Best practice on properly support OffsetArrays

README.md

@@ -132,6 +132,21 @@ julia> Origin(AO)(D)
 
 Here, `Origin(AO)` is able to automatically infer and use the indices of `AO`.
 
+## FAQ
+
+**How to properly write functions that support OffsetArrays?**
+
+It turns out that many packages don't handle custom axes well
+
+- Don't loop the vector using explicit `for i in 1:length(x)` -- use `for i in eachindex(x)` instead.
+- For multi-dimensional array, there are also `LinearIndices` and `CartesianIndices`.
+- When comparing the arrays, check the `axes(A) == axes(B)` instead of `size(A) == size(B)`.
+- Use `OffsetArrays.no_offset_view(x)`, handle it in 1-based indexing scenario, and optionally add the
+  offsets back. But don't overuse this, because
+- not every operation have a well-defined and unambiguous behavior for generic offset array. If that
+  is the case, use `Base.require_one_based_indexing` to early check the axes and early throw errors.
+  Let the user to the decision.
+
 <!-- badges -->
 
 [pkgeval-img]: https://juliaci.github.io/NanosoldierReports/pkgeval_badges/O/OffsetArrays.svg