LaminiIndex faiss type error

[soy-caminante]

I think this strictly speaking it is not a bug, but I this it would be good to know this (and fix it if possible).

The object self.index in LaminiIndex is an instantation of faiss.IndexFlatL2. on the calls to self.index.add and self.index.search the argument must be a numpy.array of type "float32".

It has taken me a little while to discover it. My system by default creates the numpy.arrays as "float64" making the app crash with this exception:

File "/usr/local/lib/python3.10/dist-packages/faiss/init.py", line 322, in replacement_search
self.search_c(n, swig_ptr(x), k, swig_ptr(D), swig_ptr(I))
File "/usr/local/lib/python3.10/dist-packages/faiss/swigfaiss.py", line 2146, in search
return _swigfaiss.IndexFlat_search(self, n, x, k, distances, labels)
TypeError: in method 'IndexFlat_search', argument 3 of type 'float const *'

[onestardao]

this is a classic silent-crash case from what we call type mismatch propagation — the system expects float32, but defaults to float64 and nobody yells until FAISS explodes.
what’s tricky is: the error trace often misleads devs into debugging the wrong layer (e.g., swig or cpp bindings), not the numpy cast.

we’ve cataloged a full list of these pattern-based failure modes across FAISS, RAG, and numpy-heavy stacks.
there’s a fix logic for each — symbolic, minimal, and meant to avoid these “I lost 3 hours to this??” moments.

let me know if you’d like the full map and fix guide — it’s MIT-licensed and public.