Incompatibility of new pvlib with solarfactors run_timeseries_engine

[sophie-pelland]

Describe the bug
We had code that was working with pvlib 0.13.1 that used run_timeseries_engine. It fails with pvlib 0.15.0.

The error mentions the code cannot find components["horizon"]. After looking at the inner workings of the code, solarfactors calls on irradiance.perez. And the output of this has been changed such that a "poa_" prefix has been added, i.e. "horizon" has become "poa_horizon". I believe this is the source of the error.

[markcampanelli]

This was listed as an expected breaking change in the v0.14.0 release on 16 Jan 2026. https://pvlib-python.readthedocs.io/en/stable/whatsnew.html

[markcampanelli]

So, despite multiple ongoing breakage examples such as this one, IRL, I will probably never win this argument: solarfactors could better protect against such breakages by putting a hard upper bound on its dependency on (a pre-v1.0 version of) pvlib-python:

https://github.com/pvlib/solarfactors/blob/ecbfc863657e239817603a43898ae173c7ccad9c/pyproject.toml#L29

[sophie-pelland]

Thanks for the info, Mark. On our end, for now we reverted to pvlib 0.13.1, but obviously that is not a great solution.

[markcampanelli]

The history of pvfactors/solarfactors appears to have complicated this matter as well. Now there appears to be a circular dependency in which pvlib imports solarfactors and vice versa, essentially leading to the need for coupled releases of supposedly "independent" repos to prevent the problem you hit. As a quickish fix, I think a change needs to be made in pvfactors that keys those irradiance fields differently depending on the detected version of pvlib. A better, longer term solution should perhaps try to engineer away this circular dependency.