NEWS

GFisher 0.3.1

The compiled code now links BLAS and LAPACK explicitly, through src/Makevars and src/Makevars.win. The C++ sources use RcppArmadillo, whose matrix operations call BLAS and LAPACK routines; R links a BLAS implicitly on Linux and macOS, but the Windows (Rtools) toolchain does not, so without this the Windows build failed at link time with undefined references to dgemm_ and similar. There is no change to any computation.

This fix has been on the repository's main branch since 2026-06-19, but no released tag carried it: v0.3.0 points at a tree published before the fix. Installing ZWuLab/[email protected] now gets it.
The package maintainer is now Zheyang Wu ([email protected]). Hong Zhang remains an author. Bug reports go to https://github.com/ZWuLab/GFisher/issues.

Initial public release.

Omnibus generalized Fisher's combination tests (oGFisher) that adapt the weights and degrees of freedom to the data, alongside the base GFisher family (Fisher's combination, Good's, Lancaster's, weighted-Z).
Accurate p-value calculation under dependence via moment-ratio matching and joint-distribution approximations (methods HYB, MR, GB).
High-performance C++ (RcppArmadillo) implementations with a three-tier fallback (cached C++ -> non-cached C++ -> pure R) verified to agree.
Fast *_ind paths for independent inputs via the optional coga package, with check_coga() to detect/install it.