PIL (Pillow) it would be 11.1.0 - https://www.piwheels.org/project/pillow/ - so you should be able to pip install pillow==11.1.0.

In both cases it should be sufficient to use (and I will probably update our installer with):

pip install --only-binary :all: numpy pillow

But this is roughly analogous to apt install anyway.

(I installed 32bit full OS to test this and, of course, python3-numpy is preinstalled, masking the problem!)

Edit: Actually, better option is pip install --prefer-binary numpy pillow since this wont break on long lists of requirements where a binary package might be completely unavailable.